research articles on purchase

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

Purchasing under threat: Changes in shopping patterns during the COVID-19 pandemic

Roles Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

Affiliation Department of Psychology, Clinical Psychology, Experimental Psychopathology, and Psychotherapy, Philipps University Marburg, Marburg, Germany

Roles Conceptualization, Methodology, Writing – review & editing

Affiliations Department of Psychology, Clinical Psychology, Experimental Psychopathology, and Psychotherapy, Philipps University Marburg, Marburg, Germany, Center for Mind, Brain and Behavior (CMBB), Philipps University Marburg and Justus Liebig University Giessen, Gießen, Germany

• Sebastian Schmidt, 
• Christoph Benke, 
• Christiane A. Pané-Farré

• Published: June 9, 2021
• https://doi.org/10.1371/journal.pone.0253231
• Peer Review
• Reader Comments

The spreading of COVID-19 has led to panic buying all over the world. In this study, we applied an animal model framework to elucidate changes in human purchasing behavior under COVID-19 pandemic conditions. Purchasing behavior and potential predictors were assessed in an online questionnaire format ( N = 813). Multiple regression analyses were used to evaluate the role of individually Perceived Threat of COVID-19 , anxiety related personality traits (trait-anxiety, intolerance of uncertainty) and the role of media exposure in predicting quantity and frequency of purchasing behavior. High levels of Perceived Threat of COVID-19 were associated significantly with a reported reduction in purchasing frequency ( b = -.24, p < .001) and an increase in the quantity of products bought per purchase ( b = .22, p < .001). These results are comparable to observed changes in foraging behavior in rodents under threat conditions. Higher levels of intolerance of uncertainty ( b = .19, p < .001) and high extend of media exposure ( b = .27, p < .001) were positively associated with Perceived Threat of COVID-19 and an increase in purchasing quantity. This study contributes to our understanding of aberrated human purchasing behavior and aims to link findings from animal research to human behavior beyond experimental investigations.

Citation: Schmidt S, Benke C, Pané-Farré CA (2021) Purchasing under threat: Changes in shopping patterns during the COVID-19 pandemic. PLoS ONE 16(6): e0253231. https://doi.org/10.1371/journal.pone.0253231

Editor: Marta Andreatta, Erasmus University Rotterdam: Erasmus Universiteit Rotterdam, GERMANY

Received: January 14, 2021; Accepted: May 31, 2021; Published: June 9, 2021

Copyright: © 2021 Schmidt et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All files (data set, R code) are available from the data_UMR repository under the following URL: https://data.uni-marburg.de/handle/dataumr/110 .

Funding: The authors received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

The spreading of the coronavirus disease (COVID-19) has led to worldwide stockpiling of food and hygiene products which caused temporally shortages [ 1 ]. In early March 2020, when the number of daily COVID-19 infections reached its peak in Germany [ 2 ], the German Federal Statistical Office recorded an enormous increase in sales of goods of sanitary and daily needs [ 3 ]: e.g., early in March 2020, a 150% increase for pasta, 153% for soap, and 751% for disinfectants. Similar changes in shopping behavior were recorded in the USA [ 4 ] and the UK [ 5 ]. At the same time, studies indicated an increase in fear and worries related to the virus [ 6 , 7 ].

The modulation of foraging behavior by threat has extensively been studied in the animal model [ 8 ]. In the natural environment, animals need to ensure a sufficient calorie intake while trying to avoid predatory attack. To parallel the natural habitat, animal studies use a safe nest area that must be left to obtain food. To evaluate threat related changes in foraging, the animals are confronted with a threat stimulus in the foraging area, such as the smell of a predator [ 9 ] or an electric shock [ 8 ]. In response to such threat encounter animals show an increase in risk assessment behaviors, e.g., attentive head-scanning [ 10 ], an inhibition of appetitive behavior [ 11 ], an increased latency in the procurement of food pellets [ 12 ] as well as a reduction in number of meals accompanied by an increase of the size of portions to maintain caloric intake [ 8 ].

A recent study investigated factors influencing stockpiling during the COVID-19 pandemic. Increased COVID-related worry (e.g., “I will become very ill.”; “I will not have access to food.”) was associated with stockpiling of more products indicating that negative affect like worries and anxiety influence shopping behavior [ 13 ].

In parallel to a predatory attack which constitutes a threat during natural foraging, the possibility of an infection with COVID-19 constitutes a threat in a human purchasing situation under pandemic conditions. In line with the described animal and human findings, we hypothesized that individually perceived threat resulting from possible COVID-19 infection will predict changes in human purchasing behavior under the current pandemic. Human purchasing is not only limited to food items. Increased selling rates were also reported for hygiene products such as disinfectant and toilet paper. Therefore, it seemed reasonable to consider purchasing of these necessities as a part of human foraging. Based on findings from animal research [ 8 ], we expected that perceived Threat of COVID-19 will lead to (1) a reduction in purchasing frequency and (2) an increase in purchasing quantity per purchase.

Additionally, we were interested in the influence of other factors known to influence feelings of anxiety that thus might be associated with threat perception of COVID-19 and changes in purchasing behavior. It has been demonstrated that psychological vulnerability factors such as trait-anxiety (i.e., the tendency to experience anxiety and perceive situations as threatening) and intolerance of uncertainty (i.e., the tendency of an individual to experience possible negative future events as unacceptable and threatening) increase the risk to fearfully respond to potentially negative or uncertain stimuli, events or situations such as those arising during the current pandemic. Both psychological factors have been linked with occurrence of anxiety-related disorders [ 14 – 16 ]. Initial evidence from the current COVID-19 pandemic revealed that trait-anxiety and intolerance of uncertainty are associated with higher levels of threat perception and fear of the coronavirus [ 17 , 18 ]. Another relevant factor that has been discussed to increase fear and threat perception of COVID-19 via transmission of threat information is the level of exposure to media. Studies from current COVID-19 pandemic higlight the role of increased media exposure on elevated anxiety and stress responses as well as increased fear of COVID-19 under the COVID-19 pandemic [ 17 , 19 ]. In the present study, we tested whether Perceived Threat of COVID-19 explains changes in purchasing behavior beyond these factors.

Purpose of the present study and hypotheses

Understanding the causes for changes in consumers purchasing behavior under the COVID-19 pandemic is of high relevance for governments and policymakers, e.g., to avoid panic buying which in turn may cause shortage of important goods. As stated in a perspective article by Van Bavel and colleagues [ 20 ] there are several research topics relevant to the COVID-19 pandemic which have to be addressed by social and behavioral sciences. Fear is a central emotional response during a pandemic which shapes information processing (e.g., risk perception) and behavior (e.g., shopping behavior). Based on a theoretical framework derived from rodent foraging behavior under threat, we examined the role of perceived threat originating from the present COVID-19 pandemic situation in predicting changes in purchasing patterns of groceries and hygiene products in an online questionnaire study. Our main hypotheses are that higher levels of Perceived Threat of COVID-19 would be (1) associated with a reduction of purchasing frequency and (2) an increase in purchasing quantity per purchase. We also expected a positive correlation between Perceived Threat of COVID-19 and an increase in purchasing quantity for individual products. In face of the known relevance of trait-anxiety and intolerance of uncertainty as risk factors for anxiety disorders and depression, we hypothesized that these constructs would be positively related to Perceived Threat of COVID-19 . Additionally, we expected individuals with high vulnerability to develop anxiety disorders (high trait-anxiety, high intolerance of uncertainty) to show a decrease in shopping frequency and an increase in purchasing quantity per purchase. Besides we hypothesized that being part of a risk group for a severe course of an infection with COVID-19 or having regular contact with a high-risk person would be associated with higher levels of Perceived Threat of COVID-19 and changes in shopping behavior as described above. A high extend of media exposure was also hypothesized to be positively associated with Perceived Threat of COVID-19 and changes in purchasing patterns (increased purchasing quantity while reducing shopping frequency).

The study was conducted from April 23 rd to May 18 th , 2020. In this time window the total amount of confirmed COVID-19 cases in Germany had reached 175.896. The implementation of public health measures by the German federal states started in March 2020 (e.g., prohibition to meet with others in public places, closure on non-essential shops, or closure of kindergartens or daycare institutions [ 21 ] while risk communication increased in the media, e.g., daily report of case numbers or information that infection with COVID-19 may cause a life-threatening disease and recommendations on how to avoid infection [ 19 ]. In effect, for March 2020 massive increases in sales figures were reported [ 3 ]. We asked participants to retrospectively rate their purchasing behavior for this month. The online questionnaire was realized using SoSci Survey [ 22 ] and was published on soscisurvey.de (see supplementary information for a German ( S1 Appendix ) and an English version ( S2 Appendix ) of the questionnaire).

Participants

In total 1074 individuals completed the online questionnaire and gave an answer to every question. Participants who did not finish the questionnaire were excluded. Data analysis was further limited to those participants for whom buying groceries constituted an actual risk of COVID-19 infection at time of assessment, i.e., we excluded participants who had already gone through a COVID-19 infection ( n = 3), or did not actually visit any stores during the assessment period due to either being in quarantine ( n = 30) or exclusively shopping online ( n = 49). To achieve a valid assessment of purchasing behavior changes from pre-pandemic to pandemic, we also excluded participants who did not make their own purchases (because, e.g., the partner did) before ( n = 65) and during the pandemic ( n = 150) leaving a final sample of 813 respondents (78% female). Participants were aged between 18 and 79 ( M = 42.42, SD = 15.00) (see Table 1 for descriptive statistics). The survey was advertised via the central e-mail system of Philipps University Marburg and on social media platforms. In order to motivate as many people as possible to participate in the online study, the raffle of three food delivery vouchers worth € 39.99 each was announced. The study was approved by the ethics committee of the Department of Psychology at the Phillips University of Marburg. Participants were informed that participation is voluntary and can be cancelled at any time without giving reasons, and that data will be stored anonymously. Written informed consent was obtained on the first page of the online questionnaire.

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

https://doi.org/10.1371/journal.pone.0253231.t001

Predictors.

Perceived Threat of COVID-19 . Perceived Threat of COVID-19 was measured using six semantic differential seven-point rating scales. The six items were introduced with “ The novel coronavirus is for me . . .. ” following two oppositely poled adjectives, (“concerning” vs. “not concerning”, “frightening” vs. “not frightening”, “something I am thinking about all the time” vs. “something I am not thinking about all the time”, “something I feel helpless about” vs. “something I can actively do something about”, “burdensome” vs. “not burdensome”, “close” vs. “far away”). These items were taken from the COSMO Snapshot Monitoring study conducted by the University of Erfurt [ 7 ]. The internal consistency of this scale was good (Cronbach´s α = .86). A principle component analysis indicated a one-dimensional construct, so we used the mean score of all six items as an indicator of perceived threat.

Intolerance of Uncertainty Scale . We used the 12-item short version of the Intolerance of Uncertainty Scale which maps the tendency of an individual to experience possible negative future events as unacceptable and threatening (e.g., “Unforeseen events upset me greatly.”) and is associated with worry, state-anxiety and related to anxiety pathologies [ 23 ]. The reported internal consistency of the short version is Cronbach´s α = .91. The internal consistency in this sample was good (Cronbach´s α = .87). A German validation study reported similar results (Cronbach´s α = .90) and reported intolerance of uncertainty to be predictive for worrying [ 24 ].

State Trait Anxiety Inventory . We used the trait portion of the State Trait Anxiety Inventory (A-Trait, e.g., “I worry too much over something that really doesn´t matter.”) which consists of 20 items. The internal consistency in this sample reached an excellent value of Cronbach´s α = .94. The reported Cronbach´s α for the A-Trait lies between .86 - .95 [ 25 ], for the German version Cronbach´s α = .90 [ 26 ].

Risk Perception . Participants assessed the likelihood of being infected with COVID-19 while shopping on a continuous scale ranging from 0% (“very unlikely”) to 100% (“very likely”).

Extend of media exposure . We asked participants to indicate how often they gather information about the COVID-19 pandemic on a four-point Likert scale (1 = “never”, 2 = “less than once a day”, 3 = “once a day”, 4 = “several times a day”).

Risk Group . Based on a standardized description ( “There is an increased risk of a severe course of COVID-19 disease for persons aged 50 years or older , smokers , persons with existing heart or lung diseases , chronic liver disease , diabetes mellitus , cancer or a weakened immune system . ” ) participants indicated if they (in person) belong to a risk group for a severe course of COVID-19 or if they have regular contact to a person (e.g., household member) belonging to such a risk group (coding: 0 = “no”, 1 = “yes”).

Social Desirability Bias . The Scale for Detecting Test Manipulation through Faking Good and Social Desirability Bias consists of seven five-level Likert items [ 27 ]. We used the individual scores to control for socially desirable reporting biases.

Demographic Variables . Participants reported their age in years, sex (coding: 0 = “female”, 1 = “male”), in which federal state they live, their highest level of education (1 = “no degree”, 2 = “primary education”, 3 = “secondary school diploma”, 4 = “high school graduation”, 5 = “university degree”) and their household size (number of persons living in a household).

Outcome measures.

Purchasing Behavior . Participants indicated the change in purchasing frequency and change in purchasing quantity for the month March 2020 relative to January 2020. We used January 2020 as a reference because at that point the German government did not consider the coronavirus to be a risk for Germany [ 28 ], no infection control measures were implemented yet [ 29 ] and no changes in purchasing behavior were observed compared to the usual level [ 3 ]. Participants were able to indicate the full range of change in purchasing frequency on a seven-point rating scale: Compared to January 2020 , before the outbreak of the Corona pandemic in Germany , how often did you go shopping in March 2020 ? (options: -3 = “much less frequently”, -2 = “less frequently”, -1 = “little less frequently”, 0 = “just as often”, 1 = “little more often”, 2 = “more often”, 3 = “much more often”). In correspondence, change in purchasing quantity was assessed using the following item: Compared to January 2020 , before the outbreak of the Corona pandemic in Germany , how much (quantity) did you buy per purchase in March 2020 ? (options: -3 = “much less”, -2 = “less”, -1 = “a little less”, 0 = “just as much”, 1 = “a little more”, 2 = “more”, 3 = “much more”).

Purchasing Quantity for individual products . For a more differentiated analysis we asked respondents to rate the purchasing quantity for individual products for March 2020 relative to January 2020. The following products were rated: toilet paper, soap, disinfectants, canned food, noodles/rice and fresh products (e.g., cheese, meat). There was the additional option to choose “do not usually buy this product”.

Data analysis

Purchasing Frequency and Purchasing Quantity were analyzed in separate multiple regressions controlled for gender, age, education, household size, and social desirability bias. In a next step, we entered all COVID-19 related variables (being part of a risk group, extend of media exposure to inform about COVID-19, risk perception of getting an infection, Perceived Threat of COVID-19 ) and anxiety related personality traits (intolerance of uncertainty, trait-anxiety) as a predictor of interest and examined its specific effect above the baseline model. In a final set of analyses, we entered all significant variables in one model and compared their effects. The same model was used to analyze the change in purchasing quantity for individual products.

Since Perceived Threat of COVID-19 was our main predictor of interest, we conducted an additional multiple regression analysis with the same baseline model as explained above and included the additional factors (e.g., sex, age, intolerance of uncertainty) to examine their specific predictive value for Perceived Threat of COVID-19 . For the ease of interpretation all continuous variables were z-standardized before entered into the model. We checked for multicollinearity using the variance inflation factor (VIF). All VIFs were smaller than two and thus considered unproblematic [ 30 ]. Since the dependent variables (purchasing frequency and purchasing quantity) were not normally distributed, we decided to additionally report confidence intervals (95% CI) based on bootstrapping [ 31 ] to bypass the assumptions for multiple linear regression. 2000 samples were generated to obtain an empirical distribution (using the boot.ci-function from the R package “boot”). Note that these results were highly comparable to the results of the parametric test. Additionally, we report non-parametric analyses (e.g., ordinal logistic regression) for the main findings as supplementary information (see S3 Appendix ), again showing highly comparable results. All analyses were conducted with R [ 32 ].

Change in purchasing frequency

The distribution of participants’ rating of change in purchasing frequency (see Fig 1 ) shows that 32.1% of study participants indicated that they went shopping for groceries as often in March as they did in January. 57.3% of the participants indicated that they went shopping less often and 10.6% indicated that they went shopping for groceries more often in March as compared to January. Overall, a one-sample t -test revealed a significant decrease in purchasing frequency from January to March ( M = -0.86, SD = 1.35), t (812) = -18.274, p < .001, d = 0.64.

N = 813. Note that categories “less” and “more” each comprise three gradations of the original scale (see section Outcome Measures ).

https://doi.org/10.1371/journal.pone.0253231.g001

Regression models were calculated for the full range scale (see S3 Table ) and, for clarity of hypothesis testing regarding a decrease in foraging frequency, excluding those 10.6% of participants who report an increase in purchasing frequency (see text below). Bivariate correlations between all variables are presented in S1 and S2 Tables (full range scale) in the supporting information. In the baseline model (see Table 2 ) sex and educational level were the only significant predictors for purchasing frequency. Female sex was associated with a decrease in purchasing frequency in March 2020 compared to January 2020, b = 0.32, t (672) = 3.46, p = .001, 95% CI [.13, .49]. Higher education was associated with a reduction of purchasing frequency, b = -.13, t (672) = 3.49, p = .001, 95% CI [-.21, -.06]. Adding Perceived Threat of COVID-19 to the model revealed that higher subjective threat was associated with a decrease in purchasing frequency, b = -.30, t (671) = 8.21, p < .001, 95% CI [-.39, -.23]. Intolerance of uncertainty and trait-anxiety revealed suppression effects, b = -.08, t (671) = 2.12, p = .035, 95% CI [-.16, -.01] respectively b = -.10, t (671) = 2.57, p = .010, 95% CI [-.18, -.03] (see S1 Table for correlations). The perception for being at high risk for infection with COVID-19 during shopping was associated with a decrease in purchasing frequency, b = -.19, t (671) = 5.03, p < .001, 95% CI [-.26, -.10]. Adding media exposure significantly improved the model, b = -.18, t (671) = 4.62, p < .001, 95% CI [-.25, -.10]. Belonging to a risk group was not a significant predictor of change in purchasing frequency ( b = -.17, t (671) = 1.92, p = .056, 95% CI [-.35, .01]) nor was having regular contact with a risk person ( b = -.11, t (671) = 1.46, p = .145, 95% CI [-.26, .05]).

https://doi.org/10.1371/journal.pone.0253231.t002

Finally, to check whether Perceived Threat of COVID-19 , risk perception and media exposure explained specific variance above and beyond the baseline model, all three predictors were entered in one block after the baseline model (see Table 3 ). The analysis revealed that Perceived Threat of COVID-19 ( b = -.24, t (667) = 5.60, p < .001, 95% CI [-.33, -.15]), risk perception ( b = -.10, t (667) = 2.63, p = .006, 95% CI [-.18, -.02]) and media exposure ( b = -.11, t (667) = 2.78, p = .009, 95% CI [-.19, -.03]) added incremental variance to the baseline model. The overall model explained 12.9% of the variance in change in purchasing frequency, F (10, 667) = 11.07, p < .001.

https://doi.org/10.1371/journal.pone.0253231.t003

Change in purchasing quantity

The distribution of participants’ rating of change in purchasing quantity (see Fig 2 ) shows that 45.5% of the participants indicated that they bought the same number of products per purchase in March as in January 2020. 8.6% of the sample indicated that they bought less products per purchase and 45.9% indicated that they bought more products per purchase in March as compared to January. A one-sample t -test confirmed a significant increase in purchasing quantity ( M = 0.58, SD = 1.12), t (812) = 14.673, p < .001, d = 0.51.

https://doi.org/10.1371/journal.pone.0253231.g002

Regression models were calculated for the full range scale (see S3 Table ) and, for clarity of hypothesis testing regarding an increase in purchasing quantity, excluding those 8.6% of participants who report a decrease in purchasing quantity (see text below). The baseline model (see Table 2 ) revealed that female sex ( b = -.19, t (672) = 2.05, p = .040, 95% CI [-.32, .02]), younger age ( b = -.09, t (667) = 2.38, p = .018, 95% CI [-.17, -.02]) and higher education ( b = .11, t (667) = 2.95, p = .003, 95% CI [.03, .19]) were associated with an increase in purchasing quantity. As expected, there was a positive association between Perceived Threat of COVID-19 and change in purchasing quantity, b = .29, t (671) = 7.86, p < .001, 95% CI [.22, .37]. Subjects who felt more threatened by COVID-19 increased their quantity of bought products per purchase. Intolerance of uncertainty and trait-anxiety explained significant variance and were both positively associated with changes in purchasing quantity, b = .11, t (671) = 2.70, p = .007, 95% CI [.02, .18] respectively b = .10, t (671) = 2.42, p = .016, 95% CI [.01, .18]. Higher risk perception for an infection during shopping was associated with an increase in purchased quantity, b = .13, t (671) = 3.72, p < .001, 95% CI [.11, .27]. People who indicated to inform themselves more frequently about COVID-19 (extend of media exposure) showed in increase in purchasing quantity, b = .19, t (671) = 5.20, p < .001, 95% CI [.12, .26]. Belonging to a risk group was not a significant predictor of change in purchasing quantity ( b = .03, t (671) = .39, p = .700, 95% CI [-.15, .20]) nor was having regular contact with a risk person ( b = -.01, t (671) = .07, p = .943, 95% CI [-.16, .13] ) .

Finally, all significant predictors were added to the baseline model (see Table 3 ). As observed for change in purchasing frequency, Perceived Threat of COVID-19 ( b = .22, t (667) = 5.02, p < .001, 95% CI [.13, .30]), the extend of media exposure ( b = .12, t( 667 ) = 3.10, p = .002, 95% CI [.05, .20]) and the perceived risk of getting infected while shopping (risk perception; b = .11, t (667) = 2.90, p = .004, 95% CI [.03, .20]) remained significant when adding all five variables together to the baseline model. The overall model explained 12.3% of the variance in change in purchasing quantity, F (10, 667) = 10.51, p < .001. Note that Perceived Threat of COVID-19 ( b = .08, p = .026) remained significant predictor for purchasing quantity when controlling for purchasing frequency (see S4 Table ).

Change in purchasing quantity for individual products

We analyzed the change of purchasing quantity for the individual products by entering all variables in the model. For an easier interpretation, we calculated the mean of change in purchasing quantity for “hygiene products”, i.e., toilet paper, soap, and disinfectants. In the same manner ratings for pasta/rice and canned food were aggregated to form the variable “non-perishable food”. For clarity of data interpretation, participants indicating that they bought less of a product were excluded from the analysis (3.3% for non-perishable food, 1.1% for hygiene products and 8.2% for fresh food). Perceived Threat of COVID-19 (see S5 Table ) was associated with an increase in purchasing quantity for non-perishable food ( b = .21, t (777) = 5.56, p < .001, 95% CI [.14, .29]), hygiene products ( b = .17, t (789) = 4.32, p < .001, 95% CI [.09, .25]), and fresh food ( b = .10, t (737) = 2.29, p = .028, 95% CI [.01, .18]). Risk Perception explained additional variance only for non-perishable food ( b = .11, t (777) = 2.94, p = .003, 95% CI [.04, .18]). High intolerance of uncertainty was associated with an increase in purchasing of non-perishable food ( b = .10, t (777) = 2.77, p = .006, 95% CI [.03, .17]), hygiene products ( b = .14, t (789) = 3.83, p < .001, 95% CI [.07, .21]), and fresh food ( b = .09, t (737) = 2.21, p = .028, 95% CI [.01, .16]). The extend of media exposure increased explained variance for non-perishable food ( b = .11, t (777) = 3.00, p = .003, 95% CI [.04, .18]) and hygiene products ( b = .13, t (789) = 3.47, p < .001, 95% CI [.05, .20]). Having regular contact to a risk person was associated with an increase in purchasing of non-perishable food ( b = .17, t (777) = 2.33, p = .020, 95% CI [.03, .31]). Belonging to a risk group oneself also was associated positively with an increase in purchasing of non-perishable food ( r = .102) but did not remain significant in the multiple regression analysis. The results for the full range scale are reported as supplementary information ( S6 Table ).

Perceived Threat of COVID-19

Since Perceived Threat of COVID-19 was our main predictor of interest, we conducted an additional multiple regression analysis (see S7 Table for all predictors) on Perceived Threat of COVID-19 . We entered the baseline model and all variables to the model that revealed a significant bivariate correlation with Perceived Threat of COVID-19 (see S2 Table ) to analyze which variables add specific variance to Perceived Threat of COVID-19 . Female subjects indicated higher Perceived Threat of COVID-19 , b = -.33, t (803) = 4.46, p < .001, 95% CI [-.47, -.18]. Age showed a negative association with Perceived Threat of COVID-19 , b = -.09, t(803) = 2.66, p = .008, 95% CI [-.15, -.02]. Educational level was positively related with Perceived Threat of COVID-19 , b = .13, t (803) = 4.48, p < .001, 95% CI [.08, .19]. Trait-anxiety ( b = .21, t (803) = 5.44, p < .001, 95% CI [.13, .29]) and risk perception ( b = .26, t (803) = 8.30, p < .001, 95% CI [.20, .32]) were positively related with Perceived Threat of COVID-19 and added specific variance to the model. Besides, higher frequency of information gathering (media exposure) was positively associated with Perceived Threat of COVID-19 , b = .27, t (803) = 8.57, p < .001, 95% CI [.21, .33]. The model explained 28.3% of the variance of Perceived Threat of COVID-19 . Note that due to the high correlation between trait-anxiety and intolerance of uncertainty ( r = . 61 ), intolerance of uncertainty did not reach significance ( p = .050). Intolerance of uncertainty added incremental variance when trait-anxiety was removed from the model, b = .19, t (804) = 5.98, p < .001, 95% CI [.13, .25].

The COVID-19 pandemic affected purchasing behavior all over the world. For future pandemics or a new flaring up of the COVID-19 infections it is important to understand relevant factors that influence panic buying. The aim of the study therefore was to investigate the role of Perceived Threat of COVID-19 and anxiety related measures on purchasing behavior. So far, studies investigating the influence of threat and anxiety on changes in purchasing behavior are scarce (e.g., Garbe and colleagues who have investigated the role of threat on purchasing of toilet paper [ 33 ] and Bentall and colleagues who also used a foraging framework [ 34 ]). In the present study, we investigated the role of Perceived Threat of COVID-19 and anxiety on purchasing behavior on a more general level and for different individual products.

The current study provides the following main findings: First and in line with our hypotheses, we found that the extend of Perceived Threat of COVID-19 is a significant predictor for changes in purchasing behavior, i.e., high threat was associated with a tendency to buy larger quantities per purchase and a reduction in purchasing frequency in March 2020 as compared to January 2020. Second, high intolerance of uncertainty was associated with an increase in purchasing quantity but not purchasing frequency (but significant suppression effect); trait-anxiety, which was highly correlated with intolerance of uncertainty, revealed a similar pattern, although there was a significant but small correlation with purchasing frequency ( r = -.08). Third, participants indicating a high extend of information gathering about COVID-19 tended to buy larger quantities and reduced purchasing frequency in March as compared to January 2020. Contrary to our expectations, being part of a risk group for a severe course of a COVID-19 infection or having contact to a person being part of such a group was not predictive for changes in purchasing behavior. All reported effects were controlled for gender, age, educational level, household size and a social desirability bias. Entering all significant predictors in one model revealed that Perceived Threat of COVID-19 was the best predictor for change in purchasing frequency as well as for change in purchasing quantity. For change in purchasing frequency Perceived Threat of COVID-19 , the extend of media exposure and participants’ risk perception of getting infected with COVID-19 while shopping were the only predictors that remained significant. The overall analysis for change in purchasing quantity revealed the same pattern of significant effects.

The observed purchasing pattern in our study shows resemblance to the strategic behavior seen in rodents. After the experience of an electrical shock in a foraging area, animals modified their foraging behavior to reduce the possibility of experiencing an aversive event by reducing the number of entrances to the foraging area while increasing meal size [ 8 ]. According to the threat imminence model, there are three defensive modes, each associated with a specific set of behaviors [ 35 ]. The mode activated depends on predatory imminence, i.e., the probability to encounter a predator. The pre-encounter mode is the first mode in the threat continuum and is activated when entering an area indicating some predatory potential. This mode is associated with meal pattern reorganization or protective nest maintenance which can be observed in animals. Our study provides evidence that humans also show similar adaptions in the face of the threat of a virus: buying larger quantities reduces the number of visits to stores necessary to maintain food supply and thus reduces the risk of an infection in the store. Importantly, the observed pattern of purchasing behavior was also predicted by the participant´s risk perception of being infected while shopping, which was correlated positively with Perceived Threat of COVID-19 ( r = .36). These findings suggest that the subjective assessment of infection risk is associated with feelings of threat and influences purchasing behavior. Similar results were observed in another online survey which also used an animal foraging framework to explain changes in purchasing behavior [ 34 ]. In this study, perceived probability of getting an infection was positively associated with increased purchasing quantity. In contrast to the present study, the authors emphasized on threat due to scarcity which is not covered in our study. The moderate correlation between risk perception and Perceived Threat of COVID-19 as found in our study suggests that additional factors–as for instance threat of scarcity–might explain additional variance in perceived threat. According to Bentall and colleagues [ 34 ], perceived risk of infection is a factor influencing scarcity vulnerability. Future studies ought to include threat due to scarcity to test whether Perceived Threat of COVID-19 remains a meaningful predictor for changes in purchasing behavior after controlling for threat due to scarcity. Unlike the rodents in the experiments by Fanselow and colleagues [ 8 ], participants did not experience an aversive event (e.g., electrical shock). Experimental studies (instructed fear paradigms) show that next to direct experience, fear and anxiety can be acquired also by informational transmission [ 36 , 37 ]. Since the outbreak of COVID-19, information about the virus and current numbers of new infections are reported on a daily basis. As reported elsewhere regular media exposure is a predictor of fear of the coronavirus [ 17 ]. In line with these findings, our analyses revealed that a greater extend of media exposure was associated with a higher level of Perceived Threat of COVID-19 suggesting its possible role as a form of verbal instruction of threat during the corona pandemic. At the same time, media exposure was associated with an increase in purchasing quantity and a decrease in purchasing frequency. Another study, using structural equation models, revealed that cyberchondria—that is, excessive information gathering about COVID-19 combined with feelings of frustration and anxiety—is positively associated with the intention to make unusual purchases [ 38 ]. A qualitative study on contents on twitter about toilet paper hoarding found out that nearly half of the analyzed tweets expressed negative feelings toward panic buying [ 39 ]. The authors hypothesize that this might lead to emotional distress, depression and anxiety-driven panic buying (see also [ 40 ]).

To get a more differentiated view, we also assessed the change in purchasing quantity for individual products. Our study extends the results reported by Garbe and colleagues [ 33 ] who investigated the role of perceived threat by COVID-19 and personality traits on purchasing of toilet paper. The authors found that high perceived threat by COVID-19 and high levels of emotionality predicted the amount of stockpiled toilet paper. In line with this finding, our data revealed that Perceived Threat of COVID-19 was positively associated with an increase in purchasing quantity for non-perishable food (canned foods, pasta/rice) and hygiene products (soap, toilet paper, disinfectants). Unexpectedly, high threat was also associated with an increase in the purchasing of fresh products, although this model showed the least variance explanation (see S5 Table ). Next to Perceived Threat of COVID-19 , intolerance of uncertainty added incremental variance for all product categories indicating that anxiety as a personality trait drives changes in purchasing behavior under threat.

Although only included as control variable, we found out that female sex was associated with a decrease in shopping frequency. This result could be interpreted as a more cautious behavior in female compared to male individuals. A study investigating the role of messaging and gender on intentions to wear a face covering under COVID-19 pandemic revealed that woman more than men intend to wear a face covering [ 41 ]. A mediating factor was the subjective likelihood to get the disease, supporting our post-hoc hypothesis that women behave more cautious under the COVID-19 pandemic. Note that in our data female sex was associated with higher levels of Perceived Threat of COVID-19 .

A limitation of this study is the retrospective rating of purchasing behavior in March 2020 which could be affected by memory biases. Longitudinal data would be important to see if subjective ratings of pre- and post-pandemic purchasing behavior differ and are associated with changes in perceived threat. The explained variance for change in purchasing frequency and purchasing quantity was rather small, indicating that additional factors were associated with a change in purchasing behavior. Recent studies indicate that e.g., right political affiliation [ 13 , 34 ], the extend of engaging in social distancing [ 13 ], and higher levels of paranoia [ 34 ] are associated with more stockpiling. Due to the correlational nature of this study no claims about causality can be made. Therefore, we cannot rule out that the found correlations between purchasing behavior and Perceived Threat of COVID-19 are coincidental although data from the German Federal Statistical Office suggests that there was indeed an unusual increase in sales figures in March compared to the mean of August 2019 to January 2020 [ 3 ]. More experimental studies should try to translate findings from animal experiments to human behavior to test whether certain behaviors are associated with different threat levels as reported in a study by Bach and colleagues [ 42 ]. Such studies could provide further evidence that foraging behavior is a relevant behavioral component of anxiety and fear in humans. Another limitation is the sex bias observed in the current study (78% of respondents were female) and the high proportion of high educated respondents which reduces generalizability although we controlled for sex and educational level. Two major strengths of this study can be mentioned: First, the derivation of hypotheses based on an animal model contributes to link findings from animal literature to human behavior. Second, this study collected purchasing behavior, anxiety ratings and Perceived Threat of COVID-19 around the peak of the COVID-19 pandemic in Germany and thus provides unique data about behavior and related predictors under an extreme event.

In conclusion perceived Perceived Threat of COVID-19 influences purchasing behavior in a twofold way: high levels of threat are associated with an increase in purchasing quantity and a reduction in purchasing frequency. The positive relation between Perceived Threat of COVID-19 and an increase of purchasing quantity was confirmed for individual products, too. Next to the Perceived Threat of COVID-19 , intolerance of uncertainty and the level of perceived risk for an infection during shopping also were significant predictors for purchasing behavior (quantity and frequency). While intolerance of uncertainty might be a relative stable personality trait, a reduction of risk perception could help to mitigate maladaptive changes in purchasing behavior like panic buying. Our data suggests that the extend of media exposure is associated with feeling of threat and change in purchasing pattern. This highlights the importance of appropriate risk communication. Information about effective protection measures while shopping could reduce high risk perception of being infected during shopping and might help to prevent panic buying. Additionally, recommendations about the amount of information gathering in media could have beneficial effects (e.g., informing only once per day to reduce negative effects).

Supporting information

S1 appendix. german version of the online questionnaire..

https://doi.org/10.1371/journal.pone.0253231.s001

S2 Appendix. English version of the online questionnaire.

https://doi.org/10.1371/journal.pone.0253231.s002

S3 Appendix. Non-parametric data analysis.

https://doi.org/10.1371/journal.pone.0253231.s003

S1 Table. Bivariate correlations.

https://doi.org/10.1371/journal.pone.0253231.s004

S2 Table. Bivariate correlations for the full range scale.

https://doi.org/10.1371/journal.pone.0253231.s005

S3 Table. Multiple regression analysis for the full range scale.

https://doi.org/10.1371/journal.pone.0253231.s006

S4 Table. Multiple regression analysis controlling for change in purchasing frequency.

https://doi.org/10.1371/journal.pone.0253231.s007

S5 Table. Multiple regression analysis for individual products.

https://doi.org/10.1371/journal.pone.0253231.s008

S6 Table. Multiple regression analysis for individual products for the full range scale.

https://doi.org/10.1371/journal.pone.0253231.s009

S7 Table. Multiple regression analysis for Perceived Threat of COVID-19.

https://doi.org/10.1371/journal.pone.0253231.s010

• 1. Statista. Coronavirus: consumers stockpiling food, by country worldwide 2020 | Statista. 24.04.2021. Accessed 24 Apr 2021.
• 2. RKI COVID-19 Germany. 17.12.2020. https://experience.arcgis.com/experience/478220a4c454480e823b17327b2bf1d4 . Accessed 17 Dec 2020.
• 3. Statistisches Bundesamt. Corona-Krise: Experimentelle Daten zeigen aktuelles Kaufverhalten. 28.04.2020. https://www.destatis.de/DE/Presse/Pressemitteilungen/2020/04/PD20_146_61.html . Accessed 17 Dec 2020.
• 4. How COVID-19 Has Transformed Consumer Spending Habits. 23.04.2021. https://www.jpmorgan.com/solutions/cib/research/covid-spending-habits . Accessed 24 Apr 2021.
• 5. O’Connell M, Paula Á de, Smith K. Preparing for a pandemic: spending dynamics and panic buying during the COVID-19 first wave: The IFS; 2020.
• View Article
• PubMed/NCBI
• Google Scholar
• 26. Laux L, Glanzmann P, Schaffner P, Spielberger CD. STAI: State-Trait-Angstinventar. Göttingen: Beltz Test GmbH.; 1981.
• 29. Our World in Data. COVID-19: Government Response Stringency Index. 17.12.2020. Accessed 17 Dec 2020.
• 32. Team RC. R: A language and environment for statistical computing; 2013.

Customer perception, purchase intention and buying decision for branded products: measuring the role of price discounts

• Research Article
• Published: 13 March 2021
• Volume 20 , pages 194–203, ( 2021 )

Cite this article

• Amit Dangi 1 ,
• Chand P. Saini 1 ,
• Vijay Singh 2 &
• Jayant Hooda 3  

874 Accesses

4 Citations

Explore all metrics

The purpose of this paper is to explore the antecedents of customer perception and its effect on the purchase intention and finally on buying decision-making about branded products especially luxury products, finally the role of price discounts in converting intentions into buying decision. This research has been carried in NCR with a collection of primary data by including statements related to the customer perception, buying intentions regarding branded luxury products and one section of the questionnaire included statements of Price discounts and buying decisions. The study used Exploratory Factor Analysis, Structure Equation Modeling, and Mediation through AMOS 19 to analyze the data. Results explored four major determinants named Quality, Trust, Psychological, and Social which were considered to contribute to building the perception of any customer for branded products and creates the purchase intention which will finally be converted into buying decisions making. The price discount plays a role of partial mediation, where due to price discount available for luxury branded products the buying decision-making has been reduced significantly.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Similar content being viewed by others

The Relationship Between Product Quality, Brand Image, Purchase Decision, and Repurchase Intention

Human values as added value(s) in consumer brand congruence: a comparison with traits and functional requirements

R. J. J. Voorn, G. van der Veen, … A. T. H. Pruyn

Consumer Desire for Luxury Brands: Individual Luxury Value Perception and Luxury Consumption

Aaker, D.A. 1991. Managing Brand Equity: Capitalizing on the Value of a Brand Nam . New York, NY: The Free Press.

Google Scholar  

Ahuja, N. (2015). Effect of Branding on Customer Buying Behaviour: A Study in Relation to Fashion Industry. International Journal of Research in Humanities & Social Sciences . https://doi.org/10.20908/ijarsijhss.v3i3.8162

Ambler, T. 1992. Need-to-know Marketing . London: Century Business.

Ayanwale, A., T. Alimi, and Matthew Ayanbimipe. 2005. The influence of advertising on Customer brand preference. Journal of Social Science 10 (1): 9–16.

Branded clothes market to double in India in five years | Zee Business. n.d. https://www.zeebiz.com/companies/in-depth-branded-apparel-market-to-double-in-india-in-five-years-1198 . Accessed 30 December 2020.

Cheung, G., and R. Lau. 2008. Testing mediation and suppression effects of latent variables: Bootstrapping with structural equations models. Organizational Research Methods 11 (2): 296–325.

Article   Google Scholar  

Cuieford, J.P. 1965. Fundamental Statistics in Psychology and Education , 4th ed. New York: McGraw-Hill.

Delener, N. 1994. Religious contrasts in customer decision behaviour patterns: Their dimensions and marketing implications. European Journal of Marketing 28 (5): 36–53.

Dillon, W.R., J.M. Thomas, K. Amna, and M. Soumen. 2001. Understanding what’s in a brand rating: A model for assessing brand and attribute effects and their relationship to brand equity. Journal of Marketing Research 38 (4): 415–429.

Doostar, M., E.M. Akhlagh, and M.K.I. Abadi. 2012. Analysis of the impact of brand assets on the buying decisions of final customers. Journal of Basic and Applied Scientific Research 2 (9): 8824–8832.

Iazzi, A., D. Vrontis, O. Trio, and Y. Melanthiou. 2016. Customer preference, satisfaction, and intentional behavior: Investigating customer attitudes for branded or unbranded products. Journal of Transnational Management . https://doi.org/10.1080/15475778.2016.1167000 .

Indian Customers’ awareness towards brands having sustainable practices increased - The Financial Express. n.d. https://www.financialexpress.com/industry/indian-Customers-awareness-towards-brands-having-sustainable-practices-increased/1800226 . Accessed 30 December 2020.

Khasawneh, K., and A.B.I. Hasouneh. 2010. The effect of familiar brand names on Customer behaviour: A Jordanian perspective. International Research Journal of Finance Economics 43 (1): 34–57.

Lamba, B. 2011. Influence of brands on customer’s buying behavior. Journal of Research in Commerce & Management , 2 (11).

Lamba, B. 2013. Abhinav National Monthly refereed Journal of Research in Commerce & Management Influence of Brands on Customer’s buying behavior. Journal of Research in Commerce & Management , 2 (11).

Malik, M.E., M.M. Ghafoor, H.K. Iqbal, Q. Ali, H. Hunbal, M. Noman, and B. Ahmad. 2013. Impact of brand image and advertisement on Customer buying behavior. World Applied Sciences Journal 23 (1): 117–122.

Nasar, A., S.K. Hussani, E. Karim, M.Q. Siddiqui, A. Nasar, S.K. Hussani, et al. 2012. Analysis of influential factors on customer buying behavior of youngster towards branded products: Evidence from Karachi. KASBIT Business Journal 5: 56–61.

Shopping: How the new Indian Customer shops, Marketing & Advertising News, ET BrandEquity. n.d. https://brandequity.economictimes.indiatimes.com/news/business-of-brands/how-the-new-indian-Customer-shops-gfk/70353653 . Accessed 30 December 2020.

Shrout, P., and N. Bolger. 2002. Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychological Methods 7 (4): 422–445.

The online private label growth paradigm. 2020, February.

Download references

Author information

Authors and affiliations.

Faculty of Commerce & Management, SGT University, Gurugram, India

Amit Dangi & Chand P. Saini

Department of Commerce, Indira Gandhi University, Meerpur, Rewari, India

Vijay Singh

Department of Commerce, C.H.L. Government College Chhara, Jhajjar, India

Jayant Hooda

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Amit Dangi .

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and permissions

About this article

Dangi, A., Saini, C.P., Singh, V. et al. Customer perception, purchase intention and buying decision for branded products: measuring the role of price discounts. J Revenue Pricing Manag 20 , 194–203 (2021). https://doi.org/10.1057/s41272-021-00300-7

Download citation

Received : 08 December 2020

Accepted : 04 January 2021

Published : 13 March 2021

Issue Date : April 2021

DOI : https://doi.org/10.1057/s41272-021-00300-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Purchase intention
• Buying decision
• Branded luxury products
• Customer perception
• Find a journal
• Publish with us
• Track your research

ORIGINAL RESEARCH article

The impact of consumer purchase behavior changes on the business model design of consumer services companies over the course of covid-19.

• 1 School of Business and Administration, Shandong University of Finance and Economics, Jinan, China
• 2 School of Statistics, Shandong University of Finance and Economics, Jinan, China

The COVID-19 pandemic has had a profound psychological and behavioral impact on people around the world. Consumer purchase behaviors have thus changed greatly, and consumer services companies need to adjust their business models to adapt to this change. From the perspective of consumer psychology, this paper explores the impact of consumer purchase behavior changes over the course of the pandemic on the business model design of consumer services companies using a representative survey of 1,742 individuals. Our results show that changes in consumer purchase behavior have a significant impact on the design of consumer services firms’ business models. Specifically, changes in consumers’ purchase object, motive, and timeframe are more likely to spark a novelty-centered business model design, whereas changes in purchase method tend to inspire an efficiency-centered one. Our findings provide a theoretical reference for consumer services companies in designing business models when faced with unexpected crises.

Introduction

The COVID-19 outbreak has abruptly disrupted the global political and economic order ( Fernandes, 2020 ), significantly impacting consumer services sectors such as retailing, hospitality, and tourism ( Pantano et al., 2020 ). The pandemic has resulted in unprecedentedly large-scale lockdowns across the world ( Kuckertz et al., 2020 ), severely restricting people’s daily activities. As a result, more consumer services companies are experimenting with new technologies and platforms in order to meet the changing consumer demands, leading to new consumption patterns. To cope with the restrictions, some consumer services companies have developed alternative business models, such as “contactless delivery” and “social cinema.”

The government’s strict restriction on population movement has led to seismic shifts in people’s livelihoods and daily lives. More people are suffering from depression and loneliness, and some have resorted to alcohol, drugs, or even self-harm for relief ( Alsukah et al., 2020 ). These unhealthy emotions and behaviors have caused quite shifts in individuals’ consumption psychology: people in a dire circumstance may develop a “nothing to lose” mentality and become more prone to risk-taking, resulting in more impulse purchases ( Hill et al., 1997 ; Harris et al., 2002 ); they might also develop post-traumatic stress disorder (PTSD) and future anxiety, resulting in fewer purchases to increase savings ( Nolen-Hoeksema and Morrow, 1991 ; Kılıç and Ulusoy, 2003 ; Kun et al., 2013 ). During the COVID-19 pandemic, consumer psychology and purchase behavior have fundamentally changed.

Purchase behavior is a special and specific behavior that directly reflects people’s needs, desires, pursuit of material and spiritual interests ( Braithwaite and Scott, 1990 ). Factors that affect changes in purchase behavior include social factors, cultural factors, demographic factors, and situational factors ( Cici and Bilginer Özsaatcı, 2021 ). Therefore, the COVID-19 pandemic as a social factor is also affecting different changes in purchase behavior. Scholars generally believe that a large number of consumers showed panic buying behavior or impulsive buying behavior in the early stage of the COVID-19 pandemic ( Aljanabi, 2021 ; Stuart et al., 2021 ), and even accompanied by compulsive buying behavior ( Samet and Gözde, 2021 ). While purchase behavior in the middle of the COVID-19 pandemic is characterized by mobility ( Gao et al., 2020 ; Zhang et al., 2020 ; Lu et al., 2021 ). The application of digital technology has created favorable conditions for consumers to participate in online shopping, and consumers’ online purchase activities have increased significantly ( Jiang and Nikolaos, 2021 ). However, the changes in purchase behavior in the above literature focus on changes in a single dimension, and do not systematically sort out the changes in consumer purchase behavior under the COVID-19 pandemic. Therefore, according to the basic theory of marketing, this study systematically sorts out the multiple dimensions of changes in consumer purchase behavior under the COVID-19 pandemic, and improves the items of the purchase behavior changes in each dimension, so as to provide supplements for the theory of consumer behavior.

Countries around the world have adopted special measures such as regional blockades in the process of fighting the epidemic. These measures are a shock to traditional business models and require corresponding changes to traditional business models. However, there are currently different perspectives on the impact of purchase behavior on corporate marketing models, including traditional brick-and-mortar store purchase models, green marketing models, B2B transaction models, and online marketing models ( Beuckels and Hudders, 2016 ; Nguyen et al., 2016 ; Sundström et al., 2019 ; Wei and Ho, 2019 ). However, there is little literature analyzing the impact of purchase behavior on firms’ business models from the perspective of sudden crisis events. Besides, there are many external factors affecting business model design, such as technological change ( Øiestad and Bugge, 2014 ), contextual factors ( Zott and Amit, 2013 ; Ghezzi et al., 2015 ), local market opportunities ( Sinkovics et al., 2014 ), and third-party partnerships in the customer value proposition development ( Velu, 2015 ). Among the above-mentioned external factors affecting business model innovation, less research is based on the impact of residents’ behavior. Therefore, it is particularly important to study the impact of changes in consumer purchase behavior on business model design in the context of the COVID-19 pandemic.

To answer these questions, this paper examines consumers’ psychological changes over the course of the COVID-19 pandemic based on the theory of environmental psychology, affective psychology, and consumer psychology. The stimulus-organism- response (S-O-R) model ( Mehrabian and Russell, 1974 ) is used to explain how the pandemic triggered people’s psychological alteration, which in turn sparked changes in their purchase behavior. Then, we conduct a representative survey of 1742 individuals to explore the impact of customer purchase behavior changes on the business model design of consumer services companies using the expectation confirmation theoretical model ( Oliver, 1980 ). The remainder of this article is structured as follows: Section 2 is devoted to conceptual basis and research assumptions; Section 3 presents the research design; Section 4 is the empirical analysis; Section 5 concludes the paper.

Conceptual Basis and Research Assumptions

Consumer purchase behavior changes during the covid-19 pandemic.

According to disaster psychology, different psychological changes of residents caused by different periods of emergencies make purchasing behaviors show distinctive characteristics, such as panic buying behaviors, impulse buying behaviors, compulsive buying behaviors and online buying behaviors. In the initial stage of the COVID-19 outbreak, although only the individuals who experienced the event will be directly affected, the negative emotions caused will be transmitted to the entire society through social networks. The public is prone to irrational emotions, including anxiety and depression ( Clauw et al., 2003 ; Klitzman and Freudenberg, 2003 ). Public anxiety, especially in the face of a large-scale pandemic, can easily lead to the spread of negative emotions ( Hull et al., 2003 ). In addition, consumers’ perception of uncertainty, scarcity, and severity and other psychological factors will increase, causing customers to panic buying behavior ( Omar et al., 2021 ). The specific performance is to stock up on some necessities and reduce the purchase of non-essential items ( Roşu et al., 2021 ). The fear of out-of-stocks and supply chain disruptions brought about by the COVID-19 pandemic will also increase consumers’ impulse buying behavior. The worse consumers perceive the COVID-19 outbreak, the stronger their inner fears, and the more likely they will lead to their impulsive purchases of health products. The COVID-19 pandemic has increased the perceived pressure of consumers, and some consumers are accompanied by compulsive purchasing behavior. By increasing their buying behaviors, they can relieve their inner anxiety and tension ( Samet and Gözde, 2021 ). Besides, online purchasing behaviors have become increasingly popular with consumers after the COVID-19 outbreak. In the face of the government’s home isolation measures, it has become more and more common for consumers to use online shopping for food and other items. People who are aware of the risks of going out are more willing to buy fresh food online ( Lu et al., 2021 ). Consumer purchase behavior is no longer limited by time and space, and consumers use mobile tools such as mobile phones to achieve shopping freedom ( Zhang et al., 2020 ). Among the above-mentioned studies have carried out detailed research on a certain characteristic of changes in consumer purchase behavior, but have not systematically sorted out changes of the psychological characteristics and behavioral characteristics of consumers. Changes in consumer purchase behavior are reflected in many aspects, not just a single dimension of change.

The stimulus-organism-response (S-O-R) model reveals the influence of the environment on individual emotions. “Stimulus” refers to any environmental factor that causes an individual’s intrinsic response to the environment. “Organism” represents the individual’s emotional state and cognitive process ( Zinkhan et al., 1992 ). “Response” is the individual’s response to the external stimulus ( Hunt and Downing, 1990 ). In short, the S-O-R theory states that external stimulus triggers people’s emotional and cognitive changes, which in turn lead to different behaviors. Therefore, the COVID-19 pandemic as the external stimulus will change people’s consumption psychology and hence their purchase behavior in terms of purchase object, motive, place, timeframe, and method.

In terms of purchase object, the outbreak of the epidemic has made consumers put forward higher requirements for products or services. When consumers face an emergency, they choose problem-solving products or services over emotional healing products or services ( Yeung and Fung, 2007 ; Cai et al., 2020 ). Utilitarian products, as opposed to hedonic items, are more effective in addressing consumers’ immediate needs ( Yang et al., 2020 ). Consumers caught in the pandemic would increase their purchases of utilitarian products such as disinfectants, masks, and health foods. On the other hand, when people are under pressure or are anxious about external threats, instead of directly addressing the issues, they often activate a psychological defense mechanism—the cognitive and behavioral tendencies that individuals unconsciously adopt in the face of frustration or conflict in order to relieve tension and anxiety ( Cramer, 1991 )—to protect themselves ( Baumeister et al., 1998 ). The COVID-19 pandemic has triggered people’s psychological defense mechanism, leading to more cautious buying. Consumers are not only more price-sensitive, but they also demand higher-quality and more reliable products. In terms of purchase objects, consumers pay more attention to the quality of the objects they buy. The increase in online purchasing activities has also made consumers more willing to disclose their personal information ( Gao et al., 2020 ).

In terms of purchase motive, previous scholars can divide purchase motivation into hedonic motivation, social motivation and utilitarian motivation ( Voss et al., 2003 ). This framework, which shapes consumer motivation for product categories, has been widely used in the field of consumer behavior. In recent years, the application of new technologies has become more and more extensive. Therefore, new media is used by more and more people and brings more fun to consumers. Driven by hedonic motivation, consumers are more keen on new media shopping methods such as Douyin and Kuaishou ( Koch et al., 2020 ). The contribution of social responsibility can improve consumers’ willingness to purchase in advance ( Tong et al., 2021 ). During the COVID-19 pandemic, many Chinese companies have donated financial and material resources during the pandemic, which helped build positive customer perceptions and attitudes toward their products ( Yin et al., 2019 ). Therefore, driven by social motivation, consumers are more willing to choose brands that have contributed to society. In addition, consumers’ herd mentality makes them more utilitarian in the process of purchasing goods, and thus more willing to choose products with higher evaluation ( Samet and Gözde, 2021 ). Driven by the above motivation, consumers choose more and more brands of goods.

In terms of purchase place, the government’s home isolation measures have made consumers’ offline shopping channels difficult, and their online purchases have become more and more common ( Zhang et al., 2020 ; Lu et al., 2021 ). Specifically, consumers have gradually developed the habit of purchasing some daily necessities online. At the same time, the rapid development of social media has brought more shopping convenience to consumers. As a result, when consumers shop on social platforms such as WeChat ( Larios-Gómez et al., 2021 ), they are able to pick their favorite products more quickly ( Ali et al., 2021 ). As the number of consumers on social platforms increases, the number of consumers in offline venues decreases accordingly. Although consumers’ offline purchasing activities have decreased, consumers have become more demanding of offline shopping places. In order to reduce the risk of infection, when consumers shop offline, they pay more attention to the safety, convenience and goodwill of shopping places ( Butu et al., 2020 ). As a result, consumers have also changed significantly in terms of purchase place.

When it comes to purchase timeframe, advances in technology stimulate consumers’ perception of the value of time. The new shopping habits that consumers have formed during the COVID-19 epidemic have made their sense of time sharper than before the COVID-19 outbreak. Consumers expect the fastest way to obtain goods and services ( Kyowon et al., 2020 ), improving their shopping efficiency. The development of Internet technology and the wide application of mobile terminals have enabled consumers to satisfy their desire to shop anytime, anywhere. Therefore, consumers prefer a shopping method with unlimited time to purchase goods and less time-consuming in terms of purchase timeframe.

In terms of purchase method, in order to avoid contact with uncertain external services and reduce the risk of infection, consumers choose contactless delivery methods based on safety needs ( Larios-Gómez et al., 2021 ). Through the contactless delivery method, consumers can effectively relieve their inner anxiety and smoothly maintain the order of daily life.

Consumer Purchase Behavior Changes and Business Model Design

People’s fear and anxiety about the pandemic are unlikely to abate in the near future, and the resulting changes in consumer demand might eventually damage the supply chain performance of consumer services companies ( Ivanov, 2020 ). These companies have already been experiencing significant challenges with their existing business models due to strict social isolation, delayed return-to-work, and disrupted logistics. The pandemic is putting some major businesses to the test since consumers may not restore their previous buying habits anytime soon ( Pantano et al., 2020 ). According to the Expectation Confirmation Theory, consumer services companies have to adjust their business models to meet new customer expectations in order to obtain consumer satisfaction.

Changes in consumer purchase behavior under the COVID-19 pandemic have had an impact on the design of novelty-centered business models. Novelty-centered business models place more emphasis on exploiting new opportunities in new ways ( Foss and Saebi, 2017 ), and their essence is to satisfy new customer value propositions, need or experience through innovations in the content, structure or governance of the activity system. Although the COVID-19 pandemic has led to a decline in consumers’ purchase power, the requirements for product quality upgrades will not change. Changes in purchase object drives consumer services companies to design novelty-centered business models. With the improvement of consumers’ overall consumption level, the enhancement of consumption power and the upgrade of consumption preferences, their satisfaction with standardized products gradually decreases, and the trend of pursuing more diversified and personalized products or services will continue. As consumer preferences increase in diversification, companies must launch new products and price them appropriately in the face of a fiercely competitive market, especially in the context of environmental uncertainty exacerbated by the COVID-19 pandemic. Novelty-centered business models can bring customers better products and experience through innovative methods on the basis of product technology innovation.

When it comes to purchase motive, consumers prefer products from companies with a reputable image or a strong sense of social responsibility. Branded products increase consumers’ perceived usefulness ( Bhattacherjee, 2001 ), which is precisely what novelty-centered business models could accomplish. Therefore, consumers expect companies to design novelty-centered business models. In terms of purchase method, consumers prefer novel purchase methods and services such as mobile payment and contactless delivery. This suggests that consumer demand for novel payment methods has not yet been completely satisfied. People who get exposed to the same products or services repeatedly will eventually get bored due to the diminishing marginal utility of overexposure ( Line et al., 2016 ). Bored customers will eventually feel less satisfied. Thus, consumer services companies should adopt a novelty-centered business model design in order to re-establish customer satisfaction. Moreover, consumers tend to favor a shorter purchase timeframe and a safer purchase place, indicating their expectation to reduce perceived risks ( Garaus and Garaus, 2021 ). To meet that expectation, firms would be better served by novelty-centered business model design. Therefore, changes in consumer purchase behavior have led to the emergence of novelty-centered business models. In summary, the following assumption is made:

H1a: Changes in purchase object facilitate the design of novelty-centered business models.

H1b: Changes in purchase motive facilitate the design of novelty-centered business models.

H1c: Changes in purchase place facilitate the design of novelty-centered business models.

H1d: Changes in purchase timeframe facilitate the design of novelty-centered business models.

H1e: Changes in purchase method facilitate the design of novelty-centered business models.

Changes in consumer purchase behavior under the COVID-19 pandemic have had an impact on the design of efficiency-centered business models. Consumer purchase behavior is a process from information acquisition, formation of purchase intention to purchase decision-making problem. Consumer purchase intention is an important factor that determines the final purchase decision. And information is an important factor that affects consumers’ purchasing intention and ultimately making purchasing decisions. Generally speaking, consumers are risk-averse, so they will collect a lot of relevant information before purchasing, so as to turn the uncertainty of purchasing a certain product into certainty. With the rapid development of information technology, whether the contradiction between the explosive growth of information and the limited attention of consumers can be resolved has become an inevitable requirement for enterprises to gain a competitive advantage. The rapid development of information technology also brings the risk of personal information being infringed on consumers at all times in the transaction, especially in the field of online consumption, the black industry chain of “stealing” and “illegal use” of consumers’ personal information shows an explosive growth trend. Whether companies can keep the personal information of consumers collected in business activities strictly confidential has become a matter of close concern to consumers. Efficiency-centered business models emphasize that enterprises can improve business efficiency by reducing transaction costs, improving information transparency and sharing, and improving transaction security. With this, information can be efficiently shared between customers and enterprises, and the “information island” between the two can be reduced, so that consumers can trust enterprises and generate purchase intentions.

In terms of purchase object, people are more rational in choosing what to purchase. This increases consumer demand for efficiency in the products or services purchased from consumer services companies. In this case, companies should choose an efficiency-centered business model design since it emphasizes improving the efficiency of business transactions. Customers are satisfied, and their expectations are confirmed when they perceive that the efficiency of the goods or services exceeds the expected efficiency. In addition, with regards to purchase motive, consumers tend to favor brands that are well rated and contribute to society. Consumers perceive branded products as allowing them to make the right choice more quickly. Efficiency-centered business models are consistent with this consumer perception. In terms of purchase place, people prefer to shop online or on social media platforms, highlighting their expectations for a safe and convenient shopping environment. Efficiency-centered business models are essential for firms to meet such customer expectations. As mentioned above, consumers prefer a shorter purchase timeframe, indicating that consumers’ time efficiency expectations have not been fully satisfied and the increasing need for consumer services companies to develop efficiency-centered business models. In terms of purchase method, the fact that consumers have become more favorable in mobile payment and contactless delivery reflects the growing consumer demand for efficient payment and delivery methods. Hence, consumer services companies need to design an efficiency-centered business model in order to increase customer satisfaction. Therefore, in addition to novelty-centered business models, the change in consumer purchase behavior has also created a demand for efficiency-centered business models. In summary, the following assumption is made:

H2a: Changes in purchase object facilitate the design of efficiency-centered business models.

H2b: Changes in purchase motive facilitate the design of efficiency-centered business models.

H2c: Changes in purchase place facilitate the design of efficiency-centered business models.

H2d: Changes in purchase timeframe facilitate the design of efficiency-centered business models.

H2e: Changes in purchase method facilitate the design of efficiency-centered business models.

On the basis of drawing on relevant research and theoretical achievements, this research innovatively constructs a theoretical research model of consumer purchase behavior on business model innovation under the background of normalization of the epidemic ( Figure 1 ).

Figure 1. Conceptual model.

Research Design

Survey design and variable measurements.

The data used in this paper was obtained through a representative survey. In order to ensure the reliability of the questionnaire, the design of the changes in consumer purchase behavior questionnaire adopted the literature method to select the measurement variables and corresponding items of the related research on consumer purchase behavior. On this basis, researchers related to consumer behavior were invited to evaluate the questionnaire, and potential consumers were selected as the survey objects for interviews, and some difficult and ambiguous questions were revised and supplemented. The scales for business model design mainly refer to the mature scales of relevant international studies. Then, modifications were made to account for the unique circumstances of the COVID-19 pandemic. On this basis, 58 individuals were selected for pre-investigation. Based on the test results, questions with relatively low factor loadings were further revised. Then, a rating scale is developed. The questionnaire take a form of a five-point Likert scale, where 1 = Strongly disagree , 2 = Disagree , 3 = Not sure , 4 = Agree , and 5 = Strongly agree . The main variables include basic demographic characteristics, the changes in consumer purchase behavior (purchase object, motive, place, timeframe, and method), and business model designs (novelty- and efficiency-centered).

Dependent Variable

Business model design (BMD) was chosen as the dependent variable. Based on Zott and Amit (2009) , we categorized business model design into novelty-centered business model design (NBM) and efficiency-centered business model design (EBM). The survey questionnaire was similar to that provided by Zott and Amit, with a few modifications to account for the COVID-19 pandemic. NBM was measured by ten items: (1) ‘ The merchant offers a wider range of goods to attract new customers ’; (2) ‘ The merchant offers a wider range of services to attract new customers ’; (3) ‘ The merchant offers a broader selection of brands ’; (4) ‘ The merchant is using more of a combination of physical and online shops to offer goods or service ’; (5) ‘ The merchant has adopted a wider variety of payment methods ’; (6) ‘ The merchant has become an industry benchmark ’; (7) ‘ The merchant is more creative in its stor e design’; (8) ‘ The merchant offers more innovative products ’; (9) ‘ The merchant offers more innovative services ’; (10) ‘ The merchant’s business model is new ’. EBM was measured by eight items: (1) ‘ The merchant has made my purchase of goods or services more efficient ’; (2) The merchant made my shopping time shorter ’; (3) ‘ The merchant has given me more information about the goods ’; (4) ‘ The merchant has given me more information about the services ’; (5) ‘T he merchant gave me more ways to buy and settle my bill ’; (6) ‘ The merchant made fewer errors in the sales process ’; (7) ‘ The merchant offers cheaper goods or services ’; (8) ‘ My communication with the merchant is faster and more efficient. ’

Explanatory Variable

Consumer purchase behavior changes (CPC) were the explanatory variable. Based on Valaskova et al. (2021) and Vázquez-Martínez et al. (2021) , we described consumer purchase behavior changes from five dimensions: changes in purchase object (PO), changes in purchase motive (PR), changes in purchase place (PP), changes in purchase timeframe (PT), and changes in purchase method (PW). As before, we made a few modifications to the questions measuring these variables to account for the COVID-19 pandemic. The specific measurements of each dimension were as follows.

According to marketing theory, changes in purchase object refers to the goods or services that consumers buy. Based on Zhang and Zheng (2019) , Consumers’ choice of purchase object is mainly reflected in price, quality and service. The measurement of the purchase object is measured from the above three aspects. At the same time, combining the characteristics of purchasing behavior under the COVID-19 pandemic ( Cai et al., 2020 ; Gao et al., 2020 ; Yang et al., 2020 ) and the results of interviews with consumers, the changes in purchase object (PO) were measured by nine items. As follows: (1) ‘ I am more likely to buy technology products (e.g., sports bracelets, etc.) ’; (2) ‘ I am more likely to buy high protein products (e.g., milk, etc.) ’; (3) ‘ I am more likely to buy high-end products ’; (4) ‘ I am more likely to buy personalized items’; (5) ‘I am more cautious about buying non-essential products ’; (6) ‘ I have higher expectations of customer service for the products I buy ’; (7) ‘ I am more concerned about the quality and efficacy of products ’; (8) ‘ I am more concerned about the price of products ’; (9) ‘ I am more likely to allow merchants access to my personal information. ’

The hedonic shopping motivation research scale according to Mark and Kristy (2003) and the utilitarian shopping motivation research scale by Martínez-López et al. (2014) formed the basis of the measurement scale of change in purchasing motivation in this study. On this basis, items unrelated to the COVID-19 pandemic were eliminated, and some items were improved to form a new measurement scale. The changes in purchase motive (PR) were measured by five items: (1) ‘ I am more likely to buy highly rated products ’; (2) ‘ I am more likely to try new brands ’; (3) ‘ I am more likely to buy products recommended by acquaintances ’; (4) ‘ I am more likely to buy products recommended in short video apps such as Douyin (Chinese TikTok) and Kuaishou ’; (5) ‘ I prefer brands that have contributed to society during the COVID-19 pandemic. ’

Marketing practice differentiates the place of purchase into online and offline ( Srikanth et al., 2011 ). Based on Volpe et al. (2013) , we measured the change in offline purchase location in purchase place. Ali et al. (2021) and Larios-Gómez et al. (2021) provided us with measurement items for changes in online purchase place. The changes in purchase place (PP) were measured by five items: (1) ‘ I am more likely to shop in a one-stop store ’; (2) ‘ I am more likely to buy goods in a contactless store ’; (3) ‘ I am more concerned about the safety of the shopping environment ’; (4) ‘ I am more concerned about the reputation of merchants ’; (5) ‘ I am more willing to shop on social media platforms such as WeChat. ’

Based on Eastlick and Feinberg (1999) , Consumers’ requirements for purchase timeframe were reflected in flexibility, speed and convenience. The survey questionnaire was similar to that provided by Eastlick, with a few modifications to account for the COVID-19 pandemic. Combined the results of the interviews, the changes in purchase timeframe (PT) were measured by three items. As follows: (1) ‘ I am more likely to spend an unlimited amount of time shopping ’; (2) ‘ I am more likely to spend less time shopping ’; (3) ‘ I am more organized in my shopping activities, such as making detailed shopping lists, planning shopping routes, and so forth. ’

Finally, based on Larios-Gómez et al. (2021) and the results of interviews with consumers, the changes in purchase method (PW) were measured by three items: (1) ‘ I am more willing to accept contactless delivery services ’; (2) ‘ I am more willing to use mobile payment ’; (3) ‘ I am more willing to use self-checkout. ’

Control Variable

Following existing literature, we selected respondents’ gender (Gender), age (Age), education attainment (Edu), and monthly income level (Income) as control variables.

To ensure measurement precision and accuracy, the data were analyzed using the item response theory (IRT) model rather than factor analysis, as the latter results in information loss ( Xue et al., 2019 ). The Item Response Theory (IRT) model estimates variables through an iterative computation process, making sufficient use of existing information. The IRT model also takes into account the difficulties of survey questions, making the estimations closer to real practice ( Xue et al., 2021 ). Therefore, we utilized the IRT model to measure business model design (BMD), including novelty-centered business model design (NBM) and efficiency-centered business model design (EBM).

Rabe-Hesketh et al. (2004) propose two types of IRT model, i.e., one-parameter logistic IRT (1PL-IRT) model and two-parameter logistic IRT (2PL-IRT) model. However, it is unrealistic to apply the 1PL-IRT model in real practices. Therefore, the 2PL-IRT model is widely used to measure latent variables. Given the fact that the 2PL-IRT model can only be applied to estimate binary variables, Zheng and Rabe-Hesketh (2007) integrate the partial credit model (PCM) into the 2PL-IRT model, namely the 2PL-PCM, to measure latent variables with multiple categories. Therefore, following Xue et al. (2019) , we employed the 2PL-PCM to measure BMD and NBM. The 2PL-PCM model specifications are as follows.

This paper aims to investigate the impact of consumer purchase behavior changes on the business model design of consumer services companies during the COVID-19 pandemic. The intended population for this research was identified as individuals who have shopped during the COVID-19 pandemic and have a basic understanding of consumer services business models. We fielded the survey from 18 April 2020 to 23 July 2020. All questionnaires were anonymous, and rigorous distribution and return protocols were followed. Questionnaires were distributed in three main ways: first, upon contact confirmation, our team members conducted on-site interviews with the respondents and distributed the questionnaires; second, using the team members’ social connections, the questionnaires were distributed to those who qualified; Third, the questionnaires were distributed through email. In the end, a total of 1,887 questionnaires were distributed, and 1,742 were valid following careful screening.

The demographic profile of the respondents is as follows. Male respondents account for 43.456%, while female respondents account for 56.544%. In terms of age, 0.459% of the respondents are under the age of 18; 30.540% are between 18 and 25 years old; 25.316% are between 26 and 35 years old; 19.518% are between 36 and 45 years old; 19.346% are between 46 and 55 years old; 4.822% are 56 years and above. Regarding education attainment, 1.607% of the respondents have a junior secondary certificate or below; 6.257% have a senior secondary certificate (including high school and vocational and technical school certificate); 48.565% have a university certificate; 43.571% have a postgraduate certificate or above. Finally, in respect of monthly income level, 19.460% of the respondents earn no income; 6.889% earn less than RMB 2,000 per month; 18.657% earn RMB 2,001–5,000 per month; 24.799% earn RMB 5,001–8,000 per month; 30.195% earn RMB 8,001 or more per month.

Common method variance (CMV) is likely to lead to biased results for variables obtained from survey questionnaires ( Xue et al., 2019 ). Therefore, we employed the Harman’s single factor test to examine the existence of the CMV. The test results showed that the common factor only explains 18.733% of total variance, indicating that the common method bias is not a concern for this paper.

Empirical Analysis

This section presents the empirical analysis conducted on the collected survey questionnaires. It includes four parts: (1) descriptive statistical analysis and correlation coefficient analysis; (2) analysis of consumer purchase behavior changes by demographic characteristics (including gender, age, monthly income level, and education attainment); (3) regression modeling; (4) robustness tests.

Descriptive Statistical Analysis and Correlation Coefficient Analysis

Table 1 showed the descriptive statistics of the main variables. All variables have a relatively small mean value, indicating that respondents’ willingness to change their behavior for the pandemic is low. A plausible explanation is that people became less vigilant and concerned as the pandemic was gradually brought under control. On the other hand, the novelty-centered business model has a higher mean value than the efficiency-centered business model, suggesting that following the pandemic, respondents tend to favor the novelty-centered business model over the efficiency-centered one. This is because as the outbreak gradually subsides, people become less pessimistic and hence more interested in new things. In addition, the standard deviations of all variables are small, indicating small variations for variables used in this study. This is also reflected in the extreme deviations, with the largest extreme deviation being only 5. Moreover, all the variables range from −3 to 2, indicating no extreme values observed.

Table 1. Descriptive statistics and correlation coefficients.

Table 1 also showed the correlation coefficients between the main variables. The results indicate a significant and positive correlation between consumer purchase behavior changes and both types of business model designs. However, the correlation between consumer purchase behavior changes and novelty-centered business model design is more significant; the impact of consumer purchasing behavior changes on novelty-centered business model designs is likely to be greater. However, the exact relationships between the variables remain to be tested further below.

Analysis of Consumer Purchase Behavior Changes by Demographic Characteristics

Over the course of the COVID-19 pandemic, consumer purchase behaviors have changed dramatically. These changes exhibited a number of differences according to demographic characteristics. Figure 2 illustrated the differences in consumer purchase behavior changes by gender, age, monthly income level, and education attainment. Details were be discussed in the following four sub-sections.

Figure 2. Consumer purchase behavior changes by demographic characteristics. (A) Consumer purchase behavior changes by gender. (B) Consumer purchase behavior changes by age. (C) Consumer purchase behavior changes by monthly income level. (D) Consumer purchase behavior changes by education attainment.

Figure 2A displayed the pandemic-induced changes in consumer purchase behavior by gender. The changes in purchase object and timeframe exhibited an apparent gender variation. Females usually tend to act more impulsively than males ( Hesham et al., 2021 ). The pandemic have prompted male consumers to be more rational in shopping; therefore, the purchase behavior change of male consumers is greater. In terms of purchase place and motive, a relatively small gender variation is shown. Lastly, no significant gender variation is found for changes in purchase method.

Figure 2B showed the pandemic-induced change in consumer purchase behavior by age group. Observably, all parameters of consumer purchase behavior changed exhibit age variation. Individuals aged 18–25 and 26–35 showed a smaller change in purchase object, but those aged under 18, 36–45, 46–55, and 56+ showed the opposite. The change in purchase place followed a similar pattern, with the exception that persons aged under 18 exhibited a lesser change. In terms of purchase motive and timeframe, the variation across age groups was minor; individuals aged 36–45 and 46–55 showed a greater change while other age groups showed less change. Lastly, the change in purchase method was relatively small across all age groups. The reason for that is: young people had already adapted to the online lifestyle before the COVID-19 outbreak, therefore no significant change after; but for the elderly, although they tend to be more skeptical of the internet, they now have little choice but to purchase online due to the pandemic isolation and lockdown. Overall, the middle-aged and elderly have changed the most in their purchase behavior.

Monthly Income Level

Figure 2C depicted the pandemic-induced changes in consumer purchase behavior according to monthly income levels. As can be seen, there was a significant variation. Individuals with no income or a monthly income of less than RMB 2,000 exhibited smaller change in their purchase object, place, timeframe, and method. People with a monthly income between RMB 5,001 and RMB 8,000 or above RMB 8,001 showed a greater change in their purchase object, place, and timeframe. In terms of purchase method, less variation was shown across monthly income levels. This is because the pandemic has prevented people from returning to work, resulting in a reduction in current or future household income, and because it has also affected people’s emotions and cognitions by instilling fear and anxiety about the future in them, prompting people to save preventively.

Education Attainment

Figure 2D showed the changes in consumer purchase behavior by education attainment. The changes in consumer purchase object and motive varied less across different education attainment levels compared to the changes in purchase place, timeframe, and method. Individuals with postgraduate or higher education attainment showed a small change in all aspects of purchase behavior. Unlike other demographic characteristics, education attainment had less of an impact on consumer purchase behavior.

Regression Modeling

To examine the impact of consumer purchase behavior changes on the business model design of consumer services companies, this paper constructed a regression model as follows. As shown in equation (3), BMD represents business model design, which includes the novelty-centered business model design (NMB) and the efficiency-centered business model design (EBM). CPC is consumer purchase behavior changes, which includes the changes in purchase object (PO), the changes in purchase motive (PR), the changes in purchase place (PP), the changes in purchase timeframe (PT), and the changes in purchase method (PW). The relationships between BMDs and CPCs are examined using the following model:

Table 2 displayed the regression results for the relationship between the pandemic-induced changes in consumer purchase behavior and novelty-centered business model design. The regression result for Model 1 (M1) showed that the changes in consumer purchase object has a positive impact on the novelty-centered business model design (0.584, p < 0.001). The regression coefficient of the change in purchase object and novelty-centered business model was 0.584, and it was significantly positively correlated at the 1% level. That was, the greater the changes in the purchase object, the more inclined the consumer services companies is to design a novelty-centered business model. H1a is validated. Similarly, the results for Models 2–5 showed that the change in consumer purchase motive, place, timeframe, and method all have a positive impact on the novelty-centered business model design (0.583, p < 0.001; 0.516, p < 0.001; 0.505, p < 0.001; 0.459, p < 0.001, respectively). The regression coefficient of the change in purchase motive, place, timeframe, and method and novelty-centered business model was 0.583, 0.516, 0.505, 0.459, and it was significantly positively correlated at the 1% level. That was, the greater the changes in the purchase motive, place, timeframe, and method, the more inclined the consumer services companies is to design a novelty-centered business model. H1b, H1c, H1d, and H1e are validated. Model 6 integrated all parameters of consumer purchase behavior changes in order to test their combined impact on novelty-centered business model design. The results were consistent with Models 1–5, thus confirming the robustness of the findings. Therefore, consumer purchase behavior changes under the COVID-19 pandemic significantly contribute to the novelty-centered business model design of consumer services companies. Moreover, the variance inflation factor (VIF) of each model was less than 10. This indicated that multicollinearity in the models was not serious, and hence has no effect on the results.

Table 2. Consumer purchase behavior changes and novelty-centered business model design.

Table 3 presented the regression results for the relationship between the pandemic-induced changes in consumer purchase behavior and the efficiency-centered business model design. Models 7–11 showed that the changes in purchase object, motive, place, timeframe, and method all have a significantly positive impact on the efficiency-centered business model design under COVID-19 (0.526, p < 0.001; 0.495, p < 0.001; 0.515, p < 0.001; 0.495, p < 0.001; 0.495, p < 0.001, respectively). H2a, H2b, H2c, H2d, and H2e are validated. Model 12 examined the combined effect of all parameters of consumer purchase behavior changes on efficiency-centered business model design. The results were consistent with Models 7–11, confirming the robustness of the results. Therefore, consumer purchase behavior changes over the course of the pandemic have a positive effect on the efficiency-centered business model design of consumer services companies. The variance inflation factor (VIF) of each model was less than 10. Again, this indicated that multicollinearity was not serious in the models, and hence had limited impact on the results.

Table 3. Consumer purchase behavior changes and efficiency-centered business model design.

Based on the above evidence, consumer purchase behavior changes under the pandemic have a positive impact on both novelty- and efficiency-centered business model design. However, there is a significant variance in the magnitude of the coefficients, suggesting that consumer purchase behavior changes may have varying degrees of impact on each type of business model design. Specifically, the pandemic-induced changes in purchase object, motive, and method are more conducive to the novelty-centered business model design of consumer services companies (0.584 > 0.526; 0.593 > 0.495; 0.505 > 0.495, respectively); the changes in consumer purchase place have an equal effect on novelty- and efficiency-centered business model designs; and the changes in purchase method have a weaker impact on the novelty-business model design than the efficiency-centered business model design (0.459 < 0.495).

The changes in consumer purchase motive, object, and timeframe have a greater positive impact on the novelty-centered business model design. Consumers can retrieve much information on the internet to reduce their risks and uncertainty, thereby increasing trust in decision-making ( Hussain et al., 2020 ). People are also more likely to purchase products or services recommended by others, and the internet is an effective way to obtain such information. As such, the changes in consumer purchase motive create an opportunity for consumer services companies to develop novelty-centered business models. During the pandemic, people have become more rational and quality-oriented in shopping, and consumer demand has shifted from quantity-focused to quality-and-quantity-focused. In this context, the market demands a wider range of products and services from companies, which can be achieved through novelty-centered business models. Therefore, the changes in purchase object have a positive impact on novelty-centered business model designs. In terms of purchase timeframe, when online shopping and home delivery cannot fulfill consumer demands in a timely manner due to pandemic disruptions and limited manpower, the consumer preference for community and near-home stores emerges. Therefore, the changes in consumer purchase timeframe have promoted novelty-centered business models, such as the physical community business model.

On the other hand, the changes in purchase method have predominantly favored efficiency-centered business models. The COVID-19 pandemic has put people at unprecedented risks. In order to reduce the risk, consumers have grown more interested in contactless delivery and mobile payment, which incorporate the omnichannel supply and provides the option to shop at any time. The customer need for low-risk, efficient, mobile, and fragmented shopping experiences opens up new business prospects for efficiency-centered business models. Therefore, the changes in consumer purchase method have a positive impact on efficiency-centered business model design for consumer services companies.

Finally, the changes in purchase place have a similar impact on novelty- and efficiency-centered business model design. Since COVID-19, there has been an increasing consumer demand for more diverse, personalized, convenient, and accessible shopping locations. Consumers want to shop in an innovative one-stop store that provides a safe or contactless environment, and this can be achieved by a business model that emphasizes both novelty and efficiency. Consumer services companies need to increase the diversity of their products and services while at the same time reducing their transaction costs to improve operational efficiency. Therefore, the changes in purchase place encourage both novelty- and efficiency-centered business model designs.

Robustness Checks

Alternative measures.

In the previous section, we used the Item Response Theory (IRT) model to measure the variables related to consumer purchase behavior and business model design. To check the robustness of the results, we re-measured the variables using the weighted average method. The regression results were all significant, as shown in Table 4 . All parameters of consumer purchase behavior changes have a significant impact on both novelty- and efficiency-centered business model design. The empirical results remain consistent with our prior findings. Therefore, our baseline results are robust.

Table 4. Alternative measures.

Control for Occupation

Another concern is the influence of missing variables on the relationship between consumer behavior changes and business model design. In the survey questionnaire, the respondents also provided information about their occupations. On the one hand, consumers’ occupation might alter their consumption behavior; while on the other hand, merchants might adjust their business strategies with respect to consumers with different occupations. As such, consumers’ occupation might affect the impact of consumer behavior changes on business model design, rendering the baseline results biased. Therefore, following Xue et al. (2019) , we introduced respondents’ occupation into the baseline regressions and re-estimate the models. The results are displayed in Table 5 . It shows that the results are highly consistent with baseline findings, with all regression coefficients being highly significant and positive ( p < 0.01). Accordingly, our baseline results are again robust and reliable.

Table 5. Adding control variable (occupation).

First of all, after the outbreak of the epidemic, there have been subtle changes in consumer buying groups. Male buying behavior has changed more. For example, they will increase the purchase of some necessities ( Vázquez-Martínez et al., 2021 ). In the face of crisis, people’s utilitarian motivation is more significant ( Voss et al., 2003 ). Therefore, people’s demand for daily necessities will increase substantially. In addition, the elderly no longer reject the purchase behavior through mobile methods, and many online shopping activities have been increased. This also makes life service companies need to further segment the market in terms of population in the future, such as adding more preferential activities for online service items for the elderly, so as to facilitate such people to further enhance their willingness to purchase.

Second, our findings suggest that changes in consumer purchase behavior have a significant positive impact on business model design. This influence reflects that consumers have put forward higher requirements for the marketing model of life service companies after experiencing the impact of the COVID-19 outbreak. According to the results of the study, changes in purchase object, purchase motive and purchase timeframe have a more profound impact on novelty-centered business model design. This shows that under the impact of the epidemic, consumer services companies should take rapid response measures, and carry out business model innovation according to the characteristics of the COVID-19 outbreak and changes in purchase behavior, such as: online transfer of sales model, expansion of target market, socialization and fragmentation of marketing model, unmanned retail, contactless service and enterprise platform integration.

Third, changes in purchase place and purchase method have a significant impact on efficiency-centered business model design. This shows that consumers currently hope that consumer services companies can reduce their selection costs, procurement costs and payment costs as much as possible, so as to ensure that they can obtain the required products or services more efficiently.

According to the research conclusions, this paper draws the following management implications: First, the consumer services companies based on new technologies should reduce their costs as much as possible and provide products or services efficiently. The company makes full use of the construction of new infrastructure such as Artificial Intelligence, Industrial Internet, and Internet of Things to power it, and makes innovations on this basis. In the future, the development direction of consumer services companies should be a deep and efficient combination of online and offline. In this way, a consumer-centric dynamic management model can be realized, and business models can be flexibly adjusted to respond to transform according to changes in the external environment. Second, enterprises should deeply explore consumers’ consumption preferences and stabilize the target market. The consumer market is unstable. While continuing to invest, companies should pay attention to the improvement of quality and service models, and deeply explore the consumption preferences of different consumers. On this basis, companies should continuously improve business models and stabilize the consumer market. Third, enterprises need to carefully introduce new models and services. During the outbreak of the epidemic, marketing models of live stream, community and short video have rapidly emerged. Not only have various e-commerce platforms started to adopt this marketing model, but some brand retailers have also begun to develop the live stream industry. According to the findings, consumers are enthusiastic about these emerging marketing models. At the same time, the unmanned retail model is also arousing the interest of consumers, and various intelligent retail products and services are put into operation, such as intelligent express cabinets, contactless distribution and unmanned convenience stores. The rapid development of these two types of models is affected by the epidemic environment, and managers should also consider the resources and capabilities of their own enterprise while rapidly innovating and introducing new models. At the same time, enterprises need to maintain a sense of crisis, cautiously introduce unfamiliar industries, and reasonably adopt various business models.

There are also limitations of this study that deserve future research attention. First, we explore the positive impact of consumer behavior changes on business model design in the consumer services sector. However, such relationship might vary across different sectors, cultures, and institution backgrounds. Future studies might examine it in a different research setting. Second, the picture of the nexus between consumer behavior changes and business model design might be incomplete. Future research might zoom into the consumption process or after-consumption behavior, investigating how the key findings might change with regard to different consumption stages.

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Author Contributions

HT collected literature. XS designed the research and wrote the manuscript. XS and HT performed the empirical analysis. XL provided the data. JT cleared data. DZ did the additional tests. All authors rewrote sections of the manuscript, contributed to manuscript revision, read, and approved the submitted version.

This research was supported by the National Social Science Foundation of China (Grant Number: 21BTJ019) and the Social Science Planning Foundation of Shandong Province (21CGLJ16).

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Ali, T. V., Pencarelli, T., Škerháková, V., Fedorko, R., and Košíková, M. (2021). The use of social media and its impact on shopping behavior of slovak and Italian consumers during COVID-19 pandemic. Sustainability 13:1710. doi: 10.3390/su13041710

CrossRef Full Text | Google Scholar

Aljanabi, A. R. A. (2021). The impact of economic policy uncertainty, news framing and information overload on panic buying behavior in the time of COVID-19: a conceptual exploration. Int. J. Emerg. Mark. [Epub Online ahead of print] doi: 10.1108/IJOEM-10-2020-1181

Alsukah, A. I., Algadheeb, N. A., Almeqren, M. A., Alharbi, F. S., Alanazi, R. A., Alshehri, A. A., et al. (2020). Individuals’ self-reactions toward COVID-19 pandemic in relation to the awareness of the disease, and psychological hardiness in Saudi Arabia. Front. Psychol. 11:588293. doi: 10.3389/fpsyg.2020.588293

PubMed Abstract | CrossRef Full Text | Google Scholar

Baumeister, R. F., Dale, K., and Sommer, K. L. (1998). Freudian defense mechanisms and empirical findings in modern social psychology: reaction formation, projection, displacement, undoing, isolation, sublimation, and denial. J. Pers. 66, 1081–1124. 00043 doi: 10.1111/1467-6494

Beuckels, E., and Hudders, L. (2016). An experimental study to investigate the impact of image interactivity on the perception of luxury in an online shopping context. J. Retail. Consum. Serv. 33, 135–142. doi: 10.1016/j.jretconser.2016.08.014

Bhattacherjee, A. (2001). Understanding information systems continuance: an expectation-confirmation model. MIS Q. 25, 351–370. doi: 10.2307/3250921

Braithwaite, V. A., and Scott, W. A. (1990). Measures of Personality and Psychological Attitudes. San Diego, CA: Academic Press.

Google Scholar

Butu, A., Brumã, I. S., Tanasã, L., Rodino, S., Dinu, V. C., Dobo?, S., et al. (2020). The impact of COVID-19 crisis upon the consumer buying behavior of fresh vegetables directly from local producers. Case study: the quarantined area of suceava county, Romania. Int. J. Environ. Res. Public Health 17:5485. doi: 10.3390/ijerph17155485

Cai, H., Tu, B., Ma, J., Chen, L., Fu, L., Jiang, Y., et al. (2020). Psychological impact and coping strategies of frontline medical staff in Hunan between January and March 2020 during the outbreak of Coronavirus Disease 2019 (COVID-19) in Hubei, China. Med. Sci. Monit. 26:e924171. doi: 10.12659/MSM.924171

Cici, E. N., and Bilginer Özsaatcı, F. G. (2021). The impact of crisis perception on consumer purchasing behaviors during the COVID-19 (coronavirus) period: a research on consumers in Turkey. Eskişehir Osmangazi Üniversitesi Ýktisadi ve Ýdari Bilimler Dergisi 16, 727–754. doi: 10.17153/oguiibf.923025

Clauw, D. J., Engel, C. C., Aronowitz, R., Jones, E., Kipen, H. M., Kroenke, K., et al. (2003). Unexplained symptoms after terrorism and war: an expert consensus statement. J. Occup. Environ. Med. 45, 1040–1048. doi: 10.1097/01.jom.0000091693.43121.2f

Cramer, P. (1991). The development of defense mechanisms: theory, research, and assessment. J. Pers. 55, 597–614. doi: 10.1007/978-1-4613-9025-1

Eastlick, M. A., and Feinberg, R. A. (1999). Shopping motives for mail catalog shopping. J. Bus. Res. 45, 281–290. doi: 10.1016/S0148-2963(97)00240-3

Fernandes, N. (2020). “Economic effects of coronavirus outbreak (COVID-19) on the world economy,” in IESE Business School Working Paper No. WP-1240-E. Spain: IESE Business School. doi: 10.2139/ssrn.3557504

Foss, N. J., and Saebi, T. (2017). Fifteen years of research on business model innovation. J. Manage. 43, 200–227. doi: 10.1177/0149206316675927

Gao, X., Shi, X., Guo, H., and Liu, Y. (2020). To buy or not buy food online: the impact of the COVID-19 epidemic on the adoption of e-commerce in China. PLoS One 15:e0237900. doi: 10.1371/journal.pone.0237900

Garaus, M., and Garaus, C. (2021). The impact of the Covid-19 pandemic on consumers’ intention to use shared-mobility services in German cities. Front. Psychol. 12:646593. doi: 10.3389/fpsyg.2021.646593

Ghezzi, A., Cortimiglia, M. N., and Frank, A. G. (2015). Strategy and business model design in dynamic telecommunications industries: a study on Italian mobile network operators. Technol. Forecast. Soc. Chang. 90, 346–354. doi: 10.1016/j.techfore.2014.09.006

Harris, K. M., Duncan, G. J., and Boisjoly, J. (2002). Evaluating the role of “nothing to lose” attitudes on risky behavior in adolescence. Soc. Forces 80, 1005–1039. doi: 10.2307/3086464

Hesham, F., Riadh, H., and Sihem, N. K. (2021). What have we learned about the effects of the COVID-19 pandemic on consumer behavior? Sustainability 13:4304. doi: 10.3390/su13084304

Hill, E. M., Ross, L. T., and Low, B. S. (1997). The role of future unpredictability in human risk-taking. Hum. Nat . 8, 287–325. doi: 10.1007/BF02913037

Hull, H. F., Danila, R., and Ehresmann, K. (2003). Smallpox and bioterrorism: public-health responses. J. Lab. Clin. Med. 142, 221–228. doi: 10.1016/S0022-2143(03)00144-6

Hunt, W., and Downing, S. (1990). Mergers, acquisitions and human resource management. Int. J. Hum. Resour. Manag. 1, 195–210. doi: 10.1080/095851990000000047

Hussain, S., Song, X., and Niu, B. (2020). Consumers’ motivational involvement in eWOM for information adoption: the mediating role of organizational motives. Front. Psychol. 10:3055. doi: 10.3389/fpsyg.2019.03055

Ivanov, D. (2020). Predicting the impacts of epidemic outbreaks on global supply chains: a simulation-based analysis on the coronavirus outbreak (COVID-19/SARS-CoV-2) case. Transp. Res. E Logist. Transp. Rev. 136:101922. doi: 10.1016/j.tre.2020.101922

Jiang, Y., and Nikolaos, S. (2021). Triggers of consumers’ enhanced digital engagement and the role of digital technologies in transforming the retail ecosystem during COVID-19 pandemic. Technol. Forecast. Soc. Chang. 172:121029. doi: 10.1016/j.techfore.2021.121029

Kılıç, C., and Ulusoy, M. A. H. I. R. (2003). Psychological effects of the november 1999 earthquake in turkey: an epidemiological study. Acta Psychiatr. Scand. 108, 232–238. doi: 10.1034/j.1600-0447.2003.00119.x

Klitzman, S., and Freudenberg, N. (2003). Implications of the world trade center attack for the public health and health care infrastructures. Am. J. Public Health 93, 400–406. doi: 10.2105/AJPH.93.3.400

Koch, J., Frommeyer, B., and Schewe, G. (2020). Online shopping motives during the COVID-19 pandemic—lessons from the Crisis. Sustainability 12:10247. doi: 10.3390/su122410247

Kuckertz, A., Brändle, L., Gaudig, A., Hinderer, S., Reyes, C. A. M., Prochotta, A., et al. (2020). Startups in times of crisis-a rapid response to the covid-19 pandemic. J. Bus. Ventur. Insights 13:e00169. doi: 10.1016/j.jbvi.2020.e00169

Kun, P., Tong, X., Liu, Y., Pei, X., and Luo, H. (2013). What are the determinants of post-traumatic stress disorder: age, gender, ethnicity or other? Evidence from 2008 Wenchuan earthquake. Public Health 127, 644–652. doi: 10.1016/j.puhe.2013.04.018

Kyowon, S., Sarang, G., and Byungdo, K. (2020). Pricing strategies under markets with time gap between purchase and consumption. J. Bus. Res. 120, 312–320. 066 doi: 10.1016/j.jbusres.2020.01

Larios-Gómez, E., Fischer, L., Peñalosa, M., and Ortega-Vivanco, M. (2021). Purchase behavior in COVID-19: a cross study in Mexico, Colombia, and Ecuador. Heliyon 7:e06468. doi: 10.1016/j.heliyon.2021.e06468

Line, N. D., Hanks, L., and Kim, W. G. (2016). Hedonic adaptation and satiation: understanding switching behavior in the restaurant industry. Int. J. Hosp. Manage. 52, 143–153. doi: 10.1016/j.ijhm.2015.10.005

Lu, M., Wang, R., and Li, P. (2021). Comparative analysis of online fresh food shopping behavior during normal and COVID-19 crisis periods. Br. Food J. [Epub Online ahead of print] doi: 10.1108/BFJ-09-2020-0849

Mark, J. A., and Kristy, E. R. (2003). Hedonic shopping motivations. J. Retail. 79, 77–95. doi: 10.1016/S0022-4359(03)00007-1

Martínez-López, F. J., Pla-García, C., Gázquez-Abad, J. C., and Rodríguez-Ardura, I. (2014). Utilitarian motivations in online consumption: dimensional structure and scales. Electron. Commer. Res. Appl. 13, 188–204. doi: 10.1016/j.elerap.2014.02.002

Mehrabian, A., and Russell, J. A. (1974). An Approach to Environmental Psychology. Cambridge, MA: MIT Press.

Nguyen, T. N., Lobo, A., and Greenland, S. (2016). Pro-environmental purchase behaviour: the role of consumers’ biospheric values. J. Retail. Consum. Serv. 33, 98–108. doi: 10.1016/j.jretconser.2016.08.010

Nolen-Hoeksema, S., and Morrow, J. (1991). A prospective study of depression and posttraumatic stress symptoms after a natural disaster: the 1989 loma prieta earthquake. J. Pers. Soc. Psychol. 61, 115–121. doi: 10.1037//0022-3514.61.1.115

Øiestad, S., and Bugge, M. M. (2014). Digitisation of publishing: exploration based on existing business models. Technol. Forecast. Soc. Chang . 83, 54–65. doi: 10.1016/j.techfore.2013.01.010

Oliver, R. L. (1980). A cognitive model of the antecedents and consequences of satisfaction decisions. J. Mark. Res. 17, 460–469. doi: 10.2307/3150499

Omar, N. A., Muhamad, A. N., Mohd, H. A., and Syed, S. A. (2021). The panic buying behavior of consumers during the COVID-19 pandemic: examining the influences of uncertainty, perceptions of severity, perceptions of scarcity, and anxiety. J. Retail. Consum. Serv. 62:102600. doi: 10.1016/j.jretconser.2021.102600

Pantano, E., Pizzi, G., Scarpi, D., and Dennis, C. (2020). Competing during a pandemic? Retailers’ ups and downs during the COVID-19 outbreak. J. Bus. Res. 116, 209–213. doi: 10.1016/j.jbusres.2020.05.036

Rabe-Hesketh, S., Skrondal, A., and Pickles, A. (2004). Generalized multilevel structural equation modeling. Psychometrika 69, 167–190. doi: 10.1007/BF02295939

Roşu, M. M., Ianole-Cãlin, R., Dinescu, R., Bratu, A., Papuc, R. M., and Cosma, A. (2021). Understanding consumer stockpiling during the COVID-19 outbreak through the theory of planned behavior. Mathematics 9:1950. doi: 10.3390/math9161950

Samet, Ç, and Gözde, G. K. (2021). Mediating effect of intolerance of uncertainty in the relationship between coping styles with stress during pandemic (COVID-19) process and compulsive buying behavior. Prog. Neuro-Psychopharmacol. Biol. Psychiatry 110:110321. doi: 10.1016/j.pnpbp.2021.110321

Sinkovics, N., Sinkovics, R. R., and Yamin, M. (2014). The role of social value creation in business model formulation at the bottom of the pyramid-implications for MNEs? Int. Bus. Rev. 23, 692–707. doi: 10.1016/j.ibusrev.2013.12.004

Srikanth, B., Pradeep, R., and Gokul, D. M. (2011). To buy or not to buy: indian consumers’ choice of online versus offline channels for air travel purchase. J. Hosp. Mark. Manag. 20, 831–854. doi: 10.1080/19368623.2011.577693

Stuart, J., Barnes, M. D., and Michela, A. (2021). Understanding panic buying during COVID-19: a text analytics approach. Expert Syst. Appl. 169:114360. doi: 10.1016/j.eswa.2020.114360

Sundström, M., Hjelm-Lidholm, S., and Radon, A. (2019). Clicking the boredom away–exploring impulse fashion buying behavior online. J. Retail. Consum. Serv. 47, 150–156. doi: 10.1016/j.jretconser.2018.11.006

Tong, E., Xie, Y., and Xiao, H. (2021). Effect of CSR contribution timing during COVID-19 pandemic on consumers’ prepayment purchase intentions: evidence from hospitality industry in China. Int. J. Hosp. Manag. 97:102997. doi: 10.1016/j.ijhm.2021.102997

Valaskova, K., Durana, P., and Adamko, P. (2021). Changes in consumers’ purchase patterns as a consequence of the COVID-19 pandemic. Mathematics 9:1788. doi: 10.3390/math9151788

Vázquez-Martínez, U. J., Morales-Mediano, J., and Leal-Rodriguez, A. L. (2021). The impact of the COVID-19 crisis on consumer purchasing motivation and behavior. Eur. Res. Manag. Bus. Econ. 27:100166. doi: 10.1016/j.iedeen.2021.100166

Velu, C. (2015). Business model innovation and third-party alliance on the survival of new firms. Technovation 35, 1–11. doi: 10.1016/j.technovation.2014.09.007

Volpe, R., Okrent, A., and Leibtag, E. (2013). The effect of supercenter-format stores on the healthfulness of consumers’ grocery purchases. Am. J. Agric. Econ. 95, 568–589. doi: 10.1093/ajae/aas132

Voss, K. E., Spangenberg, E. R., and Grohmann, B. (2003). Measuring the hedonic and utilitar-ian dimensions of consumer attitude. J. Mark. Res. 40, 310–320. doi: 10.1509/jmkr.40.3.310.19238

Wei, C., and Ho, C. T. (2019). Exploring signaling roles of service providers’ reputation and competence in influencing perceptions of service quality and outsourcing intentions. J. Organ. End User Comput. 31, 86–109. doi: 10.4018/JOEUC.2019010105

Xue, R., Gepp, A., O’Neill, T. J., Stern, S., and Vanstone, B. J. (2019). Financial literacy amongst elderly Australians. Account. Finance 59, 887–918. doi: 10.1111/acfi.12362

Xue, R., Gepp, A., O’Neill, T. J., Stern, S., and Vanstone, B. J. (2021). Financial literacy and financial strategies: the mediating role of financial concerns. Aust. J. Manage. 46, 437–465. doi: 10.1177/0312896220940762

Yang, Y., Li, O., Peng, X., and Wang, L. (2020). Consumption trends during the covid-19 crisis: how awe, coping, and social norms drive utilitarian purchases. Front. Psychol . 11:588580. doi: 10.3389/fpsyg.2020.588580

Yeung, D. Y. L., and Fung, H. H. (2007). Age differences in coping and emotional responses toward SARS: a longitudinal study of Hong Kong Chinese. Aging Ment. Health 11, 579–587. doi: 10.1080/13607860601086355

Yin, C., Bi, N., and Yu, H. (2019). Do CEOs’ personal donations matter? The impact of the CEO’s personal donations on consumers’ attitudes toward his/her company in China. J. Bus. Res. 100, 184–195. doi: 10.1016/j.jbusres.2019.03.051

Zheng, X., and Rabe-Hesketh, S. (2007). Estimating parameters of dichotomous and ordinal item response models with gllamm. Stata J. 7, 313–333.

Zhang, N., and Zheng, X. (2019). Agent-based simulation of consumer purchase behaviour based on quality, price and promotion. Enterp. Inf. Syst. 13, 1427–1441. doi: 10.1080/17517575.2019.1654133

Zhang, W., Leng, X., and Liu, S. (2020). Research on mobile impulse purchase intention in the perspective of system users during COVID-19. Pers. Ubiquitous Comput. [Epub Online ahead of print] doi: 10.1007/s00779-020-01460-w

Zinkhan, G. M., Roth, M. S., and Saxton, M. J. (1992). Knowledge development and scientific status in consumer-behavior research: a social exchange perspective. J. Consum. Res . 19, 282–291. doi: 10.1086/209302

Zott, C., and Amit, R. (2009). Business model design: an activity system perspective. Long Range Plan. 43, 216–226. doi: 10.1016/j.lrp.2009.07.004

Zott, C., and Amit, R. (2013). The business model: a theoretically anchored robust construct for strategic analysis. Strateg. Organ. 11, 403–411. doi: 10.1177/1476127013510466

Keywords : consumer psychology, consumer purchase behavior, efficiency-centered business model, novelty-centered business model, consumer services company

Citation: Tao H, Sun X, Liu X, Tian J and Zhang D (2022) The Impact of Consumer Purchase Behavior Changes on the Business Model Design of Consumer Services Companies Over the Course of COVID-19. Front. Psychol. 13:818845. doi: 10.3389/fpsyg.2022.818845

Received: 20 November 2021; Accepted: 10 February 2022; Published: 03 March 2022.

Reviewed by:

Copyright © 2022 Tao, Sun, Liu, Tian and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Xin Sun, [email protected]

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Front Psychol

Accessing the Influence of Consumer Participation on Purchase Intention Toward Community Group Buying Platform

Tanaporn hongsuchon.

1 Chulalongkorn Business School, Chulalongkorn University, Bangkok, Thailand

2 Intellectual Property Research Institute, Xiamen University, Xiamen, China

Associated Data

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The rapid development of community group buying platforms has attracted a huge attention from both the practical and academic communities. Although previous research has explored the influence patterns of community group buying platform on the customers’ purchase intention, there are limited studies on how customers’ purchase intention is influenced by their participation behavior. Therefore, based on social identity theory, this study constructs a theoretical model of consumer participation influencing users’ purchase intention through community identity in the community group purchase context, and examines the moderating role of users’ privacy concerns in this process in conjunction with privacy concern theory to systematically explore the role of consumer participation on purchase intention and its boundary conditions. In this study, the data collected from 532 valid samples are analyzed by structural equation modeling. The results of the study found that customer engagement behavior had a significant effect on purchase intention through the mediation of community identity, where privacy concerns negatively moderated the effect of community identity on purchase intention. The study reveals the intrinsic mechanism of customer engagement influencing purchase intention and its boundary conditions, which provides the suggestions for the marketing management and business practice of community group platforms.

Introduction

Since 2020, there have been some major changes in community group buying platforms in China, with the launch of new brands/platforms, and a lot of budget being invested in such platforms ( Li et al., 2020 ; Hu and Shu, 2021 ). Obviously, with the rapid development of social media and the popularization of community group buying APP ( Yang et al., 2021 ), in-depth analysis of customer engagement to establish a good relationship between consumers and community group buying platforms is extremely important for community group buying platforms in a competitive business environment to maintain consumers’ continuous use. Consumer participation to a certain extent causes the occurrence of psychological changes in consumers, which leads to their purchase intentions and behaviors. It profoundly affects the behavior, interpersonal interaction rules, and related performance of members within the community group buying platform, and has become a frontier topic of current research in the field of virtual environments.

It is now believed that consumers are likely to be influenced by their engagement behaviors to purchase the community group buying platform itself and the products or services it offers, and that consumer usage behaviors are the most essential element in maintaining the “consumer-community group buying platform” relationship. The consumer behavior is the most essential element of the relationship between the “consumer and the community group buying platform,” and is the basis for the success of the community group buying platform ( Hu and Shu, 2021 ). Community group buying platform consumer “participation behavior – willingness to buy” problem research is crucial, become community group buying platform business model innovation, marketing management changes, and theoretical research boom ( Li et al., 2020 ). The research boom is mainly driven by two forces: the practical needs of community group buying platforms and the theoretical needs of academics ( Hsu et al., 2018 ). Both hope to assist community group buying platforms to explore a new way of operation and development as a way to gain competitive advantage ( Yan and Li, 2021 ). Although the above inferences can be drawn from both theoretical and practical viewpoints, and a series of valuable results have been achieved in the study of customer engagement on purchase intention. However, the relationship between consumer participation and community group buying platform consumer behavior has been studied in a wide range of areas, and deeper theoretical logic, and the ways in which community group buying platform consumer involvement works still need to be further explored.

Numerous studies have also found that, for example, online interactions between consumers can enhance customer engagement and improve perceptions of the integrity and goodwill of the community group buying platform; and the use of avatars can rely on customer engagement to enhance the sense of identity and strengthen the consumers’ perceptions of the community group buying platform ( Deng, 2016 ; Hsu et al., 2018 ; Kamboj et al., 2018 ). However, previous literature has mostly focused on individual consumer factors and system characteristics, while the mediating effect of community identity as a driver of consumer use of IT products or services is lacking, and less attention has been paid to the impact of community identity on the path between user customer engagement and purchase behavior ( Li, 2019 ). In order to expand and enrich the existing research field, it is necessary to explore how consumer participation affects consumer purchase intentions in a virtual context.

Based on the above discussion, this study aims to construct a moderated mediating model around the relationship between consumer participation and purchase intention in the context of community group purchase usage by combing the research literature on consumer participation theory, community identity, and privacy concerns to explore the mediating psychological mechanisms and boundary conditions by which consumer participation affects purchase intention. Specifically, this study employs structural equation modeling (SEM) to delve into its inherent theoretical logic and influence mechanisms to address and answer the following questions. First, this study evaluates the effect of community identity on driving purchase intention in the context of community group purchase usage. Second, this study estimates the effect of different components of customer engagement (information sharing, responsiveness taking, and interpersonal interaction) on community identity in the context of community group purchasing. Third, this study discusses the role of community identity in mediating the relationship between consumer participation and purchase intention in the context of community group purchasing. Fourth, this study assesses the moderating role of customers’ privacy concerns in the relationship between their community identity and purchase intention in the context of community group purchasing.

Theoretical Foundation and Hypotheses Development

Social identity theory.

The concept of community identity is a further application and development of social identity in specific contexts ( Sun et al., 2021 ). Tajfel (1982) defined social identity as “an individual’s perception of belonging to a particular social group and that group membership has emotional and value significance for him or her.” Since then, scholars have mostly followed this definition ( Li et al., 2017 ). Tajfel and Turner (1986) found that individuals in a particular social group setting are characterized by a tendency to embody or express self-esteem. In other words, individuals possess a tendency to seek self-concept. In a group participation setting, interactions between groups can influence individual emotions, such as positive values. The values espoused by a group tend to reflect the identity of individuals within that group ( Li and Gao, 2019 ). Individual identity is a prerequisite for the establishment of group identity, and the individual’s evaluation of the group is based on the judgment of the characteristics and value propositions of other groups ( Ashforth and Mael, 1989 ). The individual’s evaluation and judgment make the group’s self-concept to be strengthened, thus constantly enhancing the individual’s perception of its superiority ( Algesheimer et al., 2005 ; Muniz and Schau, 2005 ; Bagozzi and Dholakia, 2006 ). Therefore, when the individual needs to get the superiority of the group, the individual will choose to join the group with which he or she has commonalities. In this particular group, the individual receives a sense of superiority due to membership, such as superior status, superior position, and superior image. Community identity is a special form of expression of social identity in which people participate in interaction, communication, and shopping by creating a virtual identity for themselves, which plays a key role in interpersonal communication in a technologically mediated environment ( Turkle, 1995 ). It reflects the virtual self-role that the user wants to express and present to others. On the community group buying platform, users can build and express themselves using freely scalable symbolic tools (e.g., short videos, personalized signatures, user avatars) ( Schau and Gilly, 2003 ), creating one or even multiple online “virtual identities” to present multiple aspects of themselves. As in the real world, there is a dual motivation for users to participate in the community group buying process to satisfy a sense of belonging and to preserve individuality ( Schau et al., 2009 ), i.e., the construction of a virtual identity reflects the need for community identity in the real world: self-identity through the search for differences between “I” and “we” and community identity through the search for differences between “we” and “they” ( Tajfel and Turner, 1986 ).

Community identity is one of the components of personal self-identity, reflecting the connection of the self to a specific virtual environment based on direct or indirect experience, and a spiritual dimension. This concept can be used to explain the individual’s relationship with the physical environment as a synthesis of affective and symbolic connections to a specific physical environment that determines who the user is and also contains the emotional meaning formed by causal expressions and confirmations ( Tajfel, 1982 ). The results of Algesheimer et al. (2005) , using the virtual society as the subject of the study and analyzed by SEM, indicate that consumers’ identification with the brand community has a significant positive impact on their brand purchasing behavior. Bagozzi and Dholakia (2006) suggest that community affiliation promotes brand purchasing behavior among community members in virtual communities, and Zheng et al. (2014) show that community identification can indirectly influence consumers’ brand-related purchasing behavior through brand identification. Yu (2018) states that customers’ brand community identity can influence their attitudes and behaviors toward their brands. Hu and He (2020) further validated the findings of Yu (2018) through an empirical study that brand community customer identification has a significant positive impact on their purchase intentions and behaviors. Obviously, in terms of research on the relationship between community identity and customer buying behavior. Community group buying platform is not only a platform for communication between community group buying platform and consumers, but also has gradually become an important channel for the sales of community group buying platform related products. Therefore, it is necessary to evaluate the influence mechanism of customers’ purchase intention in the context of community group buying platform usage in order for community group buying platform to grasp the market trend and maintain its competitive advantage. However, the role of community group buying platform identity on purchase intention has been studied by scholars, but this part of the research is mostly qualitative analysis rather than quantitative empirical evidence. Based on the above analysis, this study extends the applicability of community identity to this new research area, explores the influence mechanism of consumer participation on customers’ purchase intention, and focuses on the mediating role of community identity in it.

Consumer Participation

Kelley et al. (1990) argued that consumer participation connotes the behavior of information sharing between individuals in a stimulating environment. According to Belén del Río et al. (2001) , consumer participation is a behavioral concept that refers to the resources provided and the behaviors given by customers in the process of service production or delivery, and consumer participation is a process-oriented behavioral concept ( Wang, 2022 ). Based on this, Groth (2005) refined the relevant research of the aforementioned scholars and pointed out that “customer citizenship behavior is the behavior that customers give in the process of service and delivery, and divided customer citizenship behavior into two categories: in-role behavior refers to customer cooperative production behavior, while out-role behavior refers to other customer participation behaviors involved in the process of service and delivery” ( Zhang et al., 2021 ). With further research and expansion on consumer participation, consumer participation has been considered as a multidimensional construct ( Bettencourt, 1997 ; Ennew and Binks, 1999 ; Claycomb et al., 2001 ; Lloyd, 2003 ). Lloyd (2003) pointed out through empirical analysis that “consumer participation is made up of three dimensions: information sharing, responsible behavior, and interpersonal interaction.”

A further study, Yang and Chen (2017) , in which online brand communities were the subject, classified customer engagement into information sharing, responsibility taking, and interpersonal interaction based on the synthesis of previous studies. Therefore, this study concludes that customer engagement in the context of community group purchasing is “the result of a combination of both psychological and behavioral aspects of customer engagement, which refers to the resources and behaviors that customers put into the process of service production and delivery to accomplish value creation” ( Claycomb et al., 2001 ; Groth, 2005 ; Wang, 2019 ). Consumer participation consists of three main dimensions, namely information sharing, responsibility taking, and interpersonal interaction. Information sharing in the community group buying platform refers to information sharing between community members and community managers to facilitate community services to meet the needs of community members, as well as information sharing between community members regarding the purchase, use, and experience of products/services.

In general, information sharing among community members is the main part of information sharing in the community group buying platform ( Li et al., 2016 ). Responsibility taking refers to the clarification of the responsibilities of community members and community managers. For customer participation, it is more about community members cooperating with and assisting community managers to make the delivery of community services smoother, and at this time, customers have become a new platform for community group buying platform to collect information about product innovation. Since the community group buying platform is more concerned about whether customer participation can bring new ideas and inspiration to the community group buying platform, it often collects information on the development and promotion of new products in the virtual button brand community ( Bartl et al., 2012 ; Wang, 2019 ; Zheng et al., 2022 ). Interpersonal interaction in community group buying platform consists of two main aspects: one is the interaction between community members and community managers, and the other is the interaction between community members and each other. Moreover, interpersonal interaction in the community group buying environment helps to promote community services to meet the needs of community members ( Wang, 2019 ). The interpersonal interaction among community members helps to build a good community atmosphere and promote mutual trust, support, and cooperation among community members, which constitute the main aspects of interpersonal interaction in the community group buying platform.

Hypothesis Development

As a typical virtual community, users in the community group buying platform use scenario to join the community based on the demand of seeking self-worth or due to the recommendation of others, and the community identity can bring actual benefits and values to customers. The community identity can bring actual benefits and values to customers, and customers can obtain corresponding values through the community group buying platform, thus creating community identity ( Wang, 2019 ). The study concluded that consumer participation in the production of community group buying platform can reduce the manpower input of community group buying platform, which not only allows community group buying platform to allocate human resources rationally and improve productivity, but also can utilize the wisdom of customers to improve the service quality in a more targeted manner and meet the relevant needs of consumers ( Li et al., 2016 ). Aaker (1997) pointed out that community group buying platform must continuously attract customers to build brand identity to form brand equity, and brand identity comes from the process of consumer participation in the production and dissemination of community group buying platform products/services. In the scenario of community group buying platform, consumer participation helps the formation of community identity (i.e., the formation of identification with that community group buying platform) and, generating a sense of psychological gain.

With the improved standard of living in modern society, consumers spend most of their time in a state of emotional anxiety to cope with the uncertainty in their lives ( Yun et al., 2019 ). When customers generate a certain consumer behavior, this is likely to make them more anxious when there is asymmetric information about the product/service. An effective solution to this problem is the consumer participation community group buying platform, where consumers can choose to participate in a virtual brand community and use the information communication in this virtual platform to effectively combat the information asymmetry and the associated risk of uncertainty in the shopping process; thus the virtual brand community can be used to counteract the information asymmetry and the risk of uncertainty in the shopping process, thus alleviating customers’ choice phobia and anxiety and making them happy. Consumer participation in community group buying platforms improves individual knowledge and cognitive skills, thanks to the space that community group buying platforms provide for individuals to share, exchange, and transfer information. Individuals will get feedback from others when they provide information about products/services to their customers, and in this process, the knowledge and ability of customers will be exchanged and improved together. In addition, the community group buying platform provides an online platform for individuals to exchange, communicate, and discuss with each other. Therefore, in the community group buying platform scenario, individuals can get more opportunities to communicate with others and thus gain more pleasure, and these emotional benefits will enhance customers’ identification with the platform. Based on the aforementioned arguments, this study hypothesize the following:

• H1a: Information sharing is positively associated with community identity.
• H1b: Responsibility taking is positively associated with community identity.
• H1c: Interpersonal interaction is positively associated with community identity.

Social identity theory suggests that identity works by two main mechanisms: on the one hand, there is the effect on the individual’s self-esteem, and on the other hand, the distinction the individual makes between inside and outside the group ( Tajfel, 1982 ). Firstly, individuals identify themselves with a group through self-categorization and develop a sense of identity with that group. Individuals increase their self-esteem through this positive social identity ( Turner, 1985 ). Secondly, social identity allows individuals to develop in-group identity and out-group bias. In-group identification enhances individuals’ self-esteem, while out-group prejudice leads individuals to in-group advocacy and out-group exclusion behaviors. With the increase and enrichment of the literature on community identity, academics have begun to verify the applicability of social identity theory in the application of virtual communities because the concept of community group buying platform’s customer community identity, from its introduction to its continuous development, is essentially a social identity ( Wang et al., 2013 ). Since community group buying platform community identity has the same internal logic as social identity in essence, the past studies of scholars are summarized and analyzed to deeply explore the meaning and connotation of community group buying platform identity. It can be seen that in the community group buying platform scenario, customer community identity is a special kind of social identity in its essence, and therefore has the fundamental properties and functions of social identity.

According to the theory of community identity, customers divide themselves into a community group buying platform, agree and accept the system and norms of the community group buying platform, and participate in the activities of the community group buying platform. In this process, customers become dependent on and identify with the community group buying platform, and this psychological change in the process stimulates the generation of community identity. Therefore, customers see the community group buying platform as a key element of self-presentation and identity, and by enhancing the connection between themselves and the community group buying platform, they increase their perception of the value symbolic ability of the community group platform, which in turn has a significant positive impact on their intention to continue using the community group buying platform. Thus, community identity plays an important role in the construction of the community group buying platform and continues to influence customers’ behavior related to participation in the community as they shop through the platform. Studies in the context of virtual community platforms have also found that individuals’ community identity influences their intention to engage in related behaviors ( Algesheimer et al., 2005 ).

Bagozzi and Dholakia (2006) further found that community identity in virtual situations not only has a positive effect on purchase intentions, but also indirectly influences individuals’ behavioral intentions through the mediating role of brand identity. In addition, Schau et al. (2009) , based on an empirical study, pointed out that individual community identity in virtual platforms stimulates consumers’ brand perceptions and brand behaviors, and this view was confirmed by He and Yan (2018) . He and Yan (2018) study concluded that community identity has a significant effect on their loyalty behavior. It can be seen that customers’ community identity in the virtual environment will further stimulate purchase intention and other related behaviors. As such, this study proposed the following hypothesis.

• H2: Community identity is positively associated with purchase intention.

The above analysis shows that in community group purchase usage scenarios, consumer participation has a positive impact on their community identity ( Aaker, 1997 ; Carlson et al., 2008 ; Li et al., 2016 ), and customer community identity can in turn have a positive impact on purchase intention ( Belén del Río et al., 2001 ; Teo et al., 2003 ; Algesheimer et al., 2005 ; Bagozzi and Dholakia, 2006 ; Schau et al., 2009 ; He and Yan, 2018 ). It can be seen that community identity is the mediating variable that connects consumer participation and purchase intention in the context of community group buying platform usage.

Further research suggests that community identity is driven by consumer participation, which in turn motivates customers to generate a series of related behaviors, such as sharing shopping behavior, recommending products, and willingness to pay ( Lin et al., 2017 ). It can be seen that community identity plays a connecting role between consumer participation behavior and purchase intention. Based on the above analysis, this study finds that in the context of community group buying platform use, customer community identity has a connecting role between consumer participation and purchase intention, i.e., community identity plays a role in the relationship between consumer participation pairs and purchase intention. In other words, community identity plays an intermediary role in the relationship between consumer participation and purchase intention. Based on a comprehensive analysis of the above, this article thus hypothesizes:

• H3a: Community identity plays a mediating role between users’ information sharing and purchase intention.
• H3b: Community identity plays a mediating role between users’ responsibility taking and purchase intention.
• H3c: Community identity plays a mediating role between users’ interpersonal interaction and purchase intention.

In virtual environments (e.g., social media or virtual platforms), privacy concerns refer to users’ perceptions and concerns about the collection, acquisition, monitoring, and use of personal information ( Jia et al., 2021 ). With the widespread use of the Internet and big data technologies, lots of social media platforms are recognizing the value and importance of consumer data. Collecting, storing, and using consumers’ private data in various direct or indirect ways is gradually becoming a regular marketing practice for social media platforms ( Krafft et al., 2017 ; Yu, 2018 ). For example, the ability of companies to hold and analyze massive amounts of information, including consumers’ personal privacy data, is the basis for enabling behavioral targeting and product recommendations. At the same time, due to the relative lag of consumer privacy-related legislation and industry regulations in China, public policies have not yet been able to form effective legal constraints and institutional regulations on enterprises, and the importance that enterprises attach to obtaining customer privacy data and the neglect of protecting it coexist. Incidents of large-scale leakage of customer privacy are frequent and even escalate into scandals that seriously damage consumers’ interests and corporate reputation ( Janakiraman et al., 2018 ). And with the increasing public awareness of personal privacy, consumers’ concern about companies’ behavior in collecting, keeping and using personal privacy data is increasing. Moreover, they also show a tendency to give substantial feedback on these actions of companies.

Consumers not only allow companies to collect information about themselves for benefits such as convenience or experience ( Gabisch and Milne, 2014 ), but also create concerns about privacy leaks or misuse of information. The privacy concern in the consumer domain has become an inevitable and important issue in the virtual environment ( Ashforth and Mael, 1989 ; Martin and Murphy, 2017 ). It has been shown that “in general, individuals psychologically perceive that they can share and communicate carefree on a secure virtual platform because they perceive high levels of perceived value and develop a sense of belonging in their continued use of that virtual platform. The impact of the privacy breach results is that users with higher privacy concerns are concerned about the collection, control, and use of personal information during the use of the medium, thus creating a perceived loss factor for using that community group purchase” ( Wang et al., 2021 ). This illustrates that if a user’s privacy concerns are more sensitive, then that user will expect enhanced privacy protection in their purchase intentions, resulting in lower purchase intentions. It can be seen that in the context of community group purchase usage, the influence of community identity on purchase intention is not significant for users with low levels of privacy concerns. Users with high levels of privacy concerns, on the other hand, still had an effect on purchase intention, but did not have the same lack of privacy protection and security awareness as users with low levels of privacy concerns. Therefore, this study infers that the relationship between community identity and purchase intention is influenced by privacy concerns. Thus, this paper derived the following hypothesis:

• H4: Users’ privacy concern plays a moderating role between community identity and purchase intention.

In summary, this study proposes a research model as shown in Figure 1 .

Theoretical model.

Materials and Methods

Participants and procedure.

The data collection of this study fully considered the representativeness of the sample, and selected “Chengxin Youxuan,” “Duoduo Maicai,” and “Meituan Youxuan” users. This is because “Meituan Youxuan” belongs to a new generation of community group buying characterized by platform, “Chengxin Youxuan” belongs to a comprehensive community group buying integrating graphic information, video advertisement, and chat, and “Duoduo Maicai” represents a traditional community group buying platform. All three have a sizeable user base in China. In this study, 532 valid samples were collected from 12 July 2021 to 15 October 2021. The results of descriptive statistics are shown in Table 1 .

Descriptive statistical analysis.

Consumer participation, a 12-item scale, proposed by Ennew and Binks (1999) , Claycomb et al. (2001) , Groth (2005) , and Kim et al. (2010) , were adopted to measure three dimensions of consumer participation, among which 4 items were used to measure information sharing, four items for measuring responsibility taking, and four items for assessing interpersonal interaction. Community Identity scale of this research mainly refers to the research views of Zheng et al. (2017) , mainly used to measure consumers’ attitudinal preferences and emotional responses to community group purchasing, including three items. Privacy concern is modified from Son and Kim (2008) , Shin (2010) , and Tan et al. (2012) . Purchase intention, including four items, adopted from Grewal et al. (1998) and Papagiannidis et al. (2014) .

All the measurement items for this research were translated from English into Chinese following the back-translation procedure advocated by Cha et al. (2007) , and we modified the measurement items according to the actual usage situation of community group buying. These measurement items are rated on a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree).

Data Analysis and Results

Confirmatory factor analysis.

The results of the confirmatory factor analysis (CFA) are shown in Table 2 . This study evaluates and revises the measurement model of CFA according to the approach of Anderson and Gerbing (1988) via IBM SPSS AMOS 24. That is, CFA should demonstrate standardized factor loading, reliability (i.e., Cronbach’s Alpha and composite reliability), and Average Variance Extracted (AVE) for all constructs, and only after these metrics pass the tests, structural model evaluation can be performed ( Kline, 2011 ). Fornell and Lacker (1981) and Hair et al. (2017) clearly stated that when the factor loadings should be greater than 0.7, Cronbach’s Alpha and composite reliability should be greater than 0.7, and AVEs are greater than 0.5, then the measurement model has the convergent validity (as shown in Table 2 ).

Confirmatory factor analysis.

Table 3 reports the discriminant validity for the measurement model. The discriminant validity is an analysis used to test whether any two variables in the theoretical model are identical to each other. If there is no identity between any two variables, then the measurement model has reasonable discriminant validity. This study uses the generic discriminant validity analysis method, namely confidence interval method ( Torkzadeh et al., 2003 ). The confidence interval method is used to confirm the confidence interval of the correlation coefficient between variables. If the confidence interval of correlations between constructs includes zero, the empirical data can’t pass the test of discriminant validity. As Hancock and Nevitt (1999) suggested, bootstrap method was conducted in this research, 95% confidence interval of the correlation coefficient does not involve 1, which shows the discriminant validity among constructs (as shown in Table 3 ).

Discriminant validity for the measurement model.

IS, information sharing; RT, responsibility taking; II, interpersonal interaction; CI, community identity; PI, purchase intention.

Structural Model Analysis

The results of the model fit degree are shown in Table 4 . The study by Jackson et al. (2009) states that structural models should report model fit metrics as a way to assess, correct, and judge the goodness of measurement models. As suggested by Jackson et al. (2009) , Normed Chi-square (χ 2 /df), SRMR, TLI (NNFI), and CFI are the common metrics used to test the fit of research models. Therefore, academics usually use these nine metrics to test whether the model fit is good or not. In principle, the lower χ 2 is better, but since χ 2 is very sensitive to the sample size, so it is evaluated with the assistance of χ 2 /df, whose ideal value should be less than 5. The judgment criteria of all indicators are shown in Table 4 . The results of model fitness are shown in Table 4 , and all of them meet the suggested criteria of Jackson et al. (2009) . Therefore, the structural model has a good fit in this study.

Model fit criteria and the test results.

The path coefficients are shown in Table 5 . Information sharing (IS) ( b = 0.354, p < 0.001), responsibility taking (RT) ( b = 0.130, p < 0.05) and interpersonal interaction (II) ( b = 0.296, p < 0.001), are positively associated with community identity (CI). Therefore, H1a, H1b, and H1c are established. Community identity (CI) ( b = 0.832, p < 0.001) is positively associated with purchase intention (PI). Therefore, H2 is established.

Regression coefficient.

*p-value < 0.05; ***p-value < 0.001.

The results of the mediation effect analysis are shown in Table 6 . In this study, SEM was used to analyze the mediation effect, and the standard error of the mediation effect was first estimated using bootstrap estimation technique, and then the significant level of the mediation effect was further calculated. According to Hayes (2009) , a mediation effect is indicated if “0” does not include the 95% confidence interval of bias-corrected, the z -value is greater than 1.96, and the p -value is less than 0.05.

The analysis of mediation effect.

Specifically, the total effect of information sharing on purchase intention is 0.244. At the 95% confidence level, “0” does not include the bias-corrected 95% confidence interval range, the z -value > 1.96, and the p -value < 0.05. Therefore, a total effect exists. The indirect effect is 0.185, “0” does not include the bias-corrected 95% confidence interval range, the z -value > 1.96, and the p -value < 0.05. Therefore, there is an indirect effect. The direct effect is 0.058, “0” does not include the Bias-corrected 95% confidence interval range, the Z -value > 1.96, and the p -value < 0.05. Therefore, a direct effect exists.

The indirect effect of responsibility taking on purchase intention is 0.078, “0” does not include the bias-corrected 95% confidence interval range, the z -value > 1.96, and the p -value < 0.05. Therefore, there is an indirect effect. The indirect effect of interpersonal interaction on purchase intention is 0.132, “0” does not include the bias-corrected 95% confidence interval range, the z -value > 1.96, and the p -value < 0.05. Therefore, there is an indirect effect. In the same analytical approach, the results of the study show that H3a, H3b, and H3c are supported.

The moderation effect is reported in Table 7 . In the present study, privacy concern (PC) is the moderating variable. The results of SEM have shown that the moderation effect of community identity (CI) × privacy concern (PC) on purchase intention (PI) is −0.051 ( z = | −2.010| > 1.96, p < 0.05), implying the presence of a positive moderating effect of privacy concern (PC) on the relationship between community identity (CI) and purchase intention (PI). Specifically, the slope of community identity (CI) on purchase intention (PI) increases negatively by −0.051 units for each 1-unit increase in the moderating variable privacy concern (PC). That is, community identity (CI) has a negative moderating effect. Therefore, H4 is verified.

The analysis of moderation effect.

*p-value < 0.05; **p-value < 0.01; ***p-value < 0.001.

Research Results and Discussion

First, the effect of community group buying platform consumer participation on community identity is verified. The results of this study indicate that consumer participation has an influential role on community identity, which is consistent with existing research ( Yun et al., 2019 ). In the context of community group buying platform, asymmetric information about products/services is likely to make customers more anxious when they engage in a certain consumer behavior. The effective way to solve this problem is the consumer participation community group buying platform, where consumers can choose to participate in the virtual brand community and use the information communication in the virtual platform to effectively combat the information asymmetry and the associated risk of uncertainty in the shopping process, so as to improve the knowledge and cognitive ability of individuals, which is due to the fact that the community group buying platform provides individuals with a space for information sharing, communication, and transmission. Individuals will get feedback from others when they provide product/service related information to other users, and in this process, the knowledge and ability of customers will be exchanged and improved together. In addition, the virtual group buying platform provides a place for users to be able to communicate. This allows users to have the opportunity to communicate with others and gain more pleasure, and these emotional benefits will enhance the customers’ identification with the platform. The community group buying platform creates an interactive context for customers, providing them with a channel to communicate and share with other customers, thus making them happier and increasing their sense of identity and belonging to the community platform. Therefore, the empirical test results of this study found that consumer participation in community group buying platform can positively affect their community identity. Thus, it is verified that consumer participation in community group buying platform context has a positive effect on customers’ community identity and is a very significant effect on enhancing members’ community identity.

Second, the effect of community group buying platform customers’ community identity on purchase intention was examined. The result indicates that customer community identification has a significant effect on their purchase intention in the context of community group buying platform usage. The findings are consistent with those of Belén del Río et al. (2001) , Bagozzi and Dholakia (2006) , and He and Yan (2018) . Belén del Río et al. (2001) examined the relationship between community identity and purchase intention at both the individual and social identity levels, and found that both individual and social identity have a facilitating effect on customers’ purchase intention, and the effect of individual identity on purchase intention is stronger than that of social identity. Bagozzi and Dholakia (2006) further found that the virtual context. Bagozzi and Dholakia (2006) further found that community identity in virtual situations not only has a positive effect on purchase intentions, but also indirectly influences individuals’ behavioral intentions through the mediating role of brand identity. In addition, Schau et al. (2009) , based on an empirical study, pointed out that individual community identity in virtual platforms stimulates consumers’ brand perception and brand behavior, which is confirmed by He and Yan (2018) . He and Yan’s (2018) study concluded that community identity has a significant effect on their loyalty behavior. Therefore, this study concludes that community group buying platform context, customers’ community identity stimulates their purchase intention and thus has a significant impact.

Third, the mediating role of community identity between consumer participation and purchase intention is examined. This result shows that community identity mediates the relationship between consumer participation and purchase intention, which is consistent with existing research. Lin et al. (2017) further showed through an empirical study that individual and social identities of customers are influenced by customer participation, which leads to a series of related behaviors, such as sharing shopping behavior, recommending products, and willingness to pay. It can be seen that community identity plays a role in linking customer engagement behavior and purchase intention. Based on the above analysis, this study suggests that community identity plays a mediating role in the relationship between customer participation and purchase intention. To this end, a theoretical model of customer engagement behavior influencing users’ purchase intention through community identity in a local community group purchase context is constructed. This study uses Orange Heart Preferred, Duo Buy, and Meituan Preferred as the samples for data collection and investigation to explore the role of consumer participation on purchase intention and its boundary conditions. The results of the study revealed that community identification is an intermediate variable between consumer participation and customer purchase intention. consumer participation is a mediating effect on the purchase intention and purchase behavior of community residents, and its main means is participation in community activities. By communicating with other consumers in community activities, sharing information, and sharing responsibilities as a member of the community group buying platform, customers form a sense of identification with the community group buying platform, and on this basis, increase their purchase intentions. The results of the empirical study show that the enhancement of customers’ purchase intention by customer participation is predicated on allowing consumers to establish their identification with the community group-buying platform.

Fourth, this study proved the moderating effect of privacy concerns on the relationship between community identity and purchase intention in community group buying platform. The findings are consistent with the logical reasoning of Krafft et al. (2017) , Yu (2018) , and Wang et al. (2021) . It is hypothesized that the main reason for this may be the role of privacy concerns not only in terms of its direct impact, but also in terms of its effectiveness as an important moderating variable that can change the effect of other marketing factors on consumer behavioral responses. Moreover, there is a wide range of contexts in which privacy concerns play a moderating role ( Tan et al., 2012 ; Wang et al., 2021 ). Specifically, privacy concerns affect the strength of the role of customer community identity on the intention to use community group buying; inhibit consumers with community identity on community group buying from showing positive rather than negative attitudes; inhibit consumers from having significant resistance to community group buying based on personalized recommendation technology, which in turn inhibits purchase intention and changes the effectiveness of consumer community identity on enhancing purchase intention and behavior. Not only that, privacy concern also decreases community group buying users’ distant willingness to disclose privacy, and there is a non-negligible gap between the distant willingness and the behavior itself, which is the root cause of the contradiction between privacy attitudes and behaviors ( Hong et al., 2019 ). From the perspective of moderating focus theory, the difference in the psychological characteristic of privacy concern causes consumers to show negative moderating effects on the positive outcomes of community identification on community group buying.

Theoretical Contributions

First, this study introduces the theory of consumer participation into the context of community group purchasing, which expands the research context of consumer participation theory. In recent years, the theory of consumer participation has been a hot topic of attention in both theoretical and practical circles, and it has become a research hotspot for further enrichment and development in the subject areas of product innovation, customer curiosity, and customer innovation. However, with the booming and widespread use of community group buying platform, the differences between the community group buying platform context and the virtual environment in the broad sense are more obvious, except for the similarities, such as user characteristics as likes, comments, and retweets. For example, users’ characteristics such as liking, commenting, and retweeting, and the boundary between “virtual-reality” in the context of community group buying platform use are in constant conflict. Based on this, this study applies the theory of consumer participation to the local community group buying platform usage context, extending a new application context for the enrichment of the theory of consumer participation.

Second, a theoretical model of consumer participation influencing purchase intention is constructed. Based on the virtual context perspective, this paper constructs a theoretical model of consumer participation to predict users’ willingness to buy, discuss in depth the relationship between different components of consumer participation, community identity, and willingness to buy, and validates the relationship between community group buying and platform identity through the relevant analysis of identity theory. The relationship between platform identification and purchase intention is verified through the correlation analysis of identity theory. The results show that consumer participation stimulates the generation of their purchase intention mediated by community identity, where privacy concerns negatively moderate the effect of community identity on purchase intention. This study reveals the intrinsic mechanism of customer generation influencing purchase intention and its boundary conditions, which provides a reference for the innovative management and business practice of community group platform.

Third, this paper explores and examines the boundary conditions of privacy concerns in the process of consumer participation influencing purchase intention in the community group purchase usage context. In this study, privacy concerns are introduced as a moderating variable in the community group purchase usage context to extend the existing research on consumer participation influencing purchase intention and the intermediate mechanisms, and to confirm the variability of users’ purchase intention. This study found that users’ awareness of the role of community identity on purchase intention in the community group purchase context is inhibited by privacy concerns in addition to users’ attitudes and behaviors that change their purchase intention, indicating that users’ desire to protect privacy and their privacy control behaviors weaken user-related behaviors, and this study enriches the boundary conditions of the role of perceived value on social attachment.

Practical Implications

First, customer participation in the context of community group buying platform is an important source of purchase intention. In the community group buying platform, the personalized recommendation function of big data is used to regularly recommend shopping information for users’ preferences, which saves customers’ cost of searching for information, improves customers’ information sharing ability, cultivates users’ loyalty and satisfaction, and builds the brand of the community group platform, so as to provide actionable strategies for customers’ purchasing behaviors and stimulate their purchasing intentions through information sharing. The responsibility of the virtual brand community mainly refers to the clarification of the responsibilities of community members and community managers. For customer participation, it is more about community members cooperating and assisting the work of community managers, so as to make the delivery process of community services smoother. In addition, the community group purchase platform can carry out related activities through the community group purchase platform, so that the purchaser to participate in the interaction of the platform, in the interaction of the customer because of its knowledge and experience advantage, so that they feel as a member of the community group purchase platform, and enjoy the community group purchase platform “master” consciousness, so as to play its enthusiasm in the community group purchase platform, and then the community group purchase platform activities to cooperate, so that to a certain extent the customer becomes the community group purchase platform think tank, so that the community group purchase platform in product development and innovation has a strong competitive edge.

Second, customer community identification in community group buying platforms is a key factor influencing customers’ purchase intention. The research results show that the impact and results caused by customer participation behavior are not the same. One is that community group buying platform cannot underestimate the utility of customer participation behavior, and the other is that community group buying platform should not deliberately exaggerate the function of customer participation. Therefore, community group-buying platforms need to accurately identify the stimulus factors of customer participation behavior and measure customer participation behavior to produce relevant results, so as to reach a correct understanding of customer participation behavior. The community group buying platform should investigate the community identity factors caused by customer engagement behavior, specifically: First, the community group buying platform should identify the changes in customer community identity caused by customer engagement. For example, by participating in the activities of the community group-buying platform, users can achieve the same purpose as the community group-buying platform, whether they can recognize and identify with the values represented by the brand, whether they can agree and recognize a certain lifestyle represented by the brand, whether they can be convinced that the brand can bring a sense of superiority, etc. What’s more, it is necessary for community group buying platforms to clarify the changes caused by customer participation in their community identity. For example, by participating in the activities of the community group-buying platform, customers can show the change of community identity among customers brought by customer participation. For example, by participating in the activities of the community group-buying platform, users form a sense of belonging and identity to the community group-buying platform and consider themselves as important participants of the virtual community group-buying platform.

Third, the interpersonal interaction of customers in community group-buying platform is an effective way to form customers’ purchase. It is necessary for community group buying platforms to effectively guide and manage the interpersonal interactions of customers. Community group-buying platforms use a variety of methods to train existing and potential customers, help them improve their relevant knowledge and skills, and enable them to gain the strength to be able to participate in the service contact process. First of all, customer participation needs are classified, and customer needs are used as the starting point for targeted services to achieve the platform’s proposition of creating customer participation initiatives as well as shaping brand advantages by satisfying customer participation needs. Therefore, it is necessary to push the psychological changes of customers, cater to their psychological needs, and enhance the identification of customer participation. Second, it is important to create unique engagement experiences tailored to the customer’s characteristics. It is necessary for the community group buying platform to tailor the participation experience to meet the characteristics of the users themselves. At the same time, it is necessary to highlight the brand image of the community, so that customers can actively create brand associations and associate the participation experience, platform, and products to establish the community identity of the users of the virtual shopping platform. Third, the platform should achieve a highly interactive participation experience for customers. Ultimately, the platform and users, users and users can interact with each other through the community group shopping platform, and the community group shopping platform can monitor the changes in their needs and meet them in a timely manner according to the interaction between customers and get effective feedback messages to the platform, so as to maintain a long and positive interaction with customers and establish a sustainable customer communication mechanism.

Research Limitations and Future Research Directions

In this study, the theoretical model of consumer participation influencing purchase intention is tested through SEM, and the intrinsic mechanism of consumer participation influencing purchase intention and its boundary conditions are better verified. However, as the psychological characteristics of customers’ purchase intentions, which have strong individual experiences and feelings, involved in this study, it is obviously not enough to conduct quantitative research only from a relatively static and horizontal perspective, which cannot fully penetrate the dynamic characteristics of purchase intentions itself.” Therefore, this study encourages researchers to adopt research methods that have the advantage of dynamism, such as qualitative research, in order to better capture and describe in detail the development of purchase intention in a dynamic way.

Data Availability Statement

Author contributions.

TH: conceptualization, methodology, and validation. JL: investigation. TH and JL: formal analysis, visualization, and writing – original draft preparation, review and editing. Both authors read and agreed to the published version of the manuscript.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

• Aaker J. L. (1997). Dimensions of brand personality. J. Mark. Res. 34 347–356. [ Google Scholar ]
• Algesheimer R., Dholakia U. M., Herrmann A. (2005). The social influence of brand community:Evidence from European car clubs. J. Mark. 69 19–34. 10.1509/jmkg.69.3.19.66363 [ CrossRef ] [ Google Scholar ]
• Anderson J. C., Gerbing D. W. (1988). Structural equation modeling in practice: a review and recommended two-step approach. Psychol. Bull. 103 411–423. 10.1037/0033-2909.103.3.411 [ CrossRef ] [ Google Scholar ]
• Ashforth B. E., Mael F. (1989). Social identity theory and the organization. Acad. Manag. Rev. 14 20–39. 10.1017/cbo9781139136983.004 [ CrossRef ] [ Google Scholar ]
• Bagozzi R. P., Dholakia U. M. (2006). Antecedents and purchase consequences of customer participation in small group brand communities. Internat. J. Res. Mark. 23 45–61. 10.1016/j.ijresmar.2006.01.005 [ CrossRef ] [ Google Scholar ]
• Bartl M., Fuller J., Muhlbacher H., Ernst H. (2012). A Manager’s Perspective on Virtual Customer Integration for New Product Development. J. Prod. Innov. Manag. 29 : 6 . 10.3389/fpsyg.2022.858900 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Belén del Río A., Vazquez R., Iglesias V. (2001). The effects of brand associations on consumer response. J. Cons. Mark. 18 410–425. 10.1108/07363760110398808 [ CrossRef ] [ Google Scholar ]
• Bettencourt L. A. (1997). Customer voluntary performance: customers as partners in service delivery. J. Retail. 73 383–406. 10.1016/s0022-4359(97)90024-5 [ CrossRef ] [ Google Scholar ]
• Carlson B. D., Suter T. A., Brown T. J. (2008). Social versus psychological brand community: the role of psychological sense of brand community. J. Bus. Res. 61 284–291. 10.1016/j.jbusres.2007.06.022 [ CrossRef ] [ Google Scholar ]
• Cha E. S., Kim K. H., Erlen J. A. (2007). Translation of scales in cross-cultural research: issues and techniques. J. Adv. Nurs. 58 386–395. 04242.x 10.1111/j.1365-2648.2007 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Claycomb C., Lengnick-Hall C. A., Inks L. W. (2001). The customer as a productive resource: A pilot study and strategic implications. J. Bus. Strat. 18 47–69. 10.54155/jbs.18.1.47-70 [ CrossRef ] [ Google Scholar ]
• Deng Z. H. (2016). A Study on Factors Influencing Online Group-buying Repurchase Intentions of Chinese Consumers. East Chin. Econ. Manag. 2 141–148. 10.3969/j.issn.1007-5097.2016.02.022 [ CrossRef ] [ Google Scholar ]
• Ennew C. T., Binks M. R. (1999). Impact of participative service relationships on quality, satisfaction and retention: an exploratory study. J. Bus. Res. 46 121–132. 10.1093/milmed/usab464 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Fornell C. R., Lacker D. F. (1981). Structural equation models with unobservable variables and measurement error. J. Mark. Res. 18 382–388. 10.2307/3150980 [ CrossRef ] [ Google Scholar ]
• Gabisch J. A., Milne G. R. (2014). The impact of compensation on information ownership and privacy control. J. Consum. Mark. 31 13–26. [ Google Scholar ]
• Grewal D., Krishnan R., Baker J., Borin N. (1998). The effect of store name, brand name and price discounts on consumers’ evaluations and purchase intentions. J. Retail. 74 331–352. 10.1016/s0022-4359(99)80099-2 [ CrossRef ] [ Google Scholar ]
• Groth M. (2005). Customers as good soldiers: Examining citizenship behaviors in internet service deliveries. J. Manag. 31 7–27. 10.1177/0149206304271375 [ CrossRef ] [ Google Scholar ]
• Hair J. F., Hult G. T. M., Ringle C. M., Sarstedt M., Thiele K. O. (2017). Mirror, mirror on the wall: a comparative evaluation of composite-based structural equation modeling methods. J. Acad. Mark. Sci. 45 616–632. 10.1007/s11747-017-0517-x [ CrossRef ] [ Google Scholar ]
• Hancock G. R., Nevitt J. (1999). Bootstrapping and the identification of exogenous latent variables within structural equation models. Struct. Equ. Model. 6 394–399. 10.1080/10705519909540142 [ CrossRef ] [ Google Scholar ]
• Hayes A. F. (2009). Beyond Baron and Kenny: statistical Mediation Analysis in the New Millennium. Comm. Monogr. 76 408–420. 10.1080/03637750903310360 [ CrossRef ] [ Google Scholar ]
• He J. S., Yan H. J. (2018). Research on customer experience, brand identity and loyalty of fresh food e-commerce based on online and offline interactive integration. Comm. Times 7 73–76. 10.3969/j.issn.1002-5863.2018.07.022 [ CrossRef ] [ Google Scholar ]
• Hong W. Y., Chan F. K. Y., Thong J. Y. L. (2019). Drivers and inhibitors of internet privacy concern: A multidimensional development theory perspective. J. Bus. Ethic. 2019 : 1 . 10.1007/s10551-019-04237-1 [ CrossRef ] [ Google Scholar ]
• Hsu S. W., Qing F., Wang C. C., Hsieh H. L. (2018). Evaluation of service quality in facebook-based group-buying. Electr. Res. Appl. 2018 2830–2836. 10.1016/j.elerap.2018.01.006 [ CrossRef ] [ Google Scholar ]
• Hu B., He D. J. (2020). The Construction and Maintenance of Value Identity of Virtual Community Members. Chin. Publish. J. 3 34–40. 10.3969/j.issn.1002-4166.2020.03.010 [ CrossRef ] [ Google Scholar ]
• Hu Z. H., Shu X. G. (2021). Analysis of Media Public Opinion Emotional Tendency on the Platform of Consumers’ Community Group Purchase and the Suggestions. Theory Pract. Fin. Econ. 42 119–124. 10.16339/j.cnkihdxbcjb.2021.02.016 [ CrossRef ] [ Google Scholar ]
• Jackson D. L., Gillaspy J. A., Purc-Stephenson R. (2009). Reporting practices in confirmatory factor analysis: an overview and some recommendations. Psychol. Methods 14 6–23. 10.1037/a0014694 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Janakiraman R., Lim J. H., Rishika R. (2018). The effect of a data breach announcement on customer behavior: evidence from a multichannel retailer. J. Mark. 82 85–105. [ Google Scholar ]
• Jia R. N., Wang L. W., Fan X. C. (2021). Research on factors influencing personal information security privacy protection behavior of social network users. Mod. Intellig. 41 105–114. [ Google Scholar ]
• Kamboj S., Sarmah B., Gupta S., Dwivedi Y. (2018). Examining branding co-creation in brand communities on social media: applying the paradigm of Stimulus-Organism-Response. Internat. J. Inform. Manag. 39 169–185. 10.1016/j.ijinfomgt.2017.12.001 [ CrossRef ] [ Google Scholar ]
• Kelley S. W., Donnelly J. H., Jr., Skinner S. J. (1990). Customer participation in service production and delivery”. J. Retail. 66 315–334. [ Google Scholar ]
• Kim H. W., Gupta S., Koh J. (2010). Investigating the intention to purchase digital items in social networking communities: a customer value perspective. Inf. Manage. 8 228–234. [ Google Scholar ]
• Kline R. B. (2011). Principles and Practice of Structural Equation Modeling , 3th Edn. New York, NY: Guilford. [ Google Scholar ]
• Krafft M., Arden C. M., Verhoef P. C. (2017). Permission marketing and privacy concerns—why do customers (not) grant permissions? J. Interact. Mark. 39 39–54. [ Google Scholar ]
• Li C. Y. (2019). How social commerce constructs influence customers’ social shopping intention? An empirical study of a social commerce website. Technolog. Forecast. Soc. Chang. 2019 144282–144294. 10.1016/j.techfore.2017.11.026 [ CrossRef ] [ Google Scholar ]
• Li Q., Gao X. Y. (2019). The Users’ Sustained “Concern” Intension over Corporate Micro-blogs Based on Attachment Theory. J. Dal. Univ. Technol. 40 7–14. 10.19525/j.issn1008-407x.2019.01.002 [ CrossRef ] [ Google Scholar ]
• Li Q., Li X., Wei X. J. (2020). Study of consumer community group purchase integrating SOR and commitment trust theory. J. Xian Jiaotong Univ. 2 25–35. 10.15896/j.xjtuskxb.202002004 [ CrossRef ] [ Google Scholar ]
• Li X. G., Chen N. J., Zhang X. S. (2017). The Influence Mechanism of Virtual Brand Community’s Perceived Value on Purchase Intention of New Product. China Bus. Mark. 31 93–100. [ Google Scholar ]
• Li Y., Zhou M., Wang X. X. (2016). Customer originality value research: review, exploration and prospect. For. Econ. Manag. 3 73–85. 10.16538/j.cnki.fem.2016.03.006 [ CrossRef ] [ Google Scholar ]
• Lin J. B., Yan Y. M., Chen S. J., Luo X. (2017). Understanding the Impact of Social Commerce Website Technical Features on Repurchase Intention: a Chinese Guanxi Perspective. J. Electr. Comm. Res. 18 225–244. [ Google Scholar ]
• Lloyd A. E. (2003). The role of culture on customer participation in services. Hong Kong: The Hong Kong Polytechnic University. [ Google Scholar ]
• Martin K. D., Murphy P. E. (2017). The role of data privacy in marketing. J. Acad. Mark. Sci. 45 135–155. [ Google Scholar ]
• Muniz A. M., Schau H. J. (2005). Religiosity in the abandoned apple newton brand community. J. Cons. Res. 31 737–747. 10.1086/426607 [ CrossRef ] [ Google Scholar ]
• Papagiannidis S., See-To E., Bourlakis M. (2014). Virtual test-driving: the impact of simulated products on purchase intention. J. Ret. Cons. Servi. 21 877–887. 10.1016/j.jretconser.2014.02.010 [ CrossRef ] [ Google Scholar ]
• Schau H. J., Gilly M. C. (2003). We are what We Post? Self-Presen-tation in Personal Webspace. J. Con. Res. 30 385–404. 10.1086/378616 [ CrossRef ] [ Google Scholar ]
• Schau H. J., Muniz A. M., Arnould E. J. (2009). How brand community practices create value. J. Mark. 73 30–51. 10.1509/jmkg.73.5.30 [ CrossRef ] [ Google Scholar ]
• Shin D. H. (2010). The effects of trust, security and privacy in social networking: a security-based approach to understand the pattern of adoption. Interact. Comput. 22 428–438. [ Google Scholar ]
• Son J.-Y., Kim S. S. (2008). Internet users’ information privacy-protective responses: a taxonomy and a nomological model. MIS Q. 32 503–529. [ Google Scholar ]
• Sun B., Mao H., Kang M. (2021). Research on the Relationship Between Cross-Level Motivation, Integrated Emotion, and Virtual Knowledge Community Commitment. Front. Psychol. 12 : 563024 . 10.3389/fpsyg.2021.563024 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Tajfel H., Turner J. C. (1986). “ The Social Identity Theory of Intergroup Behavior ,” in Psychology of Intergroup Relations , eds Worchel S., Austin W. (Chicago: Nelson-Hall; ), 1986. [ Google Scholar ]
• Tajfel H. (1982). Social psychology of intergroup relations. Annu. Rev. Psychol. 1 1–39. [ Google Scholar ]
• Tan X., Qin L., Kim Y., Hsu J. (2012). Impact of Privacy Concern in Social Networking Web Sites. Internet Res. 22 211–233. 10.1108/10662241211214575 [ CrossRef ] [ Google Scholar ]
• Teo H. H., Chan H. C., Wei K. K., Zhang Z. (2003). Evaluating informationaccessibility and community adaptivity features for sustaining virtual learning communities. Int. J. Hum. Comput. Stud. 59 671–697. [ Google Scholar ]
• Torkzadeh G., Koufteros X., Pflughoeft K. (2003). Confirmatory analysis of computer self-efficacy. Struct. Equ. Mod. 10 263–275. 10.1207/s15328007sem1002_6 [ CrossRef ] [ Google Scholar ]
• Turkle S. (1995). Life on the Screen: Identity in the Age of the Internet. New York, NY: Simon & Schuster. [ Google Scholar ]
• Turner J. C. (1985). Social categorization and the self-concept: a social cognitive theory of group behavior. Adv. Group Proc. 2 77–122. [ Google Scholar ]
• Wang J. (2022). Research on the Impact of Customer Participation in Virtual Community on Service Innovation Performance. Front. Psychol. 13 : 847713 . 10.3389/fpsyg.2022.847713 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Wang L. (2019). Operating mechanism for the virtual customer engagement platform: a theoretical and empirical study. Sci. Res. Manag. 40 182–191. 10.19571/j.cnki.1000-2995.2019.07.018 [ CrossRef ] [ Google Scholar ]
• Wang Q., Yang M., Zhang W. (2021). Accessing the Influence of Perceived Value on Social Attachment: developing Country Perspective. Front. Psychol. 12 : 760774 . 10.3389/fpsyg.2021.760774 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Wang Q. Z., Yang Y., Wang Q., Ma Q. G. (2013). The effect of human image in B2C website design: an eyetracking study. Enterp. Inf. Syst. 8 582–605. [ Google Scholar ]
• Yan D. C., Li F. (2021). Creating familiar customers: the purpose of online emotional marketing in the social media era: the example of community group purchase “leader”. Mod. C omm. 43 129–134. 10.3969/j.issn.1007-8770.2021.07.021 [ CrossRef ] [ Google Scholar ]
• Yang I. M., Chen Y. (2017). Study on customer participation in city brand value co-creation strategy. Pionee. Sci. Technol. Mon. 30 75–78. 10.3969/j.issn.1665-2272.2017.15.027 [ CrossRef ] [ Google Scholar ]
• Yang M. S., Zhang W. S., Ruangkanjanases A., Zhang Y. (2021). Understanding the Mechanism of Social Attachment Role in Social Media: a Qualitative Analysis. Front. Psychol. 2021 : 720880 . 10.3389/fpsyg.2021.720880 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Yu L. (2018). Research on the Relationship between Virtual Community Support Pattern, Social Identity and Members’Continual Participation. China Bus. Market. 32 102–111. 10.14089/j.cnki.cn11-3664/f.2018.10.011 [ CrossRef ] [ Google Scholar ]
• Yun H. J., Lee G., Kim D. J. (2019). A chronological review of empirical research on personal information privacy concerns: an analysis of contexts and research constructs. Informat. Manag. 56 570–601. 10.1016/j.im.2018.10.001 [ CrossRef ] [ Google Scholar ]
• Zhang J., Qi S., Lyu B. (2021). A Receiver Perspective on Knowledge Sharing Impact on Consumer–Brand Relationship in Virtual Communities. Front. Psychol. 12 : 685959 . 10.3389/fpsyg.2021.685959 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Zheng F. C., Zhang N., Yu L. Q., Li G. F. (2017). How Does Brand Community Identity Affect Brand Loyalty and Brand Recommendation. Transdiscipl. Eng. 5 205–213. 10.3233/978-1-61499-779-5-205 [ CrossRef ] [ Google Scholar ]
• Zheng S. Y., Li J. Y., Wang H. J., Dukhaykh S., Wang L., Li B. Q., et al. (2022). Do Product Characteristics Affect Customers’ Participation in Virtual Brand Communities? An Empirical Study. Front. Psychol. 12 : 792706 . 10.3389/fpsyg.2021.792706 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Zheng S. Y., Xu C. J., Zhong C. Y. (2014). The Influence Mechanism of Virtual Brand Community’s Perceived Value on Purchase Intention of New Product: based on the Double Mediation of Group Identity and Brand Identity. China Bus. Market 3 85–89. [ Google Scholar ]

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 26 March 2024

Predicting and improving complex beer flavor through machine learning

• Michiel Schreurs   ORCID: orcid.org/0000-0002-9449-5619 1 , 2 , 3   na1 ,
• Supinya Piampongsant 1 , 2 , 3   na1 ,
• Miguel Roncoroni   ORCID: orcid.org/0000-0001-7461-1427 1 , 2 , 3   na1 ,
• Lloyd Cool   ORCID: orcid.org/0000-0001-9936-3124 1 , 2 , 3 , 4 ,
• Beatriz Herrera-Malaver   ORCID: orcid.org/0000-0002-5096-9974 1 , 2 , 3 ,
• Christophe Vanderaa   ORCID: orcid.org/0000-0001-7443-5427 4 ,
• Florian A. Theßeling 1 , 2 , 3 ,
• Łukasz Kreft   ORCID: orcid.org/0000-0001-7620-4657 5 ,
• Alexander Botzki   ORCID: orcid.org/0000-0001-6691-4233 5 ,
• Philippe Malcorps 6 ,
• Luk Daenen 6 ,
• Tom Wenseleers   ORCID: orcid.org/0000-0002-1434-861X 4 &
• Kevin J. Verstrepen   ORCID: orcid.org/0000-0002-3077-6219 1 , 2 , 3  

Nature Communications volume  15 , Article number:  2368 ( 2024 ) Cite this article

42k Accesses

786 Altmetric

Metrics details

• Chemical engineering
• Gas chromatography
• Machine learning
• Metabolomics
• Taste receptors

The perception and appreciation of food flavor depends on many interacting chemical compounds and external factors, and therefore proves challenging to understand and predict. Here, we combine extensive chemical and sensory analyses of 250 different beers to train machine learning models that allow predicting flavor and consumer appreciation. For each beer, we measure over 200 chemical properties, perform quantitative descriptive sensory analysis with a trained tasting panel and map data from over 180,000 consumer reviews to train 10 different machine learning models. The best-performing algorithm, Gradient Boosting, yields models that significantly outperform predictions based on conventional statistics and accurately predict complex food features and consumer appreciation from chemical profiles. Model dissection allows identifying specific and unexpected compounds as drivers of beer flavor and appreciation. Adding these compounds results in variants of commercial alcoholic and non-alcoholic beers with improved consumer appreciation. Together, our study reveals how big data and machine learning uncover complex links between food chemistry, flavor and consumer perception, and lays the foundation to develop novel, tailored foods with superior flavors.

Similar content being viewed by others

BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules

Rudraksh Tuwani, Somin Wadhwa & Ganesh Bagler

Sensory lexicon and aroma volatiles analysis of brewing malt

Xiaoxia Su, Miao Yu, … Tianyi Du

Predicting odor from molecular structure: a multi-label classification approach

Kushagra Saini & Venkatnarayan Ramanathan

Introduction

Predicting and understanding food perception and appreciation is one of the major challenges in food science. Accurate modeling of food flavor and appreciation could yield important opportunities for both producers and consumers, including quality control, product fingerprinting, counterfeit detection, spoilage detection, and the development of new products and product combinations (food pairing) 1 , 2 , 3 , 4 , 5 , 6 . Accurate models for flavor and consumer appreciation would contribute greatly to our scientific understanding of how humans perceive and appreciate flavor. Moreover, accurate predictive models would also facilitate and standardize existing food assessment methods and could supplement or replace assessments by trained and consumer tasting panels, which are variable, expensive and time-consuming 7 , 8 , 9 . Lastly, apart from providing objective, quantitative, accurate and contextual information that can help producers, models can also guide consumers in understanding their personal preferences 10 .

Despite the myriad of applications, predicting food flavor and appreciation from its chemical properties remains a largely elusive goal in sensory science, especially for complex food and beverages 11 , 12 . A key obstacle is the immense number of flavor-active chemicals underlying food flavor. Flavor compounds can vary widely in chemical structure and concentration, making them technically challenging and labor-intensive to quantify, even in the face of innovations in metabolomics, such as non-targeted metabolic fingerprinting 13 , 14 . Moreover, sensory analysis is perhaps even more complicated. Flavor perception is highly complex, resulting from hundreds of different molecules interacting at the physiochemical and sensorial level. Sensory perception is often non-linear, characterized by complex and concentration-dependent synergistic and antagonistic effects 15 , 16 , 17 , 18 , 19 , 20 , 21 that are further convoluted by the genetics, environment, culture and psychology of consumers 22 , 23 , 24 . Perceived flavor is therefore difficult to measure, with problems of sensitivity, accuracy, and reproducibility that can only be resolved by gathering sufficiently large datasets 25 . Trained tasting panels are considered the prime source of quality sensory data, but require meticulous training, are low throughput and high cost. Public databases containing consumer reviews of food products could provide a valuable alternative, especially for studying appreciation scores, which do not require formal training 25 . Public databases offer the advantage of amassing large amounts of data, increasing the statistical power to identify potential drivers of appreciation. However, public datasets suffer from biases, including a bias in the volunteers that contribute to the database, as well as confounding factors such as price, cult status and psychological conformity towards previous ratings of the product.

Classical multivariate statistics and machine learning methods have been used to predict flavor of specific compounds by, for example, linking structural properties of a compound to its potential biological activities or linking concentrations of specific compounds to sensory profiles 1 , 26 . Importantly, most previous studies focused on predicting organoleptic properties of single compounds (often based on their chemical structure) 27 , 28 , 29 , 30 , 31 , 32 , 33 , thus ignoring the fact that these compounds are present in a complex matrix in food or beverages and excluding complex interactions between compounds. Moreover, the classical statistics commonly used in sensory science 34 , 35 , 36 , 37 , 38 , 39 require a large sample size and sufficient variance amongst predictors to create accurate models. They are not fit for studying an extensive set of hundreds of interacting flavor compounds, since they are sensitive to outliers, have a high tendency to overfit and are less suited for non-linear and discontinuous relationships 40 .

In this study, we combine extensive chemical analyses and sensory data of a set of different commercial beers with machine learning approaches to develop models that predict taste, smell, mouthfeel and appreciation from compound concentrations. Beer is particularly suited to model the relationship between chemistry, flavor and appreciation. First, beer is a complex product, consisting of thousands of flavor compounds that partake in complex sensory interactions 41 , 42 , 43 . This chemical diversity arises from the raw materials (malt, yeast, hops, water and spices) and biochemical conversions during the brewing process (kilning, mashing, boiling, fermentation, maturation and aging) 44 , 45 . Second, the advent of the internet saw beer consumers embrace online review platforms, such as RateBeer (ZX Ventures, Anheuser-Busch InBev SA/NV) and BeerAdvocate (Next Glass, inc.). In this way, the beer community provides massive data sets of beer flavor and appreciation scores, creating extraordinarily large sensory databases to complement the analyses of our professional sensory panel. Specifically, we characterize over 200 chemical properties of 250 commercial beers, spread across 22 beer styles, and link these to the descriptive sensory profiling data of a 16-person in-house trained tasting panel and data acquired from over 180,000 public consumer reviews. These unique and extensive datasets enable us to train a suite of machine learning models to predict flavor and appreciation from a beer’s chemical profile. Dissection of the best-performing models allows us to pinpoint specific compounds as potential drivers of beer flavor and appreciation. Follow-up experiments confirm the importance of these compounds and ultimately allow us to significantly improve the flavor and appreciation of selected commercial beers. Together, our study represents a significant step towards understanding complex flavors and reinforces the value of machine learning to develop and refine complex foods. In this way, it represents a stepping stone for further computer-aided food engineering applications 46 .

To generate a comprehensive dataset on beer flavor, we selected 250 commercial Belgian beers across 22 different beer styles (Supplementary Fig.  S1 ). Beers with ≤ 4.2% alcohol by volume (ABV) were classified as non-alcoholic and low-alcoholic. Blonds and Tripels constitute a significant portion of the dataset (12.4% and 11.2%, respectively) reflecting their presence on the Belgian beer market and the heterogeneity of beers within these styles. By contrast, lager beers are less diverse and dominated by a handful of brands. Rare styles such as Brut or Faro make up only a small fraction of the dataset (2% and 1%, respectively) because fewer of these beers are produced and because they are dominated by distinct characteristics in terms of flavor and chemical composition.

Extensive analysis identifies relationships between chemical compounds in beer

For each beer, we measured 226 different chemical properties, including common brewing parameters such as alcohol content, iso-alpha acids, pH, sugar concentration 47 , and over 200 flavor compounds (Methods, Supplementary Table  S1 ). A large portion (37.2%) are terpenoids arising from hopping, responsible for herbal and fruity flavors 16 , 48 . A second major category are yeast metabolites, such as esters and alcohols, that result in fruity and solvent notes 48 , 49 , 50 . Other measured compounds are primarily derived from malt, or other microbes such as non- Saccharomyces yeasts and bacteria (‘wild flora’). Compounds that arise from spices or staling are labeled under ‘Others’. Five attributes (caloric value, total acids and total ester, hop aroma and sulfur compounds) are calculated from multiple individually measured compounds.

As a first step in identifying relationships between chemical properties, we determined correlations between the concentrations of the compounds (Fig.  1 , upper panel, Supplementary Data  1 and 2 , and Supplementary Fig.  S2 . For the sake of clarity, only a subset of the measured compounds is shown in Fig.  1 ). Compounds of the same origin typically show a positive correlation, while absence of correlation hints at parameters varying independently. For example, the hop aroma compounds citronellol, and alpha-terpineol show moderate correlations with each other (Spearman’s rho=0.39 and 0.57), but not with the bittering hop component iso-alpha acids (Spearman’s rho=0.16 and −0.07). This illustrates how brewers can independently modify hop aroma and bitterness by selecting hop varieties and dosage time. If hops are added early in the boiling phase, chemical conversions increase bitterness while aromas evaporate, conversely, late addition of hops preserves aroma but limits bitterness 51 . Similarly, hop-derived iso-alpha acids show a strong anti-correlation with lactic acid and acetic acid, likely reflecting growth inhibition of lactic acid and acetic acid bacteria, or the consequent use of fewer hops in sour beer styles, such as West Flanders ales and Fruit beers, that rely on these bacteria for their distinct flavors 52 . Finally, yeast-derived esters (ethyl acetate, ethyl decanoate, ethyl hexanoate, ethyl octanoate) and alcohols (ethanol, isoamyl alcohol, isobutanol, and glycerol), correlate with Spearman coefficients above 0.5, suggesting that these secondary metabolites are correlated with the yeast genetic background and/or fermentation parameters and may be difficult to influence individually, although the choice of yeast strain may offer some control 53 .

Spearman rank correlations are shown. Descriptors are grouped according to their origin (malt (blue), hops (green), yeast (red), wild flora (yellow), Others (black)), and sensory aspect (aroma, taste, palate, and overall appreciation). Please note that for the chemical compounds, for the sake of clarity, only a subset of the total number of measured compounds is shown, with an emphasis on the key compounds for each source. For more details, see the main text and Methods section. Chemical data can be found in Supplementary Data  1 , correlations between all chemical compounds are depicted in Supplementary Fig.  S2 and correlation values can be found in Supplementary Data  2 . See Supplementary Data  4 for sensory panel assessments and Supplementary Data  5 for correlation values between all sensory descriptors.

Interestingly, different beer styles show distinct patterns for some flavor compounds (Supplementary Fig.  S3 ). These observations agree with expectations for key beer styles, and serve as a control for our measurements. For instance, Stouts generally show high values for color (darker), while hoppy beers contain elevated levels of iso-alpha acids, compounds associated with bitter hop taste. Acetic and lactic acid are not prevalent in most beers, with notable exceptions such as Kriek, Lambic, Faro, West Flanders ales and Flanders Old Brown, which use acid-producing bacteria ( Lactobacillus and Pediococcus ) or unconventional yeast ( Brettanomyces ) 54 , 55 . Glycerol, ethanol and esters show similar distributions across all beer styles, reflecting their common origin as products of yeast metabolism during fermentation 45 , 53 . Finally, low/no-alcohol beers contain low concentrations of glycerol and esters. This is in line with the production process for most of the low/no-alcohol beers in our dataset, which are produced through limiting fermentation or by stripping away alcohol via evaporation or dialysis, with both methods having the unintended side-effect of reducing the amount of flavor compounds in the final beer 56 , 57 .

Besides expected associations, our data also reveals less trivial associations between beer styles and specific parameters. For example, geraniol and citronellol, two monoterpenoids responsible for citrus, floral and rose flavors and characteristic of Citra hops, are found in relatively high amounts in Christmas, Saison, and Brett/co-fermented beers, where they may originate from terpenoid-rich spices such as coriander seeds instead of hops 58 .

Tasting panel assessments reveal sensorial relationships in beer

To assess the sensory profile of each beer, a trained tasting panel evaluated each of the 250 beers for 50 sensory attributes, including different hop, malt and yeast flavors, off-flavors and spices. Panelists used a tasting sheet (Supplementary Data  3 ) to score the different attributes. Panel consistency was evaluated by repeating 12 samples across different sessions and performing ANOVA. In 95% of cases no significant difference was found across sessions ( p  > 0.05), indicating good panel consistency (Supplementary Table  S2 ).

Aroma and taste perception reported by the trained panel are often linked (Fig.  1 , bottom left panel and Supplementary Data  4 and 5 ), with high correlations between hops aroma and taste (Spearman’s rho=0.83). Bitter taste was found to correlate with hop aroma and taste in general (Spearman’s rho=0.80 and 0.69), and particularly with “grassy” noble hops (Spearman’s rho=0.75). Barnyard flavor, most often associated with sour beers, is identified together with stale hops (Spearman’s rho=0.97) that are used in these beers. Lactic and acetic acid, which often co-occur, are correlated (Spearman’s rho=0.66). Interestingly, sweetness and bitterness are anti-correlated (Spearman’s rho = −0.48), confirming the hypothesis that they mask each other 59 , 60 . Beer body is highly correlated with alcohol (Spearman’s rho = 0.79), and overall appreciation is found to correlate with multiple aspects that describe beer mouthfeel (alcohol, carbonation; Spearman’s rho= 0.32, 0.39), as well as with hop and ester aroma intensity (Spearman’s rho=0.39 and 0.35).

Similar to the chemical analyses, sensorial analyses confirmed typical features of specific beer styles (Supplementary Fig.  S4 ). For example, sour beers (Faro, Flanders Old Brown, Fruit beer, Kriek, Lambic, West Flanders ale) were rated acidic, with flavors of both acetic and lactic acid. Hoppy beers were found to be bitter and showed hop-associated aromas like citrus and tropical fruit. Malt taste is most detected among scotch, stout/porters, and strong ales, while low/no-alcohol beers, which often have a reputation for being ‘worty’ (reminiscent of unfermented, sweet malt extract) appear in the middle. Unsurprisingly, hop aromas are most strongly detected among hoppy beers. Like its chemical counterpart (Supplementary Fig.  S3 ), acidity shows a right-skewed distribution, with the most acidic beers being Krieks, Lambics, and West Flanders ales.

Tasting panel assessments of specific flavors correlate with chemical composition

We find that the concentrations of several chemical compounds strongly correlate with specific aroma or taste, as evaluated by the tasting panel (Fig.  2 , Supplementary Fig.  S5 , Supplementary Data  6 ). In some cases, these correlations confirm expectations and serve as a useful control for data quality. For example, iso-alpha acids, the bittering compounds in hops, strongly correlate with bitterness (Spearman’s rho=0.68), while ethanol and glycerol correlate with tasters’ perceptions of alcohol and body, the mouthfeel sensation of fullness (Spearman’s rho=0.82/0.62 and 0.72/0.57 respectively) and darker color from roasted malts is a good indication of malt perception (Spearman’s rho=0.54).

Heatmap colors indicate Spearman’s Rho. Axes are organized according to sensory categories (aroma, taste, mouthfeel, overall), chemical categories and chemical sources in beer (malt (blue), hops (green), yeast (red), wild flora (yellow), Others (black)). See Supplementary Data  6 for all correlation values.

Interestingly, for some relationships between chemical compounds and perceived flavor, correlations are weaker than expected. For example, the rose-smelling phenethyl acetate only weakly correlates with floral aroma. This hints at more complex relationships and interactions between compounds and suggests a need for a more complex model than simple correlations. Lastly, we uncovered unexpected correlations. For instance, the esters ethyl decanoate and ethyl octanoate appear to correlate slightly with hop perception and bitterness, possibly due to their fruity flavor. Iron is anti-correlated with hop aromas and bitterness, most likely because it is also anti-correlated with iso-alpha acids. This could be a sign of metal chelation of hop acids 61 , given that our analyses measure unbound hop acids and total iron content, or could result from the higher iron content in dark and Fruit beers, which typically have less hoppy and bitter flavors 62 .

Public consumer reviews complement expert panel data

To complement and expand the sensory data of our trained tasting panel, we collected 180,000 reviews of our 250 beers from the online consumer review platform RateBeer. This provided numerical scores for beer appearance, aroma, taste, palate, overall quality as well as the average overall score.

Public datasets are known to suffer from biases, such as price, cult status and psychological conformity towards previous ratings of a product. For example, prices correlate with appreciation scores for these online consumer reviews (rho=0.49, Supplementary Fig.  S6 ), but not for our trained tasting panel (rho=0.19). This suggests that prices affect consumer appreciation, which has been reported in wine 63 , while blind tastings are unaffected. Moreover, we observe that some beer styles, like lagers and non-alcoholic beers, generally receive lower scores, reflecting that online reviewers are mostly beer aficionados with a preference for specialty beers over lager beers. In general, we find a modest correlation between our trained panel’s overall appreciation score and the online consumer appreciation scores (Fig.  3 , rho=0.29). Apart from the aforementioned biases in the online datasets, serving temperature, sample freshness and surroundings, which are all tightly controlled during the tasting panel sessions, can vary tremendously across online consumers and can further contribute to (among others, appreciation) differences between the two categories of tasters. Importantly, in contrast to the overall appreciation scores, for many sensory aspects the results from the professional panel correlated well with results obtained from RateBeer reviews. Correlations were highest for features that are relatively easy to recognize even for untrained tasters, like bitterness, sweetness, alcohol and malt aroma (Fig.  3 and below).

RateBeer text mining results can be found in Supplementary Data  7 . Rho values shown are Spearman correlation values, with asterisks indicating significant correlations ( p  < 0.05, two-sided). All p values were smaller than 0.001, except for Esters aroma (0.0553), Esters taste (0.3275), Esters aroma—banana (0.0019), Coriander (0.0508) and Diacetyl (0.0134).

Besides collecting consumer appreciation from these online reviews, we developed automated text analysis tools to gather additional data from review texts (Supplementary Data  7 ). Processing review texts on the RateBeer database yielded comparable results to the scores given by the trained panel for many common sensory aspects, including acidity, bitterness, sweetness, alcohol, malt, and hop tastes (Fig.  3 ). This is in line with what would be expected, since these attributes require less training for accurate assessment and are less influenced by environmental factors such as temperature, serving glass and odors in the environment. Consumer reviews also correlate well with our trained panel for 4-vinyl guaiacol, a compound associated with a very characteristic aroma. By contrast, correlations for more specific aromas like ester, coriander or diacetyl are underrepresented in the online reviews, underscoring the importance of using a trained tasting panel and standardized tasting sheets with explicit factors to be scored for evaluating specific aspects of a beer. Taken together, our results suggest that public reviews are trustworthy for some, but not all, flavor features and can complement or substitute taste panel data for these sensory aspects.

Models can predict beer sensory profiles from chemical data

The rich datasets of chemical analyses, tasting panel assessments and public reviews gathered in the first part of this study provided us with a unique opportunity to develop predictive models that link chemical data to sensorial features. Given the complexity of beer flavor, basic statistical tools such as correlations or linear regression may not always be the most suitable for making accurate predictions. Instead, we applied different machine learning models that can model both simple linear and complex interactive relationships. Specifically, we constructed a set of regression models to predict (a) trained panel scores for beer flavor and quality and (b) public reviews’ appreciation scores from beer chemical profiles. We trained and tested 10 different models (Methods), 3 linear regression-based models (simple linear regression with first-order interactions (LR), lasso regression with first-order interactions (Lasso), partial least squares regressor (PLSR)), 5 decision tree models (AdaBoost regressor (ABR), extra trees (ET), gradient boosting regressor (GBR), random forest (RF) and XGBoost regressor (XGBR)), 1 support vector regression (SVR), and 1 artificial neural network (ANN) model.

To compare the performance of our machine learning models, the dataset was randomly split into a training and test set, stratified by beer style. After a model was trained on data in the training set, its performance was evaluated on its ability to predict the test dataset obtained from multi-output models (based on the coefficient of determination, see Methods). Additionally, individual-attribute models were ranked per descriptor and the average rank was calculated, as proposed by Korneva et al. 64 . Importantly, both ways of evaluating the models’ performance agreed in general. Performance of the different models varied (Table  1 ). It should be noted that all models perform better at predicting RateBeer results than results from our trained tasting panel. One reason could be that sensory data is inherently variable, and this variability is averaged out with the large number of public reviews from RateBeer. Additionally, all tree-based models perform better at predicting taste than aroma. Linear models (LR) performed particularly poorly, with negative R 2 values, due to severe overfitting (training set R 2  = 1). Overfitting is a common issue in linear models with many parameters and limited samples, especially with interaction terms further amplifying the number of parameters. L1 regularization (Lasso) successfully overcomes this overfitting, out-competing multiple tree-based models on the RateBeer dataset. Similarly, the dimensionality reduction of PLSR avoids overfitting and improves performance, to some extent. Still, tree-based models (ABR, ET, GBR, RF and XGBR) show the best performance, out-competing the linear models (LR, Lasso, PLSR) commonly used in sensory science 65 .

GBR models showed the best overall performance in predicting sensory responses from chemical information, with R 2 values up to 0.75 depending on the predicted sensory feature (Supplementary Table  S4 ). The GBR models predict consumer appreciation (RateBeer) better than our trained panel’s appreciation (R 2 value of 0.67 compared to R 2 value of 0.09) (Supplementary Table  S3 and Supplementary Table  S4 ). ANN models showed intermediate performance, likely because neural networks typically perform best with larger datasets 66 . The SVR shows intermediate performance, mostly due to the weak predictions of specific attributes that lower the overall performance (Supplementary Table  S4 ).

Model dissection identifies specific, unexpected compounds as drivers of consumer appreciation

Next, we leveraged our models to infer important contributors to sensory perception and consumer appreciation. Consumer preference is a crucial sensory aspects, because a product that shows low consumer appreciation scores often does not succeed commercially 25 . Additionally, the requirement for a large number of representative evaluators makes consumer trials one of the more costly and time-consuming aspects of product development. Hence, a model for predicting chemical drivers of overall appreciation would be a welcome addition to the available toolbox for food development and optimization.

Since GBR models on our RateBeer dataset showed the best overall performance, we focused on these models. Specifically, we used two approaches to identify important contributors. First, rankings of the most important predictors for each sensorial trait in the GBR models were obtained based on impurity-based feature importance (mean decrease in impurity). High-ranked parameters were hypothesized to be either the true causal chemical properties underlying the trait, to correlate with the actual causal properties, or to take part in sensory interactions affecting the trait 67 (Fig.  4A ). In a second approach, we used SHAP 68 to determine which parameters contributed most to the model for making predictions of consumer appreciation (Fig.  4B ). SHAP calculates parameter contributions to model predictions on a per-sample basis, which can be aggregated into an importance score.

A The impurity-based feature importance (mean deviance in impurity, MDI) calculated from the Gradient Boosting Regression (GBR) model predicting RateBeer appreciation scores. The top 15 highest ranked chemical properties are shown. B SHAP summary plot for the top 15 parameters contributing to our GBR model. Each point on the graph represents a sample from our dataset. The color represents the concentration of that parameter, with bluer colors representing low values and redder colors representing higher values. Greater absolute values on the horizontal axis indicate a higher impact of the parameter on the prediction of the model. C Spearman correlations between the 15 most important chemical properties and consumer overall appreciation. Numbers indicate the Spearman Rho correlation coefficient, and the rank of this correlation compared to all other correlations. The top 15 important compounds were determined using SHAP (panel B).

Both approaches identified ethyl acetate as the most predictive parameter for beer appreciation (Fig.  4 ). Ethyl acetate is the most abundant ester in beer with a typical ‘fruity’, ‘solvent’ and ‘alcoholic’ flavor, but is often considered less important than other esters like isoamyl acetate. The second most important parameter identified by SHAP is ethanol, the most abundant beer compound after water. Apart from directly contributing to beer flavor and mouthfeel, ethanol drastically influences the physical properties of beer, dictating how easily volatile compounds escape the beer matrix to contribute to beer aroma 69 . Importantly, it should also be noted that the importance of ethanol for appreciation is likely inflated by the very low appreciation scores of non-alcoholic beers (Supplementary Fig.  S4 ). Despite not often being considered a driver of beer appreciation, protein level also ranks highly in both approaches, possibly due to its effect on mouthfeel and body 70 . Lactic acid, which contributes to the tart taste of sour beers, is the fourth most important parameter identified by SHAP, possibly due to the generally high appreciation of sour beers in our dataset.

Interestingly, some of the most important predictive parameters for our model are not well-established as beer flavors or are even commonly regarded as being negative for beer quality. For example, our models identify methanethiol and ethyl phenyl acetate, an ester commonly linked to beer staling 71 , as a key factor contributing to beer appreciation. Although there is no doubt that high concentrations of these compounds are considered unpleasant, the positive effects of modest concentrations are not yet known 72 , 73 .

To compare our approach to conventional statistics, we evaluated how well the 15 most important SHAP-derived parameters correlate with consumer appreciation (Fig.  4C ). Interestingly, only 6 of the properties derived by SHAP rank amongst the top 15 most correlated parameters. For some chemical compounds, the correlations are so low that they would have likely been considered unimportant. For example, lactic acid, the fourth most important parameter, shows a bimodal distribution for appreciation, with sour beers forming a separate cluster, that is missed entirely by the Spearman correlation. Additionally, the correlation plots reveal outliers, emphasizing the need for robust analysis tools. Together, this highlights the need for alternative models, like the Gradient Boosting model, that better grasp the complexity of (beer) flavor.

Finally, to observe the relationships between these chemical properties and their predicted targets, partial dependence plots were constructed for the six most important predictors of consumer appreciation 74 , 75 , 76 (Supplementary Fig.  S7 ). One-way partial dependence plots show how a change in concentration affects the predicted appreciation. These plots reveal an important limitation of our models: appreciation predictions remain constant at ever-increasing concentrations. This implies that once a threshold concentration is reached, further increasing the concentration does not affect appreciation. This is false, as it is well-documented that certain compounds become unpleasant at high concentrations, including ethyl acetate (‘nail polish’) 77 and methanethiol (‘sulfury’ and ‘rotten cabbage’) 78 . The inability of our models to grasp that flavor compounds have optimal levels, above which they become negative, is a consequence of working with commercial beer brands where (off-)flavors are rarely too high to negatively impact the product. The two-way partial dependence plots show how changing the concentration of two compounds influences predicted appreciation, visualizing their interactions (Supplementary Fig.  S7 ). In our case, the top 5 parameters are dominated by additive or synergistic interactions, with high concentrations for both compounds resulting in the highest predicted appreciation.

To assess the robustness of our best-performing models and model predictions, we performed 100 iterations of the GBR, RF and ET models. In general, all iterations of the models yielded similar performance (Supplementary Fig.  S8 ). Moreover, the main predictors (including the top predictors ethanol and ethyl acetate) remained virtually the same, especially for GBR and RF. For the iterations of the ET model, we did observe more variation in the top predictors, which is likely a consequence of the model’s inherent random architecture in combination with co-correlations between certain predictors. However, even in this case, several of the top predictors (ethanol and ethyl acetate) remain unchanged, although their rank in importance changes (Supplementary Fig.  S8 ).

Next, we investigated if a combination of RateBeer and trained panel data into one consolidated dataset would lead to stronger models, under the hypothesis that such a model would suffer less from bias in the datasets. A GBR model was trained to predict appreciation on the combined dataset. This model underperformed compared to the RateBeer model, both in the native case and when including a dataset identifier (R 2  = 0.67, 0.26 and 0.42 respectively). For the latter, the dataset identifier is the most important feature (Supplementary Fig.  S9 ), while most of the feature importance remains unchanged, with ethyl acetate and ethanol ranking highest, like in the original model trained only on RateBeer data. It seems that the large variation in the panel dataset introduces noise, weakening the models’ performances and reliability. In addition, it seems reasonable to assume that both datasets are fundamentally different, with the panel dataset obtained by blind tastings by a trained professional panel.

Lastly, we evaluated whether beer style identifiers would further enhance the model’s performance. A GBR model was trained with parameters that explicitly encoded the styles of the samples. This did not improve model performance (R2 = 0.66 with style information vs R2 = 0.67). The most important chemical features are consistent with the model trained without style information (eg. ethanol and ethyl acetate), and with the exception of the most preferred (strong ale) and least preferred (low/no-alcohol) styles, none of the styles were among the most important features (Supplementary Fig.  S9 , Supplementary Table  S5 and S6 ). This is likely due to a combination of style-specific chemical signatures, such as iso-alpha acids and lactic acid, that implicitly convey style information to the original models, as well as the low number of samples belonging to some styles, making it difficult for the model to learn style-specific patterns. Moreover, beer styles are not rigorously defined, with some styles overlapping in features and some beers being misattributed to a specific style, all of which leads to more noise in models that use style parameters.

Model validation

To test if our predictive models give insight into beer appreciation, we set up experiments aimed at improving existing commercial beers. We specifically selected overall appreciation as the trait to be examined because of its complexity and commercial relevance. Beer flavor comprises a complex bouquet rather than single aromas and tastes 53 . Hence, adding a single compound to the extent that a difference is noticeable may lead to an unbalanced, artificial flavor. Therefore, we evaluated the effect of combinations of compounds. Because Blond beers represent the most extensive style in our dataset, we selected a beer from this style as the starting material for these experiments (Beer 64 in Supplementary Data  1 ).

In the first set of experiments, we adjusted the concentrations of compounds that made up the most important predictors of overall appreciation (ethyl acetate, ethanol, lactic acid, ethyl phenyl acetate) together with correlated compounds (ethyl hexanoate, isoamyl acetate, glycerol), bringing them up to 95 th percentile ethanol-normalized concentrations (Methods) within the Blond group (‘Spiked’ concentration in Fig.  5A ). Compared to controls, the spiked beers were found to have significantly improved overall appreciation among trained panelists, with panelist noting increased intensity of ester flavors, sweetness, alcohol, and body fullness (Fig.  5B ). To disentangle the contribution of ethanol to these results, a second experiment was performed without the addition of ethanol. This resulted in a similar outcome, including increased perception of alcohol and overall appreciation.

Adding the top chemical compounds, identified as best predictors of appreciation by our model, into poorly appreciated beers results in increased appreciation from our trained panel. Results of sensory tests between base beers and those spiked with compounds identified as the best predictors by the model. A Blond and Non/Low-alcohol (0.0% ABV) base beers were brought up to 95th-percentile ethanol-normalized concentrations within each style. B For each sensory attribute, tasters indicated the more intense sample and selected the sample they preferred. The numbers above the bars correspond to the p values that indicate significant changes in perceived flavor (two-sided binomial test: alpha 0.05, n  = 20 or 13).

In a last experiment, we tested whether using the model’s predictions can boost the appreciation of a non-alcoholic beer (beer 223 in Supplementary Data  1 ). Again, the addition of a mixture of predicted compounds (omitting ethanol, in this case) resulted in a significant increase in appreciation, body, ester flavor and sweetness.

Predicting flavor and consumer appreciation from chemical composition is one of the ultimate goals of sensory science. A reliable, systematic and unbiased way to link chemical profiles to flavor and food appreciation would be a significant asset to the food and beverage industry. Such tools would substantially aid in quality control and recipe development, offer an efficient and cost-effective alternative to pilot studies and consumer trials and would ultimately allow food manufacturers to produce superior, tailor-made products that better meet the demands of specific consumer groups more efficiently.

A limited set of studies have previously tried, to varying degrees of success, to predict beer flavor and beer popularity based on (a limited set of) chemical compounds and flavors 79 , 80 . Current sensitive, high-throughput technologies allow measuring an unprecedented number of chemical compounds and properties in a large set of samples, yielding a dataset that can train models that help close the gaps between chemistry and flavor, even for a complex natural product like beer. To our knowledge, no previous research gathered data at this scale (250 samples, 226 chemical parameters, 50 sensory attributes and 5 consumer scores) to disentangle and validate the chemical aspects driving beer preference using various machine-learning techniques. We find that modern machine learning models outperform conventional statistical tools, such as correlations and linear models, and can successfully predict flavor appreciation from chemical composition. This could be attributed to the natural incorporation of interactions and non-linear or discontinuous effects in machine learning models, which are not easily grasped by the linear model architecture. While linear models and partial least squares regression represent the most widespread statistical approaches in sensory science, in part because they allow interpretation 65 , 81 , 82 , modern machine learning methods allow for building better predictive models while preserving the possibility to dissect and exploit the underlying patterns. Of the 10 different models we trained, tree-based models, such as our best performing GBR, showed the best overall performance in predicting sensory responses from chemical information, outcompeting artificial neural networks. This agrees with previous reports for models trained on tabular data 83 . Our results are in line with the findings of Colantonio et al. who also identified the gradient boosting architecture as performing best at predicting appreciation and flavor (of tomatoes and blueberries, in their specific study) 26 . Importantly, besides our larger experimental scale, we were able to directly confirm our models’ predictions in vivo.

Our study confirms that flavor compound concentration does not always correlate with perception, suggesting complex interactions that are often missed by more conventional statistics and simple models. Specifically, we find that tree-based algorithms may perform best in developing models that link complex food chemistry with aroma. Furthermore, we show that massive datasets of untrained consumer reviews provide a valuable source of data, that can complement or even replace trained tasting panels, especially for appreciation and basic flavors, such as sweetness and bitterness. This holds despite biases that are known to occur in such datasets, such as price or conformity bias. Moreover, GBR models predict taste better than aroma. This is likely because taste (e.g. bitterness) often directly relates to the corresponding chemical measurements (e.g., iso-alpha acids), whereas such a link is less clear for aromas, which often result from the interplay between multiple volatile compounds. We also find that our models are best at predicting acidity and alcohol, likely because there is a direct relation between the measured chemical compounds (acids and ethanol) and the corresponding perceived sensorial attribute (acidity and alcohol), and because even untrained consumers are generally able to recognize these flavors and aromas.

The predictions of our final models, trained on review data, hold even for blind tastings with small groups of trained tasters, as demonstrated by our ability to validate specific compounds as drivers of beer flavor and appreciation. Since adding a single compound to the extent of a noticeable difference may result in an unbalanced flavor profile, we specifically tested our identified key drivers as a combination of compounds. While this approach does not allow us to validate if a particular single compound would affect flavor and/or appreciation, our experiments do show that this combination of compounds increases consumer appreciation.

It is important to stress that, while it represents an important step forward, our approach still has several major limitations. A key weakness of the GBR model architecture is that amongst co-correlating variables, the largest main effect is consistently preferred for model building. As a result, co-correlating variables often have artificially low importance scores, both for impurity and SHAP-based methods, like we observed in the comparison to the more randomized Extra Trees models. This implies that chemicals identified as key drivers of a specific sensory feature by GBR might not be the true causative compounds, but rather co-correlate with the actual causative chemical. For example, the high importance of ethyl acetate could be (partially) attributed to the total ester content, ethanol or ethyl hexanoate (rho=0.77, rho=0.72 and rho=0.68), while ethyl phenylacetate could hide the importance of prenyl isobutyrate and ethyl benzoate (rho=0.77 and rho=0.76). Expanding our GBR model to include beer style as a parameter did not yield additional power or insight. This is likely due to style-specific chemical signatures, such as iso-alpha acids and lactic acid, that implicitly convey style information to the original model, as well as the smaller sample size per style, limiting the power to uncover style-specific patterns. This can be partly attributed to the curse of dimensionality, where the high number of parameters results in the models mainly incorporating single parameter effects, rather than complex interactions such as style-dependent effects 67 . A larger number of samples may overcome some of these limitations and offer more insight into style-specific effects. On the other hand, beer style is not a rigid scientific classification, and beers within one style often differ a lot, which further complicates the analysis of style as a model factor.

Our study is limited to beers from Belgian breweries. Although these beers cover a large portion of the beer styles available globally, some beer styles and consumer patterns may be missing, while other features might be overrepresented. For example, many Belgian ales exhibit yeast-driven flavor profiles, which is reflected in the chemical drivers of appreciation discovered by this study. In future work, expanding the scope to include diverse markets and beer styles could lead to the identification of even more drivers of appreciation and better models for special niche products that were not present in our beer set.

In addition to inherent limitations of GBR models, there are also some limitations associated with studying food aroma. Even if our chemical analyses measured most of the known aroma compounds, the total number of flavor compounds in complex foods like beer is still larger than the subset we were able to measure in this study. For example, hop-derived thiols, that influence flavor at very low concentrations, are notoriously difficult to measure in a high-throughput experiment. Moreover, consumer perception remains subjective and prone to biases that are difficult to avoid. It is also important to stress that the models are still immature and that more extensive datasets will be crucial for developing more complete models in the future. Besides more samples and parameters, our dataset does not include any demographic information about the tasters. Including such data could lead to better models that grasp external factors like age and culture. Another limitation is that our set of beers consists of high-quality end-products and lacks beers that are unfit for sale, which limits the current model in accurately predicting products that are appreciated very badly. Finally, while models could be readily applied in quality control, their use in sensory science and product development is restrained by their inability to discern causal relationships. Given that the models cannot distinguish compounds that genuinely drive consumer perception from those that merely correlate, validation experiments are essential to identify true causative compounds.

Despite the inherent limitations, dissection of our models enabled us to pinpoint specific molecules as potential drivers of beer aroma and consumer appreciation, including compounds that were unexpected and would not have been identified using standard approaches. Important drivers of beer appreciation uncovered by our models include protein levels, ethyl acetate, ethyl phenyl acetate and lactic acid. Currently, many brewers already use lactic acid to acidify their brewing water and ensure optimal pH for enzymatic activity during the mashing process. Our results suggest that adding lactic acid can also improve beer appreciation, although its individual effect remains to be tested. Interestingly, ethanol appears to be unnecessary to improve beer appreciation, both for blond beer and alcohol-free beer. Given the growing consumer interest in alcohol-free beer, with a predicted annual market growth of >7% 84 , it is relevant for brewers to know what compounds can further increase consumer appreciation of these beers. Hence, our model may readily provide avenues to further improve the flavor and consumer appreciation of both alcoholic and non-alcoholic beers, which is generally considered one of the key challenges for future beer production.

Whereas we see a direct implementation of our results for the development of superior alcohol-free beverages and other food products, our study can also serve as a stepping stone for the development of novel alcohol-containing beverages. We want to echo the growing body of scientific evidence for the negative effects of alcohol consumption, both on the individual level by the mutagenic, teratogenic and carcinogenic effects of ethanol 85 , 86 , as well as the burden on society caused by alcohol abuse and addiction. We encourage the use of our results for the production of healthier, tastier products, including novel and improved beverages with lower alcohol contents. Furthermore, we strongly discourage the use of these technologies to improve the appreciation or addictive properties of harmful substances.

The present work demonstrates that despite some important remaining hurdles, combining the latest developments in chemical analyses, sensory analysis and modern machine learning methods offers exciting avenues for food chemistry and engineering. Soon, these tools may provide solutions in quality control and recipe development, as well as new approaches to sensory science and flavor research.

Beer selection

250 commercial Belgian beers were selected to cover the broad diversity of beer styles and corresponding diversity in chemical composition and aroma. See Supplementary Fig.  S1 .

Chemical dataset

Sample preparation.

Beers within their expiration date were purchased from commercial retailers. Samples were prepared in biological duplicates at room temperature, unless explicitly stated otherwise. Bottle pressure was measured with a manual pressure device (Steinfurth Mess-Systeme GmbH) and used to calculate CO 2 concentration. The beer was poured through two filter papers (Macherey-Nagel, 500713032 MN 713 ¼) to remove carbon dioxide and prevent spontaneous foaming. Samples were then prepared for measurements by targeted Headspace-Gas Chromatography-Flame Ionization Detector/Flame Photometric Detector (HS-GC-FID/FPD), Headspace-Solid Phase Microextraction-Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS), colorimetric analysis, enzymatic analysis, Near-Infrared (NIR) analysis, as described in the sections below. The mean values of biological duplicates are reported for each compound.

HS-GC-FID/FPD

HS-GC-FID/FPD (Shimadzu GC 2010 Plus) was used to measure higher alcohols, acetaldehyde, esters, 4-vinyl guaicol, and sulfur compounds. Each measurement comprised 5 ml of sample pipetted into a 20 ml glass vial containing 1.75 g NaCl (VWR, 27810.295). 100 µl of 2-heptanol (Sigma-Aldrich, H3003) (internal standard) solution in ethanol (Fisher Chemical, E/0650DF/C17) was added for a final concentration of 2.44 mg/L. Samples were flushed with nitrogen for 10 s, sealed with a silicone septum, stored at −80 °C and analyzed in batches of 20.

The GC was equipped with a DB-WAXetr column (length, 30 m; internal diameter, 0.32 mm; layer thickness, 0.50 µm; Agilent Technologies, Santa Clara, CA, USA) to the FID and an HP-5 column (length, 30 m; internal diameter, 0.25 mm; layer thickness, 0.25 µm; Agilent Technologies, Santa Clara, CA, USA) to the FPD. N 2 was used as the carrier gas. Samples were incubated for 20 min at 70 °C in the headspace autosampler (Flow rate, 35 cm/s; Injection volume, 1000 µL; Injection mode, split; Combi PAL autosampler, CTC analytics, Switzerland). The injector, FID and FPD temperatures were kept at 250 °C. The GC oven temperature was first held at 50 °C for 5 min and then allowed to rise to 80 °C at a rate of 5 °C/min, followed by a second ramp of 4 °C/min until 200 °C kept for 3 min and a final ramp of (4 °C/min) until 230 °C for 1 min. Results were analyzed with the GCSolution software version 2.4 (Shimadzu, Kyoto, Japan). The GC was calibrated with a 5% EtOH solution (VWR International) containing the volatiles under study (Supplementary Table  S7 ).

HS-SPME-GC-MS

HS-SPME-GC-MS (Shimadzu GCMS-QP-2010 Ultra) was used to measure additional volatile compounds, mainly comprising terpenoids and esters. Samples were analyzed by HS-SPME using a triphase DVB/Carboxen/PDMS 50/30 μm SPME fiber (Supelco Co., Bellefonte, PA, USA) followed by gas chromatography (Thermo Fisher Scientific Trace 1300 series, USA) coupled to a mass spectrometer (Thermo Fisher Scientific ISQ series MS) equipped with a TriPlus RSH autosampler. 5 ml of degassed beer sample was placed in 20 ml vials containing 1.75 g NaCl (VWR, 27810.295). 5 µl internal standard mix was added, containing 2-heptanol (1 g/L) (Sigma-Aldrich, H3003), 4-fluorobenzaldehyde (1 g/L) (Sigma-Aldrich, 128376), 2,3-hexanedione (1 g/L) (Sigma-Aldrich, 144169) and guaiacol (1 g/L) (Sigma-Aldrich, W253200) in ethanol (Fisher Chemical, E/0650DF/C17). Each sample was incubated at 60 °C in the autosampler oven with constant agitation. After 5 min equilibration, the SPME fiber was exposed to the sample headspace for 30 min. The compounds trapped on the fiber were thermally desorbed in the injection port of the chromatograph by heating the fiber for 15 min at 270 °C.

The GC-MS was equipped with a low polarity RXi-5Sil MS column (length, 20 m; internal diameter, 0.18 mm; layer thickness, 0.18 µm; Restek, Bellefonte, PA, USA). Injection was performed in splitless mode at 320 °C, a split flow of 9 ml/min, a purge flow of 5 ml/min and an open valve time of 3 min. To obtain a pulsed injection, a programmed gas flow was used whereby the helium gas flow was set at 2.7 mL/min for 0.1 min, followed by a decrease in flow of 20 ml/min to the normal 0.9 mL/min. The temperature was first held at 30 °C for 3 min and then allowed to rise to 80 °C at a rate of 7 °C/min, followed by a second ramp of 2 °C/min till 125 °C and a final ramp of 8 °C/min with a final temperature of 270 °C.

Mass acquisition range was 33 to 550 amu at a scan rate of 5 scans/s. Electron impact ionization energy was 70 eV. The interface and ion source were kept at 275 °C and 250 °C, respectively. A mix of linear n-alkanes (from C7 to C40, Supelco Co.) was injected into the GC-MS under identical conditions to serve as external retention index markers. Identification and quantification of the compounds were performed using an in-house developed R script as described in Goelen et al. and Reher et al. 87 , 88 (for package information, see Supplementary Table  S8 ). Briefly, chromatograms were analyzed using AMDIS (v2.71) 89 to separate overlapping peaks and obtain pure compound spectra. The NIST MS Search software (v2.0 g) in combination with the NIST2017, FFNSC3 and Adams4 libraries were used to manually identify the empirical spectra, taking into account the expected retention time. After background subtraction and correcting for retention time shifts between samples run on different days based on alkane ladders, compound elution profiles were extracted and integrated using a file with 284 target compounds of interest, which were either recovered in our identified AMDIS list of spectra or were known to occur in beer. Compound elution profiles were estimated for every peak in every chromatogram over a time-restricted window using weighted non-negative least square analysis after which peak areas were integrated 87 , 88 . Batch effect correction was performed by normalizing against the most stable internal standard compound, 4-fluorobenzaldehyde. Out of all 284 target compounds that were analyzed, 167 were visually judged to have reliable elution profiles and were used for final analysis.

Discrete photometric and enzymatic analysis

Discrete photometric and enzymatic analysis (Thermo Scientific TM Gallery TM Plus Beermaster Discrete Analyzer) was used to measure acetic acid, ammonia, beta-glucan, iso-alpha acids, color, sugars, glycerol, iron, pH, protein, and sulfite. 2 ml of sample volume was used for the analyses. Information regarding the reagents and standard solutions used for analyses and calibrations is included in Supplementary Table  S7 and Supplementary Table  S9 .

NIR analyses

NIR analysis (Anton Paar Alcolyzer Beer ME System) was used to measure ethanol. Measurements comprised 50 ml of sample, and a 10% EtOH solution was used for calibration.

Correlation calculations

Pairwise Spearman Rank correlations were calculated between all chemical properties.

Sensory dataset

Trained panel.

Our trained tasting panel consisted of volunteers who gave prior verbal informed consent. All compounds used for the validation experiment were of food-grade quality. The tasting sessions were approved by the Social and Societal Ethics Committee of the KU Leuven (G-2022-5677-R2(MAR)). All online reviewers agreed to the Terms and Conditions of the RateBeer website.

Sensory analysis was performed according to the American Society of Brewing Chemists (ASBC) Sensory Analysis Methods 90 . 30 volunteers were screened through a series of triangle tests. The sixteen most sensitive and consistent tasters were retained as taste panel members. The resulting panel was diverse in age [22–42, mean: 29], sex [56% male] and nationality [7 different countries]. The panel developed a consensus vocabulary to describe beer aroma, taste and mouthfeel. Panelists were trained to identify and score 50 different attributes, using a 7-point scale to rate attributes’ intensity. The scoring sheet is included as Supplementary Data  3 . Sensory assessments took place between 10–12 a.m. The beers were served in black-colored glasses. Per session, between 5 and 12 beers of the same style were tasted at 12 °C to 16 °C. Two reference beers were added to each set and indicated as ‘Reference 1 & 2’, allowing panel members to calibrate their ratings. Not all panelists were present at every tasting. Scores were scaled by standard deviation and mean-centered per taster. Values are represented as z-scores and clustered by Euclidean distance. Pairwise Spearman correlations were calculated between taste and aroma sensory attributes. Panel consistency was evaluated by repeating samples on different sessions and performing ANOVA to identify differences, using the ‘stats’ package (v4.2.2) in R (for package information, see Supplementary Table  S8 ).

Online reviews from a public database

The ‘scrapy’ package in Python (v3.6) (for package information, see Supplementary Table  S8 ). was used to collect 232,288 online reviews (mean=922, min=6, max=5343) from RateBeer, an online beer review database. Each review entry comprised 5 numerical scores (appearance, aroma, taste, palate and overall quality) and an optional review text. The total number of reviews per reviewer was collected separately. Numerical scores were scaled and centered per rater, and mean scores were calculated per beer.

For the review texts, the language was estimated using the packages ‘langdetect’ and ‘langid’ in Python. Reviews that were classified as English by both packages were kept. Reviewers with fewer than 100 entries overall were discarded. 181,025 reviews from >6000 reviewers from >40 countries remained. Text processing was done using the ‘nltk’ package in Python. Texts were corrected for slang and misspellings; proper nouns and rare words that are relevant to the beer context were specified and kept as-is (‘Chimay’,’Lambic’, etc.). A dictionary of semantically similar sensorial terms, for example ‘floral’ and ‘flower’, was created and collapsed together into one term. Words were stemmed and lemmatized to avoid identifying words such as ‘acid’ and ‘acidity’ as separate terms. Numbers and punctuation were removed.

Sentences from up to 50 randomly chosen reviews per beer were manually categorized according to the aspect of beer they describe (appearance, aroma, taste, palate, overall quality—not to be confused with the 5 numerical scores described above) or flagged as irrelevant if they contained no useful information. If a beer contained fewer than 50 reviews, all reviews were manually classified. This labeled data set was used to train a model that classified the rest of the sentences for all beers 91 . Sentences describing taste and aroma were extracted, and term frequency–inverse document frequency (TFIDF) was implemented to calculate enrichment scores for sensorial words per beer.

The sex of the tasting subject was not considered when building our sensory database. Instead, results from different panelists were averaged, both for our trained panel (56% male, 44% female) and the RateBeer reviews (70% male, 30% female for RateBeer as a whole).

Beer price collection and processing

Beer prices were collected from the following stores: Colruyt, Delhaize, Total Wine, BeerHawk, The Belgian Beer Shop, The Belgian Shop, and Beer of Belgium. Where applicable, prices were converted to Euros and normalized per liter. Spearman correlations were calculated between these prices and mean overall appreciation scores from RateBeer and the taste panel, respectively.

Pairwise Spearman Rank correlations were calculated between all sensory properties.

Machine learning models

Predictive modeling of sensory profiles from chemical data.

Regression models were constructed to predict (a) trained panel scores for beer flavors and quality from beer chemical profiles and (b) public reviews’ appreciation scores from beer chemical profiles. Z-scores were used to represent sensory attributes in both data sets. Chemical properties with log-normal distributions (Shapiro-Wilk test, p  <  0.05 ) were log-transformed. Missing chemical measurements (0.1% of all data) were replaced with mean values per attribute. Observations from 250 beers were randomly separated into a training set (70%, 175 beers) and a test set (30%, 75 beers), stratified per beer style. Chemical measurements (p = 231) were normalized based on the training set average and standard deviation. In total, three linear regression-based models: linear regression with first-order interaction terms (LR), lasso regression with first-order interaction terms (Lasso) and partial least squares regression (PLSR); five decision tree models, Adaboost regressor (ABR), Extra Trees (ET), Gradient Boosting regressor (GBR), Random Forest (RF) and XGBoost regressor (XGBR); one support vector machine model (SVR) and one artificial neural network model (ANN) were trained. The models were implemented using the ‘scikit-learn’ package (v1.2.2) and ‘xgboost’ package (v1.7.3) in Python (v3.9.16). Models were trained, and hyperparameters optimized, using five-fold cross-validated grid search with the coefficient of determination (R 2 ) as the evaluation metric. The ANN (scikit-learn’s MLPRegressor) was optimized using Bayesian Tree-Structured Parzen Estimator optimization with the ‘Optuna’ Python package (v3.2.0). Individual models were trained per attribute, and a multi-output model was trained on all attributes simultaneously.

Model dissection

GBR was found to outperform other methods, resulting in models with the highest average R 2 values in both trained panel and public review data sets. Impurity-based rankings of the most important predictors for each predicted sensorial trait were obtained using the ‘scikit-learn’ package. To observe the relationships between these chemical properties and their predicted targets, partial dependence plots (PDP) were constructed for the six most important predictors of consumer appreciation 74 , 75 .

The ‘SHAP’ package in Python (v0.41.0) was implemented to provide an alternative ranking of predictor importance and to visualize the predictors’ effects as a function of their concentration 68 .

Validation of causal chemical properties

To validate the effects of the most important model features on predicted sensory attributes, beers were spiked with the chemical compounds identified by the models and descriptive sensory analyses were carried out according to the American Society of Brewing Chemists (ASBC) protocol 90 .

Compound spiking was done 30 min before tasting. Compounds were spiked into fresh beer bottles, that were immediately resealed and inverted three times. Fresh bottles of beer were opened for the same duration, resealed, and inverted thrice, to serve as controls. Pairs of spiked samples and controls were served simultaneously, chilled and in dark glasses as outlined in the Trained panel section above. Tasters were instructed to select the glass with the higher flavor intensity for each attribute (directional difference test 92 ) and to select the glass they prefer.

The final concentration after spiking was equal to the within-style average, after normalizing by ethanol concentration. This was done to ensure balanced flavor profiles in the final spiked beer. The same methods were applied to improve a non-alcoholic beer. Compounds were the following: ethyl acetate (Merck KGaA, W241415), ethyl hexanoate (Merck KGaA, W243906), isoamyl acetate (Merck KGaA, W205508), phenethyl acetate (Merck KGaA, W285706), ethanol (96%, Colruyt), glycerol (Merck KGaA, W252506), lactic acid (Merck KGaA, 261106).

Significant differences in preference or perceived intensity were determined by performing the two-sided binomial test on each attribute.

Reporting summary

Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

Data availability

The data that support the findings of this work are available in the Supplementary Data files and have been deposited to Zenodo under accession code 10653704 93 . The RateBeer scores data are under restricted access, they are not publicly available as they are property of RateBeer (ZX Ventures, USA). Access can be obtained from the authors upon reasonable request and with permission of RateBeer (ZX Ventures, USA).  Source data are provided with this paper.

Code availability

The code for training the machine learning models, analyzing the models, and generating the figures has been deposited to Zenodo under accession code 10653704 93 .

Tieman, D. et al. A chemical genetic roadmap to improved tomato flavor. Science 355 , 391–394 (2017).

Article   ADS   CAS   PubMed   Google Scholar  

Plutowska, B. & Wardencki, W. Application of gas chromatography–olfactometry (GC–O) in analysis and quality assessment of alcoholic beverages – A review. Food Chem. 107 , 449–463 (2008).

Article   CAS   Google Scholar  

Legin, A., Rudnitskaya, A., Seleznev, B. & Vlasov, Y. Electronic tongue for quality assessment of ethanol, vodka and eau-de-vie. Anal. Chim. Acta 534 , 129–135 (2005).

Loutfi, A., Coradeschi, S., Mani, G. K., Shankar, P. & Rayappan, J. B. B. Electronic noses for food quality: A review. J. Food Eng. 144 , 103–111 (2015).

Ahn, Y.-Y., Ahnert, S. E., Bagrow, J. P. & Barabási, A.-L. Flavor network and the principles of food pairing. Sci. Rep. 1 , 196 (2011).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Bartoshuk, L. M. & Klee, H. J. Better fruits and vegetables through sensory analysis. Curr. Biol. 23 , R374–R378 (2013).

Article   CAS   PubMed   Google Scholar  

Piggott, J. R. Design questions in sensory and consumer science. Food Qual. Prefer. 3293 , 217–220 (1995).

Article   Google Scholar  

Kermit, M. & Lengard, V. Assessing the performance of a sensory panel-panellist monitoring and tracking. J. Chemom. 19 , 154–161 (2005).

Cook, D. J., Hollowood, T. A., Linforth, R. S. T. & Taylor, A. J. Correlating instrumental measurements of texture and flavour release with human perception. Int. J. Food Sci. Technol. 40 , 631–641 (2005).

Chinchanachokchai, S., Thontirawong, P. & Chinchanachokchai, P. A tale of two recommender systems: The moderating role of consumer expertise on artificial intelligence based product recommendations. J. Retail. Consum. Serv. 61 , 1–12 (2021).

Ross, C. F. Sensory science at the human-machine interface. Trends Food Sci. Technol. 20 , 63–72 (2009).

Chambers, E. IV & Koppel, K. Associations of volatile compounds with sensory aroma and flavor: The complex nature of flavor. Molecules 18 , 4887–4905 (2013).

Pinu, F. R. Metabolomics—The new frontier in food safety and quality research. Food Res. Int. 72 , 80–81 (2015).

Danezis, G. P., Tsagkaris, A. S., Brusic, V. & Georgiou, C. A. Food authentication: state of the art and prospects. Curr. Opin. Food Sci. 10 , 22–31 (2016).

Shepherd, G. M. Smell images and the flavour system in the human brain. Nature 444 , 316–321 (2006).

Meilgaard, M. C. Prediction of flavor differences between beers from their chemical composition. J. Agric. Food Chem. 30 , 1009–1017 (1982).

Xu, L. et al. Widespread receptor-driven modulation in peripheral olfactory coding. Science 368 , eaaz5390 (2020).

Kupferschmidt, K. Following the flavor. Science 340 , 808–809 (2013).

Billesbølle, C. B. et al. Structural basis of odorant recognition by a human odorant receptor. Nature 615 , 742–749 (2023).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Smith, B. Perspective: Complexities of flavour. Nature 486 , S6–S6 (2012).

Pfister, P. et al. Odorant receptor inhibition is fundamental to odor encoding. Curr. Biol. 30 , 2574–2587 (2020).

Moskowitz, H. W., Kumaraiah, V., Sharma, K. N., Jacobs, H. L. & Sharma, S. D. Cross-cultural differences in simple taste preferences. Science 190 , 1217–1218 (1975).

Eriksson, N. et al. A genetic variant near olfactory receptor genes influences cilantro preference. Flavour 1 , 22 (2012).

Ferdenzi, C. et al. Variability of affective responses to odors: Culture, gender, and olfactory knowledge. Chem. Senses 38 , 175–186 (2013).

Article   PubMed   Google Scholar  

Lawless, H. T. & Heymann, H. Sensory evaluation of food: Principles and practices. (Springer, New York, NY). https://doi.org/10.1007/978-1-4419-6488-5 (2010).

Colantonio, V. et al. Metabolomic selection for enhanced fruit flavor. Proc. Natl. Acad. Sci. 119 , e2115865119 (2022).

Fritz, F., Preissner, R. & Banerjee, P. VirtualTaste: a web server for the prediction of organoleptic properties of chemical compounds. Nucleic Acids Res 49 , W679–W684 (2021).

Tuwani, R., Wadhwa, S. & Bagler, G. BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules. Sci. Rep. 9 , 1–13 (2019).

Dagan-Wiener, A. et al. Bitter or not? BitterPredict, a tool for predicting taste from chemical structure. Sci. Rep. 7 , 1–13 (2017).

Pallante, L. et al. Toward a general and interpretable umami taste predictor using a multi-objective machine learning approach. Sci. Rep. 12 , 1–11 (2022).

Malavolta, M. et al. A survey on computational taste predictors. Eur. Food Res. Technol. 248 , 2215–2235 (2022).

Lee, B. K. et al. A principal odor map unifies diverse tasks in olfactory perception. Science 381 , 999–1006 (2023).

Mayhew, E. J. et al. Transport features predict if a molecule is odorous. Proc. Natl. Acad. Sci. 119 , e2116576119 (2022).

Niu, Y. et al. Sensory evaluation of the synergism among ester odorants in light aroma-type liquor by odor threshold, aroma intensity and flash GC electronic nose. Food Res. Int. 113 , 102–114 (2018).

Yu, P., Low, M. Y. & Zhou, W. Design of experiments and regression modelling in food flavour and sensory analysis: A review. Trends Food Sci. Technol. 71 , 202–215 (2018).

Oladokun, O. et al. The impact of hop bitter acid and polyphenol profiles on the perceived bitterness of beer. Food Chem. 205 , 212–220 (2016).

Linforth, R., Cabannes, M., Hewson, L., Yang, N. & Taylor, A. Effect of fat content on flavor delivery during consumption: An in vivo model. J. Agric. Food Chem. 58 , 6905–6911 (2010).

Guo, S., Na Jom, K. & Ge, Y. Influence of roasting condition on flavor profile of sunflower seeds: A flavoromics approach. Sci. Rep. 9 , 11295 (2019).

Ren, Q. et al. The changes of microbial community and flavor compound in the fermentation process of Chinese rice wine using Fagopyrum tataricum grain as feedstock. Sci. Rep. 9 , 3365 (2019).

Hastie, T., Friedman, J. & Tibshirani, R. The Elements of Statistical Learning. (Springer, New York, NY). https://doi.org/10.1007/978-0-387-21606-5 (2001).

Dietz, C., Cook, D., Huismann, M., Wilson, C. & Ford, R. The multisensory perception of hop essential oil: a review. J. Inst. Brew. 126 , 320–342 (2020).

CAS   Google Scholar  

Roncoroni, Miguel & Verstrepen, Kevin Joan. Belgian Beer: Tested and Tasted. (Lannoo, 2018).

Meilgaard, M. Flavor chemistry of beer: Part II: Flavor and threshold of 239 aroma volatiles. in (1975).

Bokulich, N. A. & Bamforth, C. W. The microbiology of malting and brewing. Microbiol. Mol. Biol. Rev. MMBR 77 , 157–172 (2013).

Dzialo, M. C., Park, R., Steensels, J., Lievens, B. & Verstrepen, K. J. Physiology, ecology and industrial applications of aroma formation in yeast. FEMS Microbiol. Rev. 41 , S95–S128 (2017).

Article   PubMed   PubMed Central   Google Scholar  

Datta, A. et al. Computer-aided food engineering. Nat. Food 3 , 894–904 (2022).

American Society of Brewing Chemists. Beer Methods. (American Society of Brewing Chemists, St. Paul, MN, U.S.A.).

Olaniran, A. O., Hiralal, L., Mokoena, M. P. & Pillay, B. Flavour-active volatile compounds in beer: production, regulation and control. J. Inst. Brew. 123 , 13–23 (2017).

Verstrepen, K. J. et al. Flavor-active esters: Adding fruitiness to beer. J. Biosci. Bioeng. 96 , 110–118 (2003).

Meilgaard, M. C. Flavour chemistry of beer. part I: flavour interaction between principal volatiles. Master Brew. Assoc. Am. Tech. Q 12 , 107–117 (1975).

Briggs, D. E., Boulton, C. A., Brookes, P. A. & Stevens, R. Brewing 227–254. (Woodhead Publishing). https://doi.org/10.1533/9781855739062.227 (2004).

Bossaert, S., Crauwels, S., De Rouck, G. & Lievens, B. The power of sour - A review: Old traditions, new opportunities. BrewingScience 72 , 78–88 (2019).

Google Scholar  

Verstrepen, K. J. et al. Flavor active esters: Adding fruitiness to beer. J. Biosci. Bioeng. 96 , 110–118 (2003).

Snauwaert, I. et al. Microbial diversity and metabolite composition of Belgian red-brown acidic ales. Int. J. Food Microbiol. 221 , 1–11 (2016).

Spitaels, F. et al. The microbial diversity of traditional spontaneously fermented lambic beer. PLoS ONE 9 , e95384 (2014).

Blanco, C. A., Andrés-Iglesias, C. & Montero, O. Low-alcohol Beers: Flavor Compounds, Defects, and Improvement Strategies. Crit. Rev. Food Sci. Nutr. 56 , 1379–1388 (2016).

Jackowski, M. & Trusek, A. Non-Alcohol. beer Prod. – Overv. 20 , 32–38 (2018).

Takoi, K. et al. The contribution of geraniol metabolism to the citrus flavour of beer: Synergy of geraniol and β-citronellol under coexistence with excess linalool. J. Inst. Brew. 116 , 251–260 (2010).

Kroeze, J. H. & Bartoshuk, L. M. Bitterness suppression as revealed by split-tongue taste stimulation in humans. Physiol. Behav. 35 , 779–783 (1985).

Mennella, J. A. et al. A spoonful of sugar helps the medicine go down”: Bitter masking bysucrose among children and adults. Chem. Senses 40 , 17–25 (2015).

Wietstock, P., Kunz, T., Perreira, F. & Methner, F.-J. Metal chelation behavior of hop acids in buffered model systems. BrewingScience 69 , 56–63 (2016).

Sancho, D., Blanco, C. A., Caballero, I. & Pascual, A. Free iron in pale, dark and alcohol-free commercial lager beers. J. Sci. Food Agric. 91 , 1142–1147 (2011).

Rodrigues, H. & Parr, W. V. Contribution of cross-cultural studies to understanding wine appreciation: A review. Food Res. Int. 115 , 251–258 (2019).

Korneva, E. & Blockeel, H. Towards better evaluation of multi-target regression models. in ECML PKDD 2020 Workshops (eds. Koprinska, I. et al.) 353–362 (Springer International Publishing, Cham, 2020). https://doi.org/10.1007/978-3-030-65965-3_23 .

Gastón Ares. Mathematical and Statistical Methods in Food Science and Technology. (Wiley, 2013).

Grinsztajn, L., Oyallon, E. & Varoquaux, G. Why do tree-based models still outperform deep learning on tabular data? Preprint at http://arxiv.org/abs/2207.08815 (2022).

Gries, S. T. Statistics for Linguistics with R: A Practical Introduction. in Statistics for Linguistics with R (De Gruyter Mouton, 2021). https://doi.org/10.1515/9783110718256 .

Lundberg, S. M. et al. From local explanations to global understanding with explainable AI for trees. Nat. Mach. Intell. 2 , 56–67 (2020).

Ickes, C. M. & Cadwallader, K. R. Effects of ethanol on flavor perception in alcoholic beverages. Chemosens. Percept. 10 , 119–134 (2017).

Kato, M. et al. Influence of high molecular weight polypeptides on the mouthfeel of commercial beer. J. Inst. Brew. 127 , 27–40 (2021).

Wauters, R. et al. Novel Saccharomyces cerevisiae variants slow down the accumulation of staling aldehydes and improve beer shelf-life. Food Chem. 398 , 1–11 (2023).

Li, H., Jia, S. & Zhang, W. Rapid determination of low-level sulfur compounds in beer by headspace gas chromatography with a pulsed flame photometric detector. J. Am. Soc. Brew. Chem. 66 , 188–191 (2008).

Dercksen, A., Laurens, J., Torline, P., Axcell, B. C. & Rohwer, E. Quantitative analysis of volatile sulfur compounds in beer using a membrane extraction interface. J. Am. Soc. Brew. Chem. 54 , 228–233 (1996).

Molnar, C. Interpretable Machine Learning: A Guide for Making Black-Box Models Interpretable. (2020).

Zhao, Q. & Hastie, T. Causal interpretations of black-box models. J. Bus. Econ. Stat. Publ. Am. Stat. Assoc. 39 , 272–281 (2019).

Article   MathSciNet   Google Scholar  

Hastie, T., Tibshirani, R. & Friedman, J. The Elements of Statistical Learning. (Springer, 2019).

Labrado, D. et al. Identification by NMR of key compounds present in beer distillates and residual phases after dealcoholization by vacuum distillation. J. Sci. Food Agric. 100 , 3971–3978 (2020).

Lusk, L. T., Kay, S. B., Porubcan, A. & Ryder, D. S. Key olfactory cues for beer oxidation. J. Am. Soc. Brew. Chem. 70 , 257–261 (2012).

Gonzalez Viejo, C., Torrico, D. D., Dunshea, F. R. & Fuentes, S. Development of artificial neural network models to assess beer acceptability based on sensory properties using a robotic pourer: A comparative model approach to achieve an artificial intelligence system. Beverages 5 , 33 (2019).

Gonzalez Viejo, C., Fuentes, S., Torrico, D. D., Godbole, A. & Dunshea, F. R. Chemical characterization of aromas in beer and their effect on consumers liking. Food Chem. 293 , 479–485 (2019).

Gilbert, J. L. et al. Identifying breeding priorities for blueberry flavor using biochemical, sensory, and genotype by environment analyses. PLOS ONE 10 , 1–21 (2015).

Goulet, C. et al. Role of an esterase in flavor volatile variation within the tomato clade. Proc. Natl. Acad. Sci. 109 , 19009–19014 (2012).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Borisov, V. et al. Deep Neural Networks and Tabular Data: A Survey. IEEE Trans. Neural Netw. Learn. Syst. 1–21 https://doi.org/10.1109/TNNLS.2022.3229161 (2022).

Statista. Statista Consumer Market Outlook: Beer - Worldwide.

Seitz, H. K. & Stickel, F. Molecular mechanisms of alcoholmediated carcinogenesis. Nat. Rev. Cancer 7 , 599–612 (2007).

Voordeckers, K. et al. Ethanol exposure increases mutation rate through error-prone polymerases. Nat. Commun. 11 , 3664 (2020).

Goelen, T. et al. Bacterial phylogeny predicts volatile organic compound composition and olfactory response of an aphid parasitoid. Oikos 129 , 1415–1428 (2020).

Article   ADS   Google Scholar  

Reher, T. et al. Evaluation of hop (Humulus lupulus) as a repellent for the management of Drosophila suzukii. Crop Prot. 124 , 104839 (2019).

Stein, S. E. An integrated method for spectrum extraction and compound identification from gas chromatography/mass spectrometry data. J. Am. Soc. Mass Spectrom. 10 , 770–781 (1999).

American Society of Brewing Chemists. Sensory Analysis Methods. (American Society of Brewing Chemists, St. Paul, MN, U.S.A., 1992).

McAuley, J., Leskovec, J. & Jurafsky, D. Learning Attitudes and Attributes from Multi-Aspect Reviews. Preprint at https://doi.org/10.48550/arXiv.1210.3926 (2012).

Meilgaard, M. C., Carr, B. T. & Carr, B. T. Sensory Evaluation Techniques. (CRC Press, Boca Raton). https://doi.org/10.1201/b16452 (2014).

Schreurs, M. et al. Data from: Predicting and improving complex beer flavor through machine learning. Zenodo https://doi.org/10.5281/zenodo.10653704 (2024).

Download references

Acknowledgements

We thank all lab members for their discussions and thank all tasting panel members for their contributions. Special thanks go out to Dr. Karin Voordeckers for her tremendous help in proofreading and improving the manuscript. M.S. was supported by a Baillet-Latour fellowship, L.C. acknowledges financial support from KU Leuven (C16/17/006), F.A.T. was supported by a PhD fellowship from FWO (1S08821N). Research in the lab of K.J.V. is supported by KU Leuven, FWO, VIB, VLAIO and the Brewing Science Serves Health Fund. Research in the lab of T.W. is supported by FWO (G.0A51.15) and KU Leuven (C16/17/006).

Author information

These authors contributed equally: Michiel Schreurs, Supinya Piampongsant, Miguel Roncoroni.

Authors and Affiliations

VIB—KU Leuven Center for Microbiology, Gaston Geenslaan 1, B-3001, Leuven, Belgium

Michiel Schreurs, Supinya Piampongsant, Miguel Roncoroni, Lloyd Cool, Beatriz Herrera-Malaver, Florian A. Theßeling & Kevin J. Verstrepen

CMPG Laboratory of Genetics and Genomics, KU Leuven, Gaston Geenslaan 1, B-3001, Leuven, Belgium

Leuven Institute for Beer Research (LIBR), Gaston Geenslaan 1, B-3001, Leuven, Belgium

Laboratory of Socioecology and Social Evolution, KU Leuven, Naamsestraat 59, B-3000, Leuven, Belgium

Lloyd Cool, Christophe Vanderaa & Tom Wenseleers

VIB Bioinformatics Core, VIB, Rijvisschestraat 120, B-9052, Ghent, Belgium

Łukasz Kreft & Alexander Botzki

AB InBev SA/NV, Brouwerijplein 1, B-3000, Leuven, Belgium

Philippe Malcorps & Luk Daenen

You can also search for this author in PubMed   Google Scholar

Contributions

S.P., M.S. and K.J.V. conceived the experiments. S.P., M.S. and K.J.V. designed the experiments. S.P., M.S., M.R., B.H. and F.A.T. performed the experiments. S.P., M.S., L.C., C.V., L.K., A.B., P.M., L.D., T.W. and K.J.V. contributed analysis ideas. S.P., M.S., L.C., C.V., T.W. and K.J.V. analyzed the data. All authors contributed to writing the manuscript.

Corresponding author

Correspondence to Kevin J. Verstrepen .

Ethics declarations

Competing interests.

K.J.V. is affiliated with bar.on. The other authors declare no competing interests.

Peer review

Peer review information.

Nature Communications thanks Florian Bauer, Andrew John Macintosh and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information, peer review file, description of additional supplementary files, supplementary data 1, supplementary data 2, supplementary data 3, supplementary data 4, supplementary data 5, supplementary data 6, supplementary data 7, reporting summary, source data, source data, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Schreurs, M., Piampongsant, S., Roncoroni, M. et al. Predicting and improving complex beer flavor through machine learning. Nat Commun 15 , 2368 (2024). https://doi.org/10.1038/s41467-024-46346-0

Download citation

Received : 30 October 2023

Accepted : 21 February 2024

Published : 26 March 2024

DOI : https://doi.org/10.1038/s41467-024-46346-0

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.

• SUGGESTED TOPICS
• The Magazine
• Newsletters
• Managing Yourself
• Managing Teams
• Work-life Balance
• The Big Idea
• Data & Visuals
• Reading Lists
• Case Selections
• HBR Learning
• Topic Feeds
• Account Settings
• Email Preferences

Research: How Different Fields Are Using GenAI to Redefine Roles

• Maryam Alavi

Examples from customer support, management consulting, professional writing, legal analysis, and software and technology.

The interactive, conversational, analytical, and generative features of GenAI offer support for creativity, problem-solving, and processing and digestion of large bodies of information. Therefore, these features can act as cognitive resources for knowledge workers. Moreover, the capabilities of GenAI can mitigate various hindrances to effective performance that knowledge workers may encounter in their jobs, including time pressure, gaps in knowledge and skills, and negative feelings (such as boredom stemming from repetitive tasks or frustration arising from interactions with dissatisfied customers). Empirical research and field observations have already begun to reveal the value of GenAI capabilities and their potential for job crafting.

There is an expectation that implementing new and emerging Generative AI (GenAI) tools enhances the effectiveness and competitiveness of organizations. This belief is evidenced by current and planned investments in GenAI tools, especially by firms in knowledge-intensive industries such as finance, healthcare, and entertainment, among others. According to forecasts, enterprise spending on GenAI will increase by two-fold in 2024 and grow to $151.1 billion by 2027 .

• Maryam Alavi is the Elizabeth D. & Thomas M. Holder Chair & Professor of IT Management, Scheller College of Business, Georgia Institute of Technology .

Partner Center

• Search Menu
• Advance articles
• Author Guidelines
• Submission Site
• Open Access
• About International Mathematics Research Notices
• Editorial Board
• Advertising and Corporate Services
• Journals Career Network
• Self-Archiving Policy
• Dispatch Dates
• Journals on Oxford Academic
• Books on Oxford Academic

Article Contents

• 1 Introduction
• 2 Background
• 3 Finite F-Representation Type
• 4 Frobenius Descent
• 5 A Formalism of F-Modules
• 6 Applications to Local Cohomology
• Acknowledgements

A Formalism of F -modules for Rings with Complete Local Finite F -Representation Type

• Article contents
• Figures & tables
• Supplementary Data

Eamon Quinlan-Gallego, A Formalism of F -modules for Rings with Complete Local Finite F -Representation Type, International Mathematics Research Notices , 2024;, rnae054, https://doi.org/10.1093/imrn/rnae054

• Permissions Icon Permissions

We develop a formalism of unit |$F$| -modules in the style of Lyubeznik and Emerton-Kisin for rings that have finite |$F$| -representation type after localization and completion at every prime ideal. As applications, we show that if |$R$| is such a ring then the iterated local cohomology modules |$H^{n_{1}}_{I_{1}} \circ \cdots \circ H^{n_{s}}_{I_{s}}(R)$| have finitely many associated primes, and that all local cohomology modules |$H^{n}_{I}(R / gR)$| have closed support when |$g$| is a nonzerodivisor on |$R$|⁠ .

Email alerts

Citing articles via.

• Recommend to your Library

Affiliations

• Online ISSN 1687-0247
• Print ISSN 1073-7928
• Copyright © 2024 Oxford University Press
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Institutional account management
• Rights and permissions
• Get help with access
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

• See us on facebook
• See us on twitter
• See us on youtube
• See us on linkedin
• See us on instagram

Pilot study shows ketogenic diet improves severe mental illness

A small clinical trial led by Stanford Medicine found that the metabolic effects of a ketogenic diet may help stabilize the brain.

April 1, 2024 - By Nina Bai

A study led by researchers at Stanford Medicine showed that diet can help those with serious mental illness. nishihata

For people living with serious mental illness like schizophrenia or bipolar disorder, standard treatment with antipsychotic medications can be a double-edged sword. While these drugs help regulate brain chemistry, they often cause metabolic side effects such as insulin resistance and obesity, which are distressing enough that many patients stop taking the medications.

Now, a pilot study led by Stanford Medicine researchers has found that a ketogenic diet not only restores metabolic health in these patients as they continue their medications, but it further improves their psychiatric conditions. The results, published March 27 in Psychiatry Research , suggest that a dietary intervention can be a powerful aid in treating mental illness.

“It’s very promising and very encouraging that you can take back control of your illness in some way, aside from the usual standard of care,” said Shebani Sethi , MD, associate professor of psychiatry and behavioral sciences and the first author of the new paper.

Making the connection

Sethi, who is board certified in obesity and psychiatry, remembers when she first noticed the connection. As a medical student working in an obesity clinic, she saw a patient with treatment-resistant schizophrenia whose auditory hallucinations quieted on a ketogenic diet.

That prompted her to dig into the medical literature. There were only a few, decades-old case reports on using the ketogenic diet to treat schizophrenia, but there was a long track record of success in using ketogenic diets to treat epileptic seizures.

“The ketogenic diet has been proven to be effective for treatment-resistant epileptic seizures by reducing the excitability of neurons in the brain,” Sethi said. “We thought it would be worth exploring this treatment in psychiatric conditions.”

A few years later, Sethi coined the term metabolic psychiatry, a new field that approaches mental health from an energy conversion perspective.

Shebani Sethi

In the four-month pilot trial, Sethi’s team followed 21 adult participants who were diagnosed with schizophrenia or bipolar disorder, taking antipsychotic medications, and had a metabolic abnormality — such as weight gain, insulin resistance, hypertriglyceridemia, dyslipidemia or impaired glucose tolerance. The participants were instructed to follow a ketogenic diet, with approximately 10% of the calories from carbohydrates, 30% from protein and 60% from fat. They were not told to count calories.

“The focus of eating is on whole non-processed foods including protein and non-starchy vegetables, and not restricting fats,” said Sethi, who shared keto-friendly meal ideas with the participants. They were also given keto cookbooks and access to a health coach. 

The research team tracked how well the participants followed the diet through weekly measures of blood ketone levels. (Ketones are acids produced when the body breaks down fat — instead of glucose — for energy.) By the end of the trial, 14 patients had been fully adherent, six were semi-adherent and only one was non-adherent.

The participants underwent a variety of psychiatric and metabolic assessments throughout the trial.

Before the trial, 29% of the participants met the criteria for metabolic syndrome, defined as having at least three of five conditions: abdominal obesity, elevated triglycerides, low HDL cholesterol, elevated blood pressure and elevated fasting glucose levels. After four months on a ketogenic diet, none of the participants had metabolic syndrome.

On average, the participants lost 10% of their body weight; reduced their waist circumference by 11% percent; and had lower blood pressure, body mass index, triglycerides, blood sugar levels and insulin resistance.

“We’re seeing huge changes,” Sethi said. “Even if you’re on antipsychotic drugs, we can still reverse the obesity, the metabolic syndrome, the insulin resistance. I think that’s very encouraging for patients.”

The participants reported improvements in their energy, sleep, mood and quality of life.

The psychiatric benefits were also striking. On average, the participants improved 31% on a psychiatrist rating of mental illness known as the clinical global impressions scale, with three-quarters of the group showing clinically meaningful improvement. Overall, the participants also reported better sleep and greater life satisfaction.

“The participants reported improvements in their energy, sleep, mood and quality of life,” Sethi said. “They feel healthier and more hopeful.”

The researchers were impressed that most of the participants stuck with the diet. “We saw more benefit with the adherent group compared with the semi-adherent group, indicating a potential dose-response relationship,” Sethi said.

Alternative fuel for the brain

There is increasing evidence that psychiatric diseases such as schizophrenia and bipolar disorder stem from metabolic deficits in the brain, which affect the excitability of neurons, Sethi said.

The researchers hypothesize that just as a ketogenic diet improves the rest of the body’s metabolism, it also improves the brain’s metabolism.

“Anything that improves metabolic health in general is probably going to improve brain health anyway,” Sethi said. “But the ketogenic diet can provide ketones as an alternative fuel to glucose for a brain with energy dysfunction.”

Likely there are multiple mechanisms at work, she added, and the main purpose of the small pilot trial is to help researchers detect signals that will guide the design of larger, more robust studies.  

As a physician, Sethi cares for many patients with both serious mental illness and obesity or metabolic syndrome, but few studies have focused on this undertreated population.

She is the founder and director of the metabolic psychiatry clinic at Stanford Medicine.

“Many of my patients suffer from both illnesses, so my desire was to see if metabolic interventions could help them,” she said. “They are seeking more help. They are looking to just feel better.”

Researchers from the University of Michigan; the University of California, San Francisco; and Duke University contributed to the study.

The study was supported by Baszucki Group Research Fund, Keun Lau Fund and the Obesity Treatment Foundation.

About Stanford Medicine

Stanford Medicine is an integrated academic health system comprising the Stanford School of Medicine and adult and pediatric health care delivery systems. Together, they harness the full potential of biomedicine through collaborative research, education and clinical care for patients. For more information, please visit med.stanford.edu .

Artificial intelligence

Exploring ways AI is applied to health care

• Download PDF
• Share X Facebook Email LinkedIn
• Permissions

Fatal Traffic Risks With a Total Solar Eclipse in the US

• 1 Department of Medicine, University of Toronto, Toronto, Ontario, Canada
• 2 Evaluative Clinical Science Platform, Sunnybrook Research Institute, Toronto, Ontario, Canada
• 3 Institute for Clinical Evaluative Sciences, Toronto, Ontario, Canada
• 4 Division of General Internal Medicine, Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada
• 5 Center for Leading Injury Prevention Practice Education & Research, Toronto, Ontario, Canada
• 6 Department of Medicine, University of British Columbia, Vancouver, British Columbia, Canada
• 7 Centre for Clinical Epidemiology & Evaluation, University of British Columbia, Vancouver, British Columbia, Canada

A total solar eclipse occurs when the moon temporarily obscures the sun and casts a dark shadow across the earth. This astronomical spectacle has been described for more than 3 millennia and can be predicted with high precision. Eclipse-related solar retinopathy (vision loss from staring at the sun) is an established medical complication; however, other medical outcomes have received little attention. 1

Read More About

Redelmeier DA , Staples JA. Fatal Traffic Risks With a Total Solar Eclipse in the US. JAMA Intern Med. Published online March 25, 2024. doi:10.1001/jamainternmed.2023.5234

Manage citations:

© 2024

Artificial Intelligence Resource Center

Best of JAMA Network 2022

Browse and subscribe to JAMA Network podcasts!

Others Also Liked

Select your interests.

Customize your JAMA Network experience by selecting one or more topics from the list below.

• Academic Medicine
• Acid Base, Electrolytes, Fluids
• Allergy and Clinical Immunology
• American Indian or Alaska Natives
• Anesthesiology
• Anticoagulation
• Art and Images in Psychiatry
• Artificial Intelligence
• Assisted Reproduction
• Bleeding and Transfusion
• Caring for the Critically Ill Patient
• Challenges in Clinical Electrocardiography
• Climate and Health
• Climate Change
• Clinical Challenge
• Clinical Decision Support
• Clinical Implications of Basic Neuroscience
• Clinical Pharmacy and Pharmacology
• Complementary and Alternative Medicine
• Consensus Statements
• Coronavirus (COVID-19)
• Critical Care Medicine
• Cultural Competency
• Dental Medicine
• Dermatology
• Diabetes and Endocrinology
• Diagnostic Test Interpretation
• Drug Development
• Electronic Health Records
• Emergency Medicine
• End of Life, Hospice, Palliative Care
• Environmental Health
• Equity, Diversity, and Inclusion
• Facial Plastic Surgery
• Gastroenterology and Hepatology
• Genetics and Genomics
• Genomics and Precision Health
• Global Health
• Guide to Statistics and Methods
• Hair Disorders
• Health Care Delivery Models
• Health Care Economics, Insurance, Payment
• Health Care Quality
• Health Care Reform
• Health Care Safety
• Health Care Workforce
• Health Disparities
• Health Inequities
• Health Policy
• Health Systems Science
• History of Medicine
• Hypertension
• Images in Neurology
• Implementation Science
• Infectious Diseases
• Innovations in Health Care Delivery
• JAMA Infographic
• Law and Medicine
• Leading Change
• Less is More
• LGBTQIA Medicine
• Lifestyle Behaviors
• Medical Coding
• Medical Devices and Equipment
• Medical Education
• Medical Education and Training
• Medical Journals and Publishing
• Mobile Health and Telemedicine
• Narrative Medicine
• Neuroscience and Psychiatry
• Notable Notes
• Nutrition, Obesity, Exercise
• Obstetrics and Gynecology
• Occupational Health
• Ophthalmology
• Orthopedics
• Otolaryngology
• Pain Medicine
• Palliative Care
• Pathology and Laboratory Medicine
• Patient Care
• Patient Information
• Performance Improvement
• Performance Measures
• Perioperative Care and Consultation
• Pharmacoeconomics
• Pharmacoepidemiology
• Pharmacogenetics
• Pharmacy and Clinical Pharmacology
• Physical Medicine and Rehabilitation
• Physical Therapy
• Physician Leadership
• Population Health
• Primary Care
• Professional Well-being
• Professionalism
• Psychiatry and Behavioral Health
• Public Health
• Pulmonary Medicine
• Regulatory Agencies
• Reproductive Health
• Research, Methods, Statistics
• Resuscitation
• Rheumatology
• Risk Management
• Scientific Discovery and the Future of Medicine
• Shared Decision Making and Communication
• Sleep Medicine
• Sports Medicine
• Stem Cell Transplantation
• Substance Use and Addiction Medicine
• Surgical Innovation
• Surgical Pearls
• Teachable Moment
• Technology and Finance
• The Art of JAMA
• The Arts and Medicine
• The Rational Clinical Examination
• Tobacco and e-Cigarettes
• Translational Medicine
• Trauma and Injury
• Treatment Adherence
• Ultrasonography
• Users' Guide to the Medical Literature
• Vaccination
• Venous Thromboembolism
• Veterans Health
• Women's Health
• Workflow and Process
• Wound Care, Infection, Healing
• Register for email alerts with links to free full-text articles
• Access PDFs of free articles
• Manage your interests
• Save searches and receive search alerts

Suggestions or feedback?

MIT News | Massachusetts Institute of Technology

• Machine learning
• Social justice
• Black holes
• Classes and programs

Departments

• Aeronautics and Astronautics
• Brain and Cognitive Sciences
• Architecture
• Political Science
• Mechanical Engineering

Centers, Labs, & Programs

• Abdul Latif Jameel Poverty Action Lab (J-PAL)
• Picower Institute for Learning and Memory
• Lincoln Laboratory
• School of Architecture + Planning
• School of Engineering
• School of Humanities, Arts, and Social Sciences
• Sloan School of Management
• School of Science
• MIT Schwarzman College of Computing

Large language models use a surprisingly simple mechanism to retrieve some stored knowledge

Press contact :.

Previous image Next image

Large language models, such as those that power popular artificial intelligence chatbots like ChatGPT, are incredibly complex. Even though these models are being used as tools in many areas, such as customer support, code generation, and language translation, scientists still don’t fully grasp how they work.

In an effort to better understand what is going on under the hood, researchers at MIT and elsewhere studied the mechanisms at work when these enormous machine-learning models retrieve stored knowledge.

They found a surprising result: Large language models (LLMs) often use a very simple linear function to recover and decode stored facts. Moreover, the model uses the same decoding function for similar types of facts. Linear functions, equations with only two variables and no exponents, capture the straightforward, straight-line relationship between two variables.

The researchers showed that, by identifying linear functions for different facts, they can probe the model to see what it knows about new subjects, and where within the model that knowledge is stored.

Using a technique they developed to estimate these simple functions, the researchers found that even when a model answers a prompt incorrectly, it has often stored the correct information. In the future, scientists could use such an approach to find and correct falsehoods inside the model, which could reduce a model’s tendency to sometimes give incorrect or nonsensical answers.

“Even though these models are really complicated, nonlinear functions that are trained on lots of data and are very hard to understand, there are sometimes really simple mechanisms working inside them. This is one instance of that,” says Evan Hernandez, an electrical engineering and computer science (EECS) graduate student and co-lead author of a paper detailing these findings .

Hernandez wrote the paper with co-lead author Arnab Sharma, a computer science graduate student at Northeastern University; his advisor, Jacob Andreas, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL); senior author David Bau, an assistant professor of computer science at Northeastern; and others at MIT, Harvard University, and the Israeli Institute of Technology. The research will be presented at the International Conference on Learning Representations.

Finding facts

Most large language models, also called transformer models, are neural networks . Loosely based on the human brain, neural networks contain billions of interconnected nodes, or neurons, that are grouped into many layers, and which encode and process data.

Much of the knowledge stored in a transformer can be represented as relations that connect subjects and objects. For instance, “Miles Davis plays the trumpet” is a relation that connects the subject, Miles Davis, to the object, trumpet.

As a transformer gains more knowledge, it stores additional facts about a certain subject across multiple layers. If a user asks about that subject, the model must decode the most relevant fact to respond to the query.

If someone prompts a transformer by saying “Miles Davis plays the. . .” the model should respond with “trumpet” and not “Illinois” (the state where Miles Davis was born).

“Somewhere in the network’s computation, there has to be a mechanism that goes and looks for the fact that Miles Davis plays the trumpet, and then pulls that information out and helps generate the next word. We wanted to understand what that mechanism was,” Hernandez says.

The researchers set up a series of experiments to probe LLMs, and found that, even though they are extremely complex, the models decode relational information using a simple linear function. Each function is specific to the type of fact being retrieved.

For example, the transformer would use one decoding function any time it wants to output the instrument a person plays and a different function each time it wants to output the state where a person was born.

The researchers developed a method to estimate these simple functions, and then computed functions for 47 different relations, such as “capital city of a country” and “lead singer of a band.”

While there could be an infinite number of possible relations, the researchers chose to study this specific subset because they are representative of the kinds of facts that can be written in this way.

They tested each function by changing the subject to see if it could recover the correct object information. For instance, the function for “capital city of a country” should retrieve Oslo if the subject is Norway and London if the subject is England.

Functions retrieved the correct information more than 60 percent of the time, showing that some information in a transformer is encoded and retrieved in this way.

“But not everything is linearly encoded. For some facts, even though the model knows them and will predict text that is consistent with these facts, we can’t find linear functions for them. This suggests that the model is doing something more intricate to store that information,” he says.

Visualizing a model’s knowledge

They also used the functions to determine what a model believes is true about different subjects.

In one experiment, they started with the prompt “Bill Bradley was a” and used the decoding functions for “plays sports” and “attended university” to see if the model knows that Sen. Bradley was a basketball player who attended Princeton.

“We can show that, even though the model may choose to focus on different information when it produces text, it does encode all that information,” Hernandez says.

They used this probing technique to produce what they call an “attribute lens,” a grid that visualizes where specific information about a particular relation is stored within the transformer’s many layers.

Attribute lenses can be generated automatically, providing a streamlined method to help researchers understand more about a model. This visualization tool could enable scientists and engineers to correct stored knowledge and help prevent an AI chatbot from giving false information.

In the future, Hernandez and his collaborators want to better understand what happens in cases where facts are not stored linearly. They would also like to run experiments with larger models, as well as study the precision of linear decoding functions.

“This is an exciting work that reveals a missing piece in our understanding of how large language models recall factual knowledge during inference. Previous work showed that LLMs build information-rich representations of given subjects, from which specific attributes are being extracted during inference. This work shows that the complex nonlinear computation of LLMs for attribute extraction can be well-approximated with a simple linear function,” says Mor Geva Pipek, an assistant professor in the School of Computer Science at Tel Aviv University, who was not involved with this work.

This research was supported, in part, by Open Philanthropy, the Israeli Science Foundation, and an Azrieli Foundation Early Career Faculty Fellowship.

Share this news article on:

Press mentions.

Researchers at MIT have found that large language models mimic intelligence using linear functions, reports Kyle Wiggers for  TechCrunch . “Even though these models are really complicated, nonlinear functions that are trained on lots of data and are very hard to understand, there are sometimes really simple mechanisms working inside them,” writes Wiggers. 

Previous item Next item

Related Links

• Evan Hernandez
• Jacob Andreas
• Language and Intelligence Group
• Computer Science and Artificial Intelligence Laboratory
• Department of Electrical Engineering and Computer Science

Related Topics

• Computer science and technology
• Artificial intelligence
• Human-computer interaction
• Computer Science and Artificial Intelligence Laboratory (CSAIL)
• Electrical Engineering & Computer Science (eecs)

Related Articles

Demystifying machine-learning systems

AI agents help explain other AI systems

3 Questions: Jacob Andreas on large language models

Solving a machine-learning mystery

More mit news.

Designing solutions to ensure equity in health care

Read full story →

Training manufacturing technologists to be future shop floor leaders

Characterizing social networks

MIT economics to launch new predoctoral fellowship program

Programming functional fabrics

Most work is new work, long-term study of U.S. census data shows

• More news on MIT News homepage →

Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA, USA

• Map (opens in new window)
• Events (opens in new window)
• People (opens in new window)
• Careers (opens in new window)
• Accessibility
• Social Media Hub
• MIT on Facebook
• MIT on YouTube
• MIT on Instagram

Advertisement

Supported by

Use of Abortion Pills Has Risen Significantly Post Roe, Research Shows

By Pam Belluck

Pam Belluck has been reporting about reproductive health for over a decade.

• Share full article

On the eve of oral arguments in a Supreme Court case that could affect future access to abortion pills, new research shows the fast-growing use of medication abortion nationally and the many ways women have obtained access to the method since Roe v. Wade was overturned in June 2022.

The Details

A study, published on Monday in the medical journal JAMA , found that the number of abortions using pills obtained outside the formal health system soared in the six months after the national right to abortion was overturned. Another report, published last week by the Guttmacher Institute , a research organization that supports abortion rights, found that medication abortions now account for nearly two-thirds of all abortions provided by the country’s formal health system, which includes clinics and telemedicine abortion services.

The JAMA study evaluated data from overseas telemedicine organizations, online vendors and networks of community volunteers that generally obtain pills from outside the United States. Before Roe was overturned, these avenues provided abortion pills to about 1,400 women per month, but in the six months afterward, the average jumped to 5,900 per month, the study reported.

Overall, the study found that while abortions in the formal health care system declined by about 32,000 from July through December 2022, much of that decline was offset by about 26,000 medication abortions from pills provided by sources outside the formal health system.

“We see what we see elsewhere in the world in the U.S. — that when anti-abortion laws go into effect, oftentimes outside of the formal health care setting is where people look, and the locus of care gets shifted,” said Dr. Abigail Aiken, who is an associate professor at the University of Texas at Austin and the lead author of the JAMA study.

The co-authors were a statistics professor at the university; the founder of Aid Access, a Europe-based organization that helped pioneer telemedicine abortion in the United States; and a leader of Plan C, an organization that provides consumers with information about medication abortion. Before publication, the study went through the rigorous peer review process required by a major medical journal.

The telemedicine organizations in the study evaluated prospective patients using written medical questionnaires, issued prescriptions from doctors who were typically in Europe and had pills shipped from pharmacies in India, generally charging about $100. Community networks typically asked for some information about the pregnancy and either delivered or mailed pills with detailed instructions, often for free.

Online vendors, which supplied a small percentage of the pills in the study and charged between $39 and $470, generally did not ask for women’s medical history and shipped the pills with the least detailed instructions. Vendors in the study were vetted by Plan C and found to be providing genuine abortion pills, Dr. Aiken said.

The Guttmacher report, focusing on the formal health care system, included data from clinics and telemedicine abortion services within the United States that provided abortion to patients who lived in or traveled to states with legal abortion between January and December 2023.

It found that pills accounted for 63 percent of those abortions, up from 53 percent in 2020. The total number of abortions in the report was over a million for the first time in more than a decade.

Why This Matters

Overall, the new reports suggest how rapidly the provision of abortion has adjusted amid post-Roe abortion bans in 14 states and tight restrictions in others.

The numbers may be an undercount and do not reflect the most recent shift: shield laws in six states allowing abortion providers to prescribe and mail pills to tens of thousands of women in states with bans without requiring them to travel. Since last summer, for example, Aid Access has stopped shipping medication from overseas and operating outside the formal health system; it is instead mailing pills to states with bans from within the United States with the protection of shield laws.

What’s Next

In the case that will be argued before the Supreme Court on Tuesday, the plaintiffs, who oppose abortion, are suing the Food and Drug Administration, seeking to block or drastically limit the availability of mifepristone, the first pill in the two-drug medication abortion regimen.

The JAMA study suggests that such a ruling could prompt more women to use avenues outside the formal American health care system, such as pills from other countries.

“There’s so many unknowns about what will happen with the decision,” Dr. Aiken said.

She added: “It’s possible that a decision by the Supreme Court in favor of the plaintiffs could have a knock-on effect where more people are looking to access outside the formal health care setting, either because they’re worried that access is going away or they’re having more trouble accessing the medications.”

Pam Belluck is a health and science reporter, covering a range of subjects, including reproductive health, long Covid, brain science, neurological disorders, mental health and genetics. More about Pam Belluck