salmon bias hypothesis definition

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: 13 April 2021

Salmon bias effect as hypothesis of the lower mortality rates among immigrants in Italy

• Anteo Di Napoli 1 ,
• Alessandra Rossi 1 ,
• Gianfranco Alicandro 2 , 3 ,
• Martina Ventura 1 ,
• Luisa Frova 2 &
• Alessio Petrelli 1  

Scientific Reports volume  11 , Article number:  8033 ( 2021 ) Cite this article

3982 Accesses

12 Citations

5 Altmetric

Metrics details

• Epidemiology
• Public health

Compared with natives, immigrants have lower all-cause mortality rates, despite their lower socioeconomic status, an epidemiological paradox generally explained by the healthy migrant effect. Another hypothesis is the so-called salmon bias effect: “statistically immortal” subjects return to their country of origin when they expect to die shortly, but their deaths are not registered in the statistics of the country of residence. This underestimation of deaths determines an artificially low immigrant mortality rate. We aimed to estimate the potential salmon bias effect on differences in mortality rates between Italians and immigrants. We used a national cohort of all Italians registered in the 2011 census and followed up for mortality from 2012 to 2016. Mortality data were retrieved from the Causes of Death Register, which included all deaths occurring in the country and the Resident Population Register, which collects also the deaths occurring abroad. We assumed as a possible salmon bias event the death of an immigrant resident in Italy that died in his/her country of origin. Considering the deaths occurring in the country of origin, we observed an 18.1% increase in the overall mortality rates for immigrants and an increase of 23.7% in the age-standardized mortality rate. Mortality rates of immigrants resident in Italy, calculated without taking into account the deaths occurring in the country of origin, are certainly underestimated. However, the salmon bias only partly explains the difference in mortality rates between immigrants and Italians.

Similar content being viewed by others

Nationwide longitudinal population-based study on mortality in Italy by immigrant status

Selective migration and mortality by economic status in Lolland-Falster, Denmark, 1992–2018

Research on social and economic factors influencing regional mortality patterns in China

Introduction.

Compared with natives, immigrants generally have a lower all-cause mortality rate despite their lower socioeconomic status, which is associated with poor health in terms of both morbidity and mortality 1 , 2 . Several explanations have been proposed for this epidemiological paradox.

The healthy migrant effect hypothesis posits that migration is selective of healthier individuals: migrants are healthier than the native people both of the country of origin and of the country of destination 1 , 3 , 4 , 5 .

A second hypothesis is the so-called salmon bias effect, an expression first used by Pablos-Mendez to describe “the compulsion to die in one’s birthplace” 6 . This hypothesis asserts that many immigrants return to their country of origin when they expect to die shortly 1 , 5 , 6 , 7 . If deaths occurring in their country of origin are not registered in the mortality statistics of the country of residence, some individuals become “statistically immortal”, resulting in an artificially low immigrant mortality rate 1 , 3 , 5 , 6 . The salmon bias was advanced as a possible explanation of the “Hispanic paradox”, the lower mortality rate of Hispanics than of non-Hispanic whites in the United States, despite the former having a more disadvantaged risk factor profile 1 , 3 , 5 , 6 .

Several studies have evaluated the hypothesis that any survival advantage for immigrants compared to natives may merely be a statistical artifact; the mobility of immigrant populations may cause an undercoverage of deaths and/or an overcoverage of the resident population in demographic registers 8 , 9 , 10 , 11 , 12 .

However, the salmon bias has not been convincingly documented 7 , and, to our knowledge, it has never been evaluated in Italy. Previous Italian studies found that immigrants showed a lower risk of mortality compared to Italians 2 , 13 , although unregistered remigration (delays in registration in municipal registries of the final return to the country of origin), which inflates the mortality rate denominators, has been postulated 13 .

We aimed to estimate the potential salmon bias effect on differences in mortality rates between Italians and immigrants resident in Italy.

The study was conducted using the Italian statistical registers, which are the main source of demographic statistics in our country. In particular, we used the 15th Census of Population and Housing (2011), the Causes of Death (CoD) Register, and the Resident Population (RP) Register. The RP collects individual data on demographic events occurring in Italy or abroad among the resident population, such as births, deaths, and migrations.

The study was based on a national cohort made up of all residents recorded in the 2011 Census with follow-up data for mortality from January 2012 up to December 2016 14 . Subjects entered the cohort on 1 January 2012 and were followed up until death, emigration, or last available year of mortality data (2016), whichever came first, yielding a maximum of 5 years of follow-up. Mortality data were obtained through a deterministic record linkage with the CoD Register by using the fiscal code (a unique personal identification number issued to all residents in Italy at birth or upon immigration) as linkage key. The reliability of the fiscal code was very high in all the registers, making it possible to link 97.1% of all deaths among the Census population occurring in Italy in the period 2012–2014 15 . Since there is no reason to believe that reliability of the fiscal code reported in all registers decreased over the subsequent years, the performance of the record linkage is expected to be equally high.

The CoD Register annually collects information on deaths occurring in Italy among the resident population but does not record deaths occurring abroad. To recover those deaths, we also linked the census archive and the RP Register with a deterministic procedure, using the fiscal code as linkage key. The RP Register was also used to recover the date of migration for those who had moved abroad.

The study used Istat official registers, which were checked for duplications by the Institute itself before the final release. The post-enumeration survey estimated an undercoverage rate of 1.07% for the 2011 Italian Census 16 ; missing deaths are unlikely since mortality data in Italy cover 100% of the population 17 .

We assumed as a possible salmon bias all events occurring among immigrants who were resident in Italy on the date of the census, who died in their country of origin, and who were then not recorded in the Causes of Death Register.

The country of origin was identified through citizenship for two reasons: first, in Italy, where the phenomenon of migration is relatively recent, citizenship represents the status of immigrants better than does the country of birth. Italian citizenship is acquired by foreign adults only after long, continuous residence or by children of foreigners born in Italy when they turn 18. Second, the information about country of birth was affected by too many missing values.

We calculated crude, age-specific and age-adjusted mortality rates, and the ratio between the age-adjusted mortality rate computed with and without the deaths occurring in the country of origin. Age-standardized mortality rates were computed using European population in 2013 as standard.

The cohort was conceived within the project “IF IST 2646 ‐ Socioeconomic differences in mortality”, which was included in the National Statistical Program and approved by the Italian Data Protection Authority.

The study cohort included 59,227,313 individuals (55,221,311 Italians and 4,006,002 immigrants). Immigrants were younger than Italians (median age class: 30–34 vs. 40–44), while the percentage of males was lower among immigrants than Italians (46.7% vs. 48.5%).

In the period 2012–2016, there were 17,158 deaths among immigrants occurring in Italy and another 3,102 deaths occurring in the country of origin, accounting for an 18.1% increase in detected deaths. Correspondingly, among immigrants, the crude mortality rate increased from 8.70 to 10.27 deaths per 10,000 person-years, while the standardized mortality rate increased from 38.19 to 47.23 deaths per 10,000 person-years.

Deaths of immigrants from Albania (N = 879), Morocco (N = 467), and Romania (N = 385), the three most numerous foreign communities in Italy (41.8% of the immigrant population), accounted for 55.8% of all deaths occurring in the country of origin. The detailed mortality rates for the 20 countries with the highest number of deaths are shown in Table 1 .

Figures  1 and 2 shows the age-specific mortality rates for Italians and for immigrants, calculated with and without deaths occurring in the country of origin, for men and women, respectively.

Age-specific mortality rates*10,000 person years among men resident in Italy, by citizenship and source of data. Istat, 2012–2016.

Age-specific mortality rates*10,000 person years among women resident in Italy, by citizenship and source of data. Istat, 2012–2016.

The crude mortality rate at all ages per 10,000 person-years calculated without the deaths occurring in the country of origin was: for Italians, 106.37 (95% CI: 106.31–106.42) for men and 108.11 (95% CI: 108.05–108.16) for women; for immigrants, 10.00 (95% CI: 9.93–10.06) for men and 7.56 (95% CI: 7.51–7.61) for women. Instead, when considering also the deaths occurring in the country of origin, it was 11.91 (95% CI: 11.84–11.98) for men and 8.84 (95% CI: 8.78–8.89) for women.

After age standardization, the mortality rate per 10,000 person-years decreased to 101.75 (95% CI: 101.58–101.92) and 67.70 (95% CI: 67.59–67.81) for Italian men and women, respectively. For immigrants, the age-standardized rates, calculated without the deaths occurring in the country of origin, were 45.06 (95% CI: 43.66–46.51) for men and 33.64 (95% CI: 33.62–34.70) for women. Instead, when also considering the deaths occurring in the country of origin, these rates increased to 57.69 (95% CI: 56.41–58.96) for men and 40.40 (95% CI: 39.46–41.34) for women.

Figure  3 shows the ratio of the age-specific mortality rates among immigrants, calculated with and without deaths occurring in the country of origin, with an average excess of about 19% for men and 17% for women. The ratio was particularly high for subjects aged 65–69 (1.27 for men and 1.24 for women), 70–74 (1.35 for men and 1.36 for women), and 75–79 (1.42 for men and 1.34 for women). Similar patterns in ratios were found for subjects aged ≥ 65 when focusing the analyses on the three most numerous foreign communities in Italy: the Romanian (men: 1.40, women: 1.29), Albanian (men: 1.77, women: 1.29) and Moroccan (men: 1.84, women: 1.96) communities (data not shown in figure).

Ratio between age-specific mortality rates of immigrants resident in Italy, calculated including and excluding deaths occurred in the country of origin, by sex. Istat, 2012–2016.

In our study we observed that considering the deaths occurring in the country of origin increased the overall mortality rates of immigrants resident in Italy by 18.1% and the age-standardized mortality rate by 23.7%.

Indeed, the age-standardized mortality rates for immigrants computed without the deaths occurring in the country of origin were much lower than those for Italians and remained lower even when we considered the deaths occurring in the country of origin.

Thus, it can be said that studies based only on the CoD Register in the country of residence underestimate the mortality rates of immigrants resident in Italy.

The estimation of mortality among the immigrant population is subject to a number of data problems due to the limits in recording mobility among foreign-born populations 8 , 11 . National vital registration systems are poorly equipped to gather information on residents dying abroad, in particular if they have foreign citizenship. The problem concerning data mismatch arises when deaths occur without an official residence change, so that person-years are still accounted for in national mortality rate estimates 11 .

The detection of “statistically immortal” subjects could fall within the definition of the salmon bias, which refers to those deaths occurring in the country of origin but not registered in Civil Registry Office of the country of residence. This results in an underestimation of the numerator and, consequently, an artificially low immigrant mortality rate 1 , 3 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 .

Another point to support the presence of the salmon bias is that we found the greatest differences between the two methods for calculating the deaths of immigrants among the elderly, supporting the hypothesis that if deteriorating health triggers return migration, this phenomenon is more pronounced at older ages, when health problems become more prevalent 5 , 7 , 13 . These differences are confirmed by our findings within the three largest foreign communities in Italy (Romania, Albania, Morocco). Some authors have described this as a differential remigration by age, which can affect especially those immigrants with health problems and/or low socioeconomic profiles (unhealthy remigration effect) 13 .

Once having emigrated, individuals are less likely to have their vital events documented in their country of origin, where they are still registered and continue to be erroneously regarded as being at risk of vital events. This represents a potential source of bias in register-based censuses, defined as overcoverage, which has been postulated as one of the explanations of the migrant mortality paradox, in particular at peak migration ages 8 , 9 , 11 .

Among immigrants, mortality generally shows a U-shaped age pattern, with excess at young and older ages compared to natives, with a large advantage at adult ages 8 . The problem of the reliability of very low estimates of migrant mortality at ages 65 and over has been raised, given the risk of immigrants’ potentially not having un-registered from host country registries 2 , 12 . In our study we observed the largest differences in mortality rates between immigrants and Italians at older ages.

However, the retrieved information on the unregistered deaths of immigrants, even if we hypothesize the presence of the salmon bias effect, is not enough to explain the large difference in mortality rates compared to those of Italians. In fact, the age-specific and standardized mortality rates of immigrants remain much lower than those observed among Italians, even with the addition of the deaths detected by the new source.

The salmon bias effect may contribute to explain the observed difference in mortality rates between Italians and immigrants; the role of the healthy migrant effect remains prominent, as observed also by other studies that tested the hypothesis of potential statistical artifact 8 , 9 .

The better health status of the immigrant population compared to that of Italians could be ascribed to the well-known positive selection effect: healthier people are more likely to migrate 2 , 10 , 13 .

An interesting research hypothesis that could confirm the effect of the salmon bias is to verify the increase in mortality rates of those immigrants who have returned compared to those of the general population of their country of origin. In fact, if the low mortality among immigrants compared to that of the country of destination is partly explained by the salmon bias, we would expect increased mortality among repatriates compared with the population in the country of origin, in particular in the first few years after repatriation 7 , 10 . Such a study could be conducted, for example, also on the population of Italian emigrants who have returned from abroad over the past few decades.

The study has some limitations.

First, our findings represent only indirect empirical evidence that could either refute or support the salmon bias hypothesis as an explanation for the lower mortality observed among immigrants compared to Italians. This evidence is thus probably insufficient to confirm the salmon bias hypothesis.

Second, we do not know the causes of death of individuals who died in their country of origin since this crucial information is not collected in the RP Register. In fact, in the presence of the salmon bias, one would expect higher mortality rates for some specific causes, for example, cancer, especially immediately after returning to one’s country of origin 10 . Moreover, it was not possible to determine whether immigrants died in their country of origin of acute causes, such as infectious diseases or accidents, but this would not in any case contribute to the salmon bias hypothesis.

Third, due to the characteristics of the data source used to collect deaths occurring abroad, the date of the return to the country of origin is unknown because for the RP Register, these individuals are still resident in Italy. The RP Register can only integrate the information about deaths of immigrants undetected by the routine CoD Register source.

Finally, we do not know whether all immigrants who die in their country of origin are captured in RP Register. In any case, this limitation would not conflict with the hypothesis of the presence of a salmon bias effect.

To our knowledge, this is the first study that has attempted to estimate whether the salmon bias effect has any role in explaining the lower mortality of immigrants in Italy than that of Italians. Another strength of the study is the integration of two complete and reliable national archives, the CoD Register and the RP Register.

Conclusions

The mortality rates of immigrants resident in Italy, calculated so far without taking into account the deaths of subjects who died abroad, are certainly underestimated. Future studies on the mortality of immigrants in Italy will necessarily have to take this into account. The lower mortality rate among immigrants compared to that of natives is a real phenomenon, but researchers must take into account the potential biases when they estimate the extent of the advantage.

Abraído-Lanza, A. F., Dohrenwend, B. P., Ng-Mak, D. S. & Turner, J. B. The Latino mortality paradox: a test of the “salmon bias” and healthy migrant hypotheses. Am J Public Health 89 , 1543–1548 (1999).

Article   Google Scholar  

Pacelli, B. et al. Differences in mortality by immigrant status in Italy. Results of the Italian network of longitudinal metropolitan studies. Eur J Epidemiol. 31 , 691–701 (2016).

Ruiz, J. M., Steffen, P. & Smith, T. B. Hispanic mortality paradox: A systematic review and meta-analysis of the longitudinal literature. Am J Public Health 103 , e52–e60 (2013).

Ullmann, S. H., Goldman, N. & Massey, D. S. Healthier before they migrate, less healthy when they return? The health of returned migrants in Mexico. Soc Sci Med. 73 , 421–428 (2011).

Turra, C. M. & Elo, I. T. The impact of salmon bias on the hispanic mortality advantage: New evidence from social security data. Popul. Res. Policy Rev. 27 , 515–530 (2008).

Pablos-Mendez, A. Mortality among Hispanics. JAMA 271 , 1237 (1994).

Article   CAS   Google Scholar  

Kristensen, P. & Bjerkedal, T. Dealing with emigration in cohort studies: follow-up of mortality and cancer incidence among Norwegians born between 1967 and 1976. Eur. J. Epidemiol. 25 , 155–161 (2010).

Guillot, M., Khlat, M., Elo, I., Solignac, M. & Wallace, M. Understanding age variations in the migrant mortality advantage: An international comparative perspective. PLoS ONE 13 (6), e0199669 (2018).

Wallace, M. & Wilson, B. Age variations and population over-coverage: Is low mortality among migrants merely a data artefact?. Popul. Stud. 10 , 1–18 (2021).

Google Scholar  

Puschmann, P., Donrovich, R. & Matthijs, K. Salmon bias or red herring? Comparing adult mortality risks (ages 30–90) between natives and internal migrants: Stayers, returnees and movers in Rotterdam, the Netherlands, 1850–1940. Hum. Nat. 28 , 481–499 (2017).

Monti, A., Drefahl, S., Mussino, E. & Härkönen, J. Over-coverage in population registers leads to bias in demographic estimates. Popul Stud. 74 (3), 451–469 (2020).

Kibele, E., Scholz, R. & Shkolnikov, V. M. Low migrant mortality in Germany for men aged 65 and older: Fact or artifact?. Eur J Epidemiol. 23 (6), 389–393 (2008).

Fedeli, U., Ferroni, E., Pigato, M., Avossa, F. & Saugo, M. Causes of mortality across different immigrant groups in Northeastern Italy. PeerJ 3 , e975. https://doi.org/10.7717/peerj.975 (2015).

Article   PubMed   PubMed Central   Google Scholar  

Alicandro, G., Grande, E., Sebastiani, G. & Frova, L. The 2011 Italian census cohort for the study of socioeconomic inequality in mortality. Int. J. Public Health 65 , 693–697 (2020).

Frova, L., Alicandro, G., Demuru, E. & Sebastiani, G. Integrazione di archivi nazionali per lo studio delle diseguaglianze socioeconomiche nella mortalità in Italia. Istat Working Papers 2/2019. ISBN: 978-88-458-1986-5 (2019). https://www.istat.it/it/files//2019/08/IWP-2-2019.pdf . Accessed 8 March 2021.

National Institute of Statistics (Istat). Post Enumeration Survey (PES) of the 15th Italian population census (2015). https://www.istat.it/en/archivio/145236 . Accessed 8 March 2021.

Eurostat, Causes of deaths, Metadata (Italy) (2017). https://ec.europa.eu/eurostat/cache/metadata/EN/hlth_cdeath_esms_it.htm . Accessed 08 March 2021.

Download references

Acknowledgements

We thank Jacqueline M. Costa for English language editing.

Author information

Authors and affiliations.

National Institute for Health, Migration and Poverty (INMP), Istituto Nazionale per la promozione della salute delle popolazioni Migranti e per il contrasto delle malattie della Povertà (INMP), Via di S. Gallicano 25/a, 00153, Roma, Italy

Anteo Di Napoli, Alessandra Rossi, Martina Ventura & Alessio Petrelli

National Institute of Statistics (Istat), Viale Liegi 13, 00198, Rome, Italy

Gianfranco Alicandro & Luisa Frova

Department of Pathophysiology and Transplantation, Università degli Studi di Milano, Milan, Italy

Gianfranco Alicandro

You can also search for this author in PubMed   Google Scholar

Contributions

A.D.N. took part in the conceptualization of the study, bibliographic research, development and implementation of methods, statistical analysis and preparation of manuscript; A.R. took part in the development and implementation of methods and statistical analysis, and preparation of manuscript; G.A. took part in the development and implementation of methods and statistical analysis, and preparation of manuscript; M.V. took part in the development and implementation of methods and statistical analysis; L.F. took part in the development and implementation of methods; A.P. took part in the conceptualization of the study, bibliographic research, development and implementation of methods, and preparation of manuscript.

Corresponding author

Correspondence to Anteo Di Napoli .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

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

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.

Di Napoli, A., Rossi, A., Alicandro, G. et al. Salmon bias effect as hypothesis of the lower mortality rates among immigrants in Italy. Sci Rep 11 , 8033 (2021). https://doi.org/10.1038/s41598-021-87522-2

Download citation

Received : 22 October 2020

Accepted : 19 March 2021

Published : 13 April 2021

DOI : https://doi.org/10.1038/s41598-021-87522-2

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

This article is cited by

Cancer, cardiovascular disease, and all-cause mortality in iraqi- and swedish-born individuals in sweden: the medim cohort study.

• Nadine Fadhel Dhaher
• Miriam Pikkemaat
• Louise Bennet

Scientific Reports (2023)

• Anteo Di Napoli
• Martina Ventura
• Alessio Petrelli

Scientific Reports (2022)

Mortality differences between migrants and Italians residing in Rome before, during, and in the aftermath of the great recession. A longitudinal cohort study from 2001 to 2015

• Eleonora Trappolini
• Claudia Marino
• Laura Cacciani

BMC Public Health (2021)

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 newsletter — what matters in science, free to your inbox daily.

Salmon Bias or Red Herring?

Comparing Adult Mortality Risks (Ages 30–90) between Natives and Internal Migrants: Stayers, Returnees and Movers in Rotterdam, the Netherlands, 1850–1940

• Open access
• Published: 17 October 2017
• Volume 28 , pages 481–499, ( 2017 )

Cite this article

You have full access to this open access article

• Paul Puschmann 1 , 2 ,
• Robyn Donrovich 2 &
• Koen Matthijs 2  

4055 Accesses

24 Citations

6 Altmetric

Explore all metrics

The purpose of this research is to empirically test the salmon bias hypothesis, which states that the “healthy migrant” effect—referring to a situation in which migrants enjoy lower mortality risks than natives—is caused by selective return-migration of the weak, sick, and elderly. Using a unique longitudinal micro-level database—the Historical Sample of the Netherlands—we tracked the life courses of internal migrants after they had left the city of Rotterdam, which allowed us to compare mortality risks of stayers, returnees, and movers using survival analysis for the study group as a whole, and also for men and women separately. Although migrants who stayed in the receiving society had significantly higher mortality risks than natives, no significant difference was found for migrants who returned to their municipality of birth (returnees). By contrast, migrants who left for another destination (movers) had much lower mortality risks than natives. Natives who left Rotterdam also had significantly lower mortality risks than natives who stayed in Rotterdam. Female migrants, in particular, who stayed in the receiving urban society paid a long-term health price. In the case of Rotterdam, the salmon bias hypothesis can be rejected because the lower mortality effect among migrants was not caused by selective return-migration. The healthy migrant effect is real and due to a positive selection effect: Healthier people are more likely to migrate.

Similar content being viewed by others

Moving beyond salmon bias: mexican return migration and health selection.

Investigating Health Selection Within Mexico and Across the US Border

Healthy migrants health selection of internal migrants in germany.

Avoid common mistakes on your manuscript.

Many European and North American studies report that migrants have lower mortality risks than the native-born population (Alter and Oris 2005 ; Markides and Eschbach 2005 ). This so-called healthy migrant effect is found in both contemporary and historical settings, and it has been explained by differences in early life conditions (Alter and Oris 2005 ), healthy lifestyles and behaviors (Abraído-Lanza et al. 2005 ; Lariscy et al. 2015 ), as well as in terms of selection effects (Oris and Alter 2001 ). With regard to the latter, most studies focus on the idea of a positive selection effect in the area of origin, in the sense that individuals who are young and healthy are more able and more likely to move than the sick, weak, elderly, and disabled (Khlat and Darmon 2003 ). For the nineteenth-century Belgian village of Sart, Oris and Alter ( 2001 ) observed, for example, that individuals from families that experienced death among their members during the previous two years were much less likely to out-migrate, compared with those from families in which everybody had survived.

Ever since the healthy migrant effect—initially referred to as an epidemiological paradox and later as the Hispanic or Latino paradox—was discovered among Latin American migrants in the US (Markides and Coreil 1986 ), scholars have doubted whether the results of such analyses are real, or if the healthy migrant effect is merely a statistical artifact resulting from measurement errors or biases toward healthy migrants. Not only are the results counterintuitive, and—at first glance—hard to reconcile with long-standing insights into health and mortality, but also the data on migrants are often of poorer quality than that of the native population (Razum 2006 ; Redstone Akresh and Frank 2008 ).

Doubts about the validity of the observed healthy migrant effect led to the formulation of the salmon bias hypothesis, which states that the observed lower mortality risks among migrants are the result of selective return-migration of the sick and elderly and those who are unable to adapt to and endure harsh working and living conditions (Deboosere and Gadeyne 2005 ). If migrants indeed have a tendency to go home before they die, their deaths do not contribute to the national death statistics in the country of study, but rather in the country of origin. If the second out-migration is not registered, this would lead to a situation in which migrants become “statistically immortal” in the society under study (Abraído-Lanza et al. 1999 ). Even if out-migration of the sick is registered, this can lead to measurement error since the presence of migrants in a society who do not die there is likely to lead to an inflated denominator, causing an artificially lowered mortality rate among migrants. These doubts suggest an overestimation of the healthy migrant effect, or even question the very existence of a health advantage of migrants.

In this paper we test the salmon bias hypothesis for internal migrants, both men and women, in Rotterdam during the latter half of the nineteenth and the early twentieth centuries by systematically comparing mortality risks among stayers and leavers, subdividing the latter category into returnees (migrants who return to their municipality of birth) and movers (migrants who moved to another destination than their municipality of birth). Given that women at the time had different migration patterns than men—they moved more often, but over shorter distances (Greefs and Winter 2016 )—and faced diverging mortality risks in later life (Mourits 2017 ), we look at mortality risks not only for the entire study group, but also separately for women and for men.

Rotterdam was selected as a case study to test the salmon bias hypothesis for two main reasons. First in a previous study (Puschmann et al. 2016 ) comparing adult (ages 30+) mortality risks among migrants and natives in Antwerp, Rotterdam, and Stockholm (1850–1930), we found that the healthy migrant effect was particularly strong among internal migrants who moved to Rotterdam. We revealed that the health advantage of migrants in the port cities under study was related to, among others, the early life environment and positive selection effects. With respect to the latter, our previous findings showed an inverse relationship between migration distance and mortality risks. This led us to the conclusion that the most physically and mentally fit were more likely to migrate over long distances. Although we censored individual migrants upon out-migration, we wanted to test in a more systematic way whether selective return migration had biased the results of our event history analysis. This decision was strengthened by the fact that we found that certain subgroups of migrants in the population actually experienced excess mortality. In the city of Rotterdam this was the case for Italian and Italian-speaking Swiss immigrant men.

The second reason for choosing Rotterdam to test the salmon bias hypothesis is related to the nature of the data. The Historical Sample of the Netherlands (Mandemakers 2000 ) allows us to follow the life course of migrants (and natives) who left Rotterdam, at least as long as they moved within the national borders (97% of all internal migrants). Typically, the life courses of leavers are truncated upon out-migration in historical and contemporary datasets. Consequently, previous studies have, at best, only been able to estimate to what degree selective return-migration might have biased their results. Such estimations have led to contradictory and inconclusive results. Whereas some studies report that the healthy migrant effect is indeed caused by selective return-migration of the sick, weak, and elderly (Lu and Qin 2014 ), others found that this phenomenon only partially contributed to the observed effect (Khlat and Courbage 1996 ; Turra and Elo 2008 ), while still others reached the conclusion that it did not have an impact at all (Abraído-Lanza et al. 1999 ; Deboosere and Gadeyne 2005 ; Wallace and Kulu 2014 ).

Because the data allow us to track the final migration destination of internal migrants within the Netherlands, this case study can shed light on whether the salmon bias hypothesis can (partly) explain the mortality advantage of internal migrants we observed in our previous study (Puschmann et al. 2016 ).

Research Setting

Like other European port cities at the time, Rotterdam received growing numbers of internal migrants during the latter half of the nineteenth and the early twentieth centuries. Thanks to urban in-migration and natural population growth, the population of Rotterdam increased from just over 90,000 inhabitants in 1850 to 332,000 in 1900, reaching 598,000 inhabitants by 1930. The majority of the internal migrants originated from the rural municipalities of the province of Zuid-Holland, which meant that the largest share of the newcomers in Rotterdam were peasants and agricultural laborers who were born in Rotterdam’s direct hinterland. The Dutch provinces of Noord-Brabant and Zeeland were also important sending areas (Van der Harst 2006 ). Among the urban immigrants in Rotterdam, women slightly outnumbered men (Puschmann 2015 ), and single women were particularly active as domestic servants (Bras 2003 ). Declining opportunities in the agricultural sector and the gradual destruction of the family economy were the main push factors for both men and women. The share of international migrants was stable at around 3% between 1850 and 1930 and consisted mainly of Germans, which is not surprising given the important trade relations with the German hinterland (Puschmann 2015 ).

Rotterdam’s attractiveness to migrants was mainly related to the growing employment opportunities in the port sector, industry, construction, and services. Thanks to the construction of the Nieuwe Waterweg—a direct connection between Rotterdam and the North Sea—and the high-speed industrialization of the German Ruhr and Rhine areas, Rotterdam swiftly turned into Europe’s largest port city. The port clearly functioned as an important pull factor for male internal and international migrants. By 1909, about 55% of the city’s working population was engaged in the port sector (Van de Laar 2000 ). The growth of the port went hand in hand with the revival of old industries and the advent of new industries, which usually concentrated on the handling of raw materials which arrived in the port. The construction of railways and tramways, as well as the introduction of busses, facilitated the migration of thousands of migrants but slowed down migration in the course of the first half of the twentieth century, as it allowed growing numbers of people to commute to Rotterdam (Puschmann 2015 ).

Bouman and Bouman ( 1955 ) showed that especially rural-to-urban migrants had a hard time integrating in Rotterdam. They put forward that newcomers were uprooted and ended up in a struggle for survival, as they landed in badly paid and dangerous jobs in the port and in construction, which made it difficult for them to make a living. Simultaneously they no longer could count on the social network in their home village, and it was difficult to create new social ties in Rotterdam, especially since the newcomers were—because of their dialect, low socioeconomic status, and different lifestyle—being looked down upon by the native Rotterdam population. More recent research has led to considerable changes to this picture. Internal migrants indeed entered the labor market at lower levels, but stayers were able to catch up with natives and even to outperform them in the long run (Puschmann 2015 ). Nevertheless, being born in the countryside was associated with substantially lower social status. Next, the facts that internal migrants in Rotterdam disproportionally stayed single, and that those who married did so on average at a later age, show that the social integration of internal migrants was indeed hampered, although this time rural-to-urban migrants were not disfavored relative to urban-to-urban migrants. Internal migrants in other port cities, including Antwerp and Stockholm, faced similar challenges (Puschmann et al. 2015 ).

The data for the analyses was retrieved from the Data Set Life Courses Release 2010.01 from the Historical Sample of the Netherlands, a large historical demographic database with life course information on individuals born in the period 1812–1922 (Mandemakers 2000 ). The data are derived from the Dutch population registers as well as the vital registration of births, marriages, and deaths. The data collection began with a random sample of birth certificates, and the database makers aimed to “reconstruct” as many full life courses as possible. The Data Set Life Courses Release 2010.01 consists of 44,252 life courses, of which 62% are complete. Since the database managers have provided start and end dates for the periods in which the life course information of the research person is complete, we can determine the risk period for all individuals, with both full and partial life courses, in our survival models.

Study Population

We selected all individuals who lived in Rotterdam and its suburbs at some point between 1850 and 1940 (after their thirtieth birthday). This resulted in a dataset consisting of 1,452 research persons (756 natives and 696 migrants). Research persons who were born in Rotterdam are considered natives ; individuals who were born elsewhere in the Netherlands and moved at some point to Rotterdam are migrants . The distinction between stayers, returnees, and movers is based on a combination of observed places of death, the declared destination of the last out-migration as it was specified in the population register of Rotterdam, and the place where a person was last recorded. The latter two criteria were only taken into account if the person was still alive at the end of the research period, or in case the place of death was unknown ( n  = 281). An overview of the classification of the research persons into natives, migrants, stayers, leavers, returnees, and movers is presented in Fig.  1 .

Overview of the different categories of migrants and natives in the analyses

For all individuals, all life course information from the database was retrieved, and individuals were only censored if they were still alive at the end of the study period or if they left the country. In our study group, 67% of all life courses are complete.

We included several fixed and time-varying variables in our analyses. The variable migration status is coded as native for those born in Rotterdam and migrant for those born outside of Rotterdam (but within the Netherlands). The variable stayers/leavers divides the study group into those who stayed in Rotterdam and those who left the city. Age at arrival notes the age that the migrant first arrived in the city and is divided into four categories: ages <15, 15–24, 25+, or unknown. In the analyses that include both men and women in the study group, we distinguish sex as women and men. Since birth dates spanned 60 years, we used birth cohort to categorize research persons into groups born 1850–1869, 1870–1889, and 1890–1910. Two variables were treated as time-varying, being updated from age 30 until the end of analysis time: civil status and occupation . Civil status was grouped into four categories: unmarried, married, separated/widowed, and unknown. Occupation is based on the HISCO codification (Van Leeuwen et al. 2002 ) and recoded into HISCLASS (Van Leeuwen and Maas 2011 ), and further categorized into four groups: professionals, foremen and skilled, day laborers and unskilled, and unknown. Finally, in order to identify the study population as natives, stayers, returnees, or movers, we included the variable last destination . This variable is first grouped as natives, migrants who stayed in Rotterdam, migrants who returned home, and migrants who migrated somewhere else. Later we further divided the native population into two groups: those who stayed and those who left.

Statistical Analysis

We conduct survival analysis, first making use of Kaplan-Meier survival estimates to get an initial impression of the mortality differences for different categories of natives and migrants. In order to adjust for other factors, such as birth cohort, age at first in-migration, civil status, and occupation, we fit Gompertz proportional hazard models with all-cause mortality specified (ages 30+) as the failure event. The Gompertz model was chosen because it has been shown to fit adult human mortality well between ages 30 and 90 (Cleves et al. 2008 ). Further, on the basis of AIC/BIC testing criteria, the Gompertz model best fit our data (with the lowest AIC and BIC scores) compared with other parametric and semi-parametric models that were also tested (Cox proportional hazards, Weibull, and the exponential model).

By including the survival history of migrants upon departure, a more formal way of testing the salmon bias hypothesis than has been the case in previous studies is possible. In this way, we compare the mortality risks of three different groups of migrants—stayers, returnees, and movers—with that of the native population in order to determine if the observed healthy migrant effect was a result of selective out-migration. In the first analysis, we divide the migrant population into stayers and leavers. Next, we further divide the leavers into migrants who returned to their municipality of birth and migrants who moved to another destination. If we find that returnees have much higher mortality risks than the native population, we can confirm the salmon bias hypothesis. If our findings show, by contrast, no significant difference between natives and returnees or a lower mortality risk among the latter category we will reject the hypothesis and conclude that it is a “red herring.” Finally, we add a new element to the discussion by dividing the native population also into stayers and leavers. Too often, natives have been considered as a static category, even though a considerable share became migrants in the course of their lives. It is worth evaluating whether mortality risks also differed between natives who never left Rotterdam and natives who did leave since the healthy migrant effect suggests a self-selection mechanism in terms of health in the place of origin. Consequently, we should be able to find such a mechanism for their native counterparts. Our findings based on this distinction of separate categories of natives (those who stayed and those who left) are displayed in the Results.

We aimed for parsimonious models in order to maximize the statistical power for the newly added variables since our sample is relatively small. In order to benefit from the largest sample size possible, our main effects models are first presented for both sexes combined. We opt for nested models in which we include only those variables that improve the fit of the model, which are organized in a series of six models in which each additional variable was tested by use of log likelihood ratio tests. Based on these tests, two variables that we tested were not included in the analyses because they did not lead to a better fit: urban-rural birthplace and distance from birthplace. The final model, incorporating the main variables of interest and other controls from early and later life, leads to the best fit. See the Electronic Supplementary Materials (Table ESM- 1 ) for descriptive statistics of all variables.

Next, we present the full models which we ran separately for men and women, given that there are sex differentials in mortality, in general, as well as that variables related to migration may differ for men and women—for example, propensities to migrate, reason to migrate, distance traveled, and ages at migration (cf. Greefs and Winter 2016 ; Greenwood 2008 ; Mourits 2017 ). These models were also designed as nested, but for simplification we present only the full models (the nested models are shown in Tables ESM- 2 and 3 ). To more directly compare men and women with each other, we included interaction terms to measure how male and female natives and different groups of migrants differed in terms of mortality risk.

Mortality Risks among Natives, Stayers, and Leavers

Figure  2 shows the Kaplan-Meier survival estimates for natives and migrants, with the latter category subdivided into stayers and leavers. The graph shows that stayers have higher mortality risks than natives shortly after they enter the risk set, which might be related to the stress of first moving to an alien environment. However, between 25 and 35 years of analysis time the mortality risk of stayers is lower than that of the natives. From 37 years on their survival rates drop below those of the natives, suggesting that the stayers paid a long-term price for their move to the city. This might be related to the lack of a social network in a society with no national pension and health care system, in which the elderly and sick were dependent on care from their family, friends, and neighbors. However, the experience of the leavers is completely different. For the first 10 analysis years their mortality risks are comparable to those of natives, but subsequently their survival rates become considerably higher than those of the natives and the stayers. It seems therefore that this group was particularly healthy.

Kaplan-Meier survival estimates by stayers, leavers, and natives

In Table  1 , our nested models for both sexes are presented. Model I contains only the migration status variable. Unsurprisingly, migrants had a lower (though insignificant) risk of dying than natives (HR = 0.89; 95%; CI: 0.74–1.07). In Model II the stayer-leaver variable is included and the migration status variable becomes significant and its effect size strengthens (HR = 0.79; 95% CI: 0.63–0.99). Stayers have a 37% higher mortality risk compared with leavers, significant at the 5% level (95% CI: 1.02–1.85). In Model III age at in-migration is added. The migration status variable now loses its significance and the effect changes (HR: 1.02; 95% CI: 0.79–1.30), although the stayer-leaver variable stays significant and the effect remains stable. Migrants who arrived before their fifteenth birthday have a much lower risk of dying (HR: 0.26; 95% CI: 0.08–0.62) than the reference category of migrants who moved to Rotterdam after their twenty-fifth birthday, which is significant at the 5% level. The same is true for migrants who arrived between their fifteenth and twenty-fifth birthdays, but the effect size is considerably smaller (HR: 0.62; 95% CI: 0.35–1.07). Migrants who arrived at unknown ages had a significantly higher risk of dying (at the 10% level) compared with the reference category of migrants arriving at the ages of 25+ (HR: 0.62; 95% CI: 0.97–1.90). Sex is adjusted for in Model IV, which has no major influence on the other variables (only unknown age at arrival becomes more significant). In Model V we adjust for birth cohort, significant at the 0.01% level, which suggests an increase in the mortality risk for each successive cohort, most likely related to industrialization. In Model VI the time-varying covariates civil status and occupation were added to the models. These adjustments led to a stronger healthy migrant effect (although the variable stays insignificant) and an increase in the hazard ratio of the stayers, now with 83% higher mortality risk than leavers. The effects for age at arrival weakened, and the category of ages 15–24 becomes insignificant. The unknown age category, however, becomes stronger and highly significant. As it turns out, singles had a higher mortality risk than the reference category of married people (HR: 1.58; 95% CI: 1.18–2.09), and widowed and separated individuals also had a higher mortality risk (HR: 1.99; 95% CI: 1.42–2.66). An even stronger effect was found for individuals with unknown marital status (HR: 3.01; 95% CI: 2.37–3.81). For occupation we found that the foremen and skilled workers had higher mortality risk than the reference category of professionals (HR: 1.30; 95% CI: 0.97–1.72).

Given that migration variables could have a different relationship with adult mortality for men and women, Table  2 shows the full models separately by sex. A first observation is that, apart from the marital status variable, the effects are in the same direction, but less strong for men than for women, and the results are considerably more often significant among women. The latter is most likely related to the somewhat smaller sample size and the fewer number of failures (deaths) among men.

Only for women do we find a significant mortality advantage for the migrants compared with the reference category of natives (HR: 0.75; 95% CI: 0.54–1.05). For men the HR is 0.86, but not significant (95% CI: 0.57–1.31). Among the women the stayers had a 2.3 times higher mortality risk than the leavers, significant at the 0.001 level (95% CI: 1.58–3.56). Among the men the effect was in the same direction, but weaker and not significant (HR: 1.37; CI: 0.75–2.51). The younger the migrant women and men were when they arrived in the city, the lower their mortality risks were. Although this effect was only significant for women in the age category <15 (at the 0.1% level), this finding is particularly strong (with a HR of 0.19) relative to the reference category of ages 25+. For both men and women, migrants who arrived at an unknown age had a much higher (and highly statistically significant) mortality risk. Next, for men and women we observe significantly higher mortality risks for the cohort 1870–1889 and 1890–1910 compared with the reference cohort of 1850–1869. Unmarried men and women had higher mortality risks than married individuals. Widowed and separated women had a highly significant higher mortality risk compared with married women (HR: 2.35; 95% CI: 1.70–3.29). Additionally, men and women with an unknown marital status had a significantly higher mortality risk compared with the reference categories of married men and married women. Regarding occupation, we found significant effects only among women with an unknown occupation. They had a higher mortality risk compared with the reference category of professionals (HR: 1.58; 95% CI: 1.02–2.44).

Mortality Risks among Natives, Stayers, Returnees, and Movers

Figure  3 shows the Kaplan-Meier curves for natives, migrants who stayed in Rotterdam, movers, and return migrants. The curves show that movers had much lower mortality risks than natives and migrants who stayed in Rotterdam. The survival rates of return migrants were somewhat below that of natives and migrants who stayed in Rotterdam, but the difference was not as pronounced as one would expect on the basis of the salmon bias hypothesis. After 20 years of analysis time the survival estimates become less reliable owing to small sample size. Judging on the basis of these K-M curves, the healthy migrant effect is caused by the group of movers, who have much lower mortality risks than all other groups.

Kaplan-Meier survival estimates by last destination

Next, Table  3 shows the three fully adjusted Gompertz models for both sexes and for women and men separately. In the first model (both sexes combined), migrants who stayed in Rotterdam had a significantly higher mortality risk than the native population (HR: 1.67; 95% CI: 1.18–2.21). However, no significant difference in the mortality risks between natives and return migrants was found, and the effect size is so small that it cannot account for any healthy migrant effect (HR: 1.13; 95% CI: 0.72–1.98). Movers, by contrast, had a considerably lower mortality risk than the native population (HR: 0.76; 95% CI: 0.59–1.00) and was highly significant. For the women we observe the same results, but the effects for stayers (HR: 1.85; 95% CI: 1.29–2.66) and movers (HR: 0.71; 95% CI: 0.50–1.00) were stronger, while the effect size for the returnees was even smaller. In the model including only men, no significant results were found.

Extending this analysis, in order to compare both men and women among natives, stayers, returnees, and movers, we ran a model including interaction terms for sex and the last destination variable (Fig.  4 ). Compared with native men, only returnees had elevated mortality risks, with about 15% higher mortality, but the result was not significant (HR: 1.15; 95% CI: 0.52–2.50). Both stayers (HR: 0.83) and movers (HR: 0.70) had lower mortality risks, with the latter category significant at the 10% level. For women, relative to the reference category of native men, there were significantly lower mortality risks for natives (HR: 0.73; 95% CI: 0.56–0.94) and for movers (HR: 0.56; 95% CI: 0.38–0.80). Female returnees also had lower mortality risks, though not statistically significant (HR: 0.85; 95% CI: 0.43–1.65). The only female group with excess mortality was found for stayers, with just over 40% higher mortality risk compared with native men, significant at the 10% level (HR: 1.42; 95% CI: 0.98–2.04). The latter result is quite striking. It suggests that women who stayed in Rotterdam were a less favorable selection of the population of origin in terms of health and human capital, and/or that they paid a higher health price for their migration. The former is underlined by a recent study by Hilde Greefs and Anne Winter, in which they showed that women who moved over a longer distance to Antwerp in the latter half of the nineteenth century were—contrary to men—more often from a more modest background (Greefs and Winter 2016 ). At the same time it is not unthinkable that women—because of their limited human capital—became marginalized upon arrival in Rotterdam. Others with more means might have moved on, whereas those who stayed in contact with their family of origin and the community in which they grew up might have returned after domestic service.

Relative mortality risks by sex and last destination (fully controlled; significant results noted)

Comparing Mortality Risks of Stayers and Leavers for both Migrants and Natives

Finally, we divide the native population also into stayers and leavers. The Kaplan-Meier curves in Fig.  5 show that natives who left Rotterdam had higher survival probabilities than natives who stayed. The K-M curve of the leaving natives is similar to that of migrants who moved to another destination in the Netherlands. However, between 15 and 35 years of analysis time, the survival probabilities of the leaving natives were even lower than that of the moving migrants.

Next, Fig.  6 shows the Gompertz model with the distinction between stayers and leavers among the native population in three models: both sexes combined, only women, and only men. The models are adjusted for age, age at arrival, sex (in the combined model), birth cohort, and time-varying civil status and occupation. Compared with the reference category of native stayers, natives who left Rotterdam have highly significant relative mortality risks at just under HR = 0.6 across all models. Similarly, nearly the same results, in terms of strength and significance, were found for migrants who left for another destination. No significant differences were found for return migrants relative to the reference category in all three models. Overall, men and women had similar results with the exception of migrants who stayed in Rotterdam. Female migrants who stayed had around 47% higher relative mortality risk than female natives, significant at the 5% level (HR: 1.47).

Relative mortality risks by last destination for both sexes, women and men separately (controlled for age, age at arrival, birth cohort, and time-varying civil status and occupation)

+ p  < 0.10, * p  < 0.05, ** p  < 0.01, *** p  < 0.001

Discussion, Policy Implications, and Future Research

The analyses in this paper show that we can reject the salmon bias hypothesis for our specific case study. Even though we found some elevated mortality risk for male returnees at first glance, the observed lower mortality among internal migrants in Rotterdam are real and were not caused by selective return migration of the sick, weak, and elderly because no significant difference in mortality risks between returnees and the native population was found. The salmon bias hypothesis is thus a red herring in the case of late-nineteenth- and early-twentieth-century Rotterdam. Movers experienced the lowest relative mortality risks of the entire population under study. If we factor in that natives who left Rotterdam also had significantly lower mortality risks than natives who stayed in the Dutch port city, we can only conclude that migration and good health are even more strongly correlated than we could imagine on the basis of the previous studies: The healthier people were, the more they moved, and this was true for both men and women, who experienced similar effects in terms of their migration patterns.

Although men and women experienced similar effects overall, we further investigated sex differences since the motivation to move and the patterns of migration differed between men and women. We find one significant distinction in our analyses between men and women. Female migrants who stayed in Rotterdam had a higher mortality risk (than female natives of Rotterdam), but men did not (they had slightly lower mortality risk compared with male natives). This seems to suggest that for migrant women the selection effect was not as strong as it was for men; the difference between migrant women and those who stayed in their region of origin was not as great (cf. Greefs and Winter 2016 ). Perhaps because of their limited human capital or the lack of a social network, they might have ended up in trouble in the city, which could have prevented them from returning home or moving to another destination. One might think about domestic servants who gave birth to an illegitimate child or about women who ended up in prostitution. The literature on migration in this era suggests indeed that female migrants were disproportionally engaged in out-of-wedlock fertility and prostitution (Fuchs and Moch 1990 ; Moch 2003 ). However, in the case of Rotterdam such claims would require further investigation.

The fact that we did not find a salmon bias effect in our data does not mean that this result can be automatically extrapolated to other populations in other times and regions. As always, the historical context in which human behavior is being shaped has to be taken into account. Certain migrant populations might be more inclined to move to their home region once they fall seriously ill, and this could in fact lead to a real salmon bias effect. It is therefore crucial to replicate formal tests of the salmon bias hypothesis for other populations, and to study migration and mortality patterns against the background of the societies migrants came from as well as the societies they moved to. This means that future research would do well to take the culture, religion, traditions, and family systems of the migrant populations under study into account.

The results of this study also have an important implication for contemporary health policy. The healthy migrant effect suggests that all migrant groups fare better than the native population, but this is only true for those migrants who are most mobile. The effect, therefore, likely underestimates the health problems among migrants who live in a receiving society for a longer period of time. The fact that stayers fare much worse than movers also suggests that migrants pay a health price for adaptation. In light of our findings, future studies on migrant health should distinguish between stayers and leavers and, within those groups, between men and women.

Abraído-Lanza, A. F., Dohrenwend, B., Ng-Mak, D., & Turner, J. (1999). The Latino mortality paradox: A test of the “salmon bias” and healthy migrant hypotheses. American Journal of Public Health, 89 (10), 1543–1548.

Article   Google Scholar  

Abraído-Lanza, A. F., Chao, M. T., & Flórez, K. R. (2005). Do healthy behaviors decline with greater acculturation? Implications for the Latino mortality paradox. Social Science and Medicine, 61 (6), 1243–1255.

Alter, G., & Oris, M. (2005). Childhood conditions, migration and mortality. Social Biology, 52 (3–4), 178–191.

Google Scholar  

Bouman, P., & Bouman, W. (1955). De groei van de grote werkstad: een studie over de bevolking van Rotterdam . Assen: Van Gorcum.

Bras, H. (2003). Maids to the city: Migration patterns of female domestic servants from the province of Zeeland, the Netherlands (1850–1950). The History of the Family, 8 , 217–246.

Cleves, M., Gould, W., Gutierrez, R., & Marchenko, Y. (2008). An introduction to survival analysis using Stata (second ed.). College Station, TX: Stata Press.

Deboosere, P., & Gadeyne, S. (2005). Adult migrant mortality advantage in Belgium: Evidence using census and register data. Population (English Edition), 60 (5–6), 655–698.

Fuchs, R., & Moch, L. (1990). Pregnant, single, and far from home: Migrant women in nineteenth-century Paris. The American Historical Review, 95 (4), 1007–1031.

Greefs, H., & Winter, A. (2016). Alone and far from home: Gender and migration trajectories of single foreign newcomers to Antwerp, 1850-1880. Journal of Urban History, 42 (1), 61–80.

Greenwood, M. J. (2008). Family and Sex-specific U.S. Immigration from Europe, 1870–1910: A panel data study of rates and composition. Explorations in Economic History, 45 (4), 356–382.

Khlat, M., & Courbage, Y. (1996). Mortality and causes of death of Moroccans in France, 1979–1991. Population: An English Selection, 8 , 59–94.

Khlat, M., & Darmon, N. (2003). Is there a Mediterranean migrants mortality paradox in Europe? International Journal of Epidemiology, 32 (6), 1115–1118.

Lariscy, J., Hummer, R., & Hayward, M. (2015). Hispanic older adult mortality in the United States: New estimates and an assessment of factors shaping the Hispanic paradox. Demography, 52 (1), 1–14.

Lu, Y., & Qin, L. (2014). Healthy migrant and salmon bias hypotheses: A study of health and internal migration in China. Social Science and Medicine, 102 , 41–18.

Mandemakers, K. (2000). Historical sample of the Netherlands. In P. Kelly Hall, R. McCaa, & G. Thorvaldsen (Eds.), Handbook of international historical microdata for population research (pp. 149–177). Minneapolis: Minnesota Population Center.

Markides, K. S., & Coreil, J. (1986). The health of Southwestern Hispanics: An epidemiologic paradox. Public Health Reports, 101 , 253–265.

Moch, L. (2003). Moving Europeans. In Migration in Western Europe since 1650 . Bloomington and Indianapolis: Indiana University Press.

Mourits, R. (2017). Later-life mortality and longevity in late-18th and 19th-century cohorts. Where are we now, and where are we heading? Historical Life Course Studies, 4 , 1–19.

Oris, M., & Alter, G. (2001). Paths to the city and roads to death. Mortality and Migration in East Belgium during the Industrial Revolution. Journal of Belgian History, 3-4 , 453–495.

Puschmann, P. (2015). Social inclusion and exclusion of urban in-migrants in northwestern European port cities: Antwerp, Rotterdam and Stockholm, ca 1850–1930. KU Leuven: Upublished PhD thesis.

Puschmann, P., Van den Driessche, N., Grönberg, P., Van de Putte, B., & Matthijs, K. (2015). From outsiders to insiders? Partner choice and marriage among internal migrants in Antwerp, Rotterdam and Stockholm, 1850-1930. Historical Social Research / Historische Sozialforschung, 40 (2), 319–358.

Puschmann, P., Donrovich, R., Grönberg, P., Dekeyser, G., & Matthijs, K. (2016). Later-life mortality differences in Antwerp, Rotterdam and Stockholm, 1850-1930. Historical Social Research / Historische Sozialforschung, 41 (4), 257–290.

Razum, O. (2006). Commentary: Of salmon and time travellers — musing on the mystery of migrant mortality. International Journal of Epidemiology, 35 (4), 919–921.

Redstone Akresh, I., & Frank, R. (2008). Health selection among new immigrants. American Journal of Public Health, 98 (11), 2058–2064.

Turra, C., & Elo, I. (2008). The impact of salmon bias on the Hispanic mortality advantage: New evidence from social security data. Population Research and Policy Review, 27 , 515–530.

Van de Laar, P. (2000). Stad van formaat: Geschiedenis van Rotterdam in de 19de en 20ste eeuw . Zwolle: Waanders.

Van der Harst, G. (2006). Brabanders en Zeeuwen in het Rotterdamse straatbeeld – De verhuisgeschiedenis van de eerste generatie migranten, 1870-1920. In P. Van de Laar, L. Lucassen, & K. Mandemakers (Eds.), Naar Rotterdam – Immigratie en levensloop in Rotterdam vanaf het einde van de negentiende eeuw (pp. 39–60). Amsterdam: Aksant.

Van Leeuwen, M. and I. Maas. 2011. HISCLASS. A Historical International Social Class Scheme . Leuven: Leuven University Press.

Van Leeuwen, M., I. Maas and A. Miles. 2002. HISCO. Historical International Standard Classification of Occupations . Leuven: Leuven University Press.

Wallace, M., & Kulu, H. (2014). Migration and health in England and Scotland: A study of migrant selectivity and salmon bias. Population, Space and Place, 20 (8), 694–708.

Download references

Acknowledgments

We would like to thank George Alter (ICPSR, University of Michigan), Romola Davenport (University of Cambridge), Angélique Janssens (Radboud University; Maastricht University), Jan Kok (Radboud University Nijmegen), Kees Mandemakers (International Institute of Social History; Erasmus University Rotterdam), Alice Reid (University of Cambridge), Richard Smith (University of Cambridge), and Jan Van Bavel (KU Leuven) for their useful suggestions and encouragements. We are grateful to Research Foundation Flanders (FWO) for the financial support we received which enabled us to conduct this research.

Author information

Authors and affiliations.

Radboud Group for Historical Demography and Family History, Department of History, Radboud University, Erasmusplein 1, 6525, HT, Nijmegen, the Netherlands

Paul Puschmann

Family and Population Studies, Centre for Sociological Research, KU Leuven, Parkstraat 45, 3000, Leuven, Belgium

Paul Puschmann, Robyn Donrovich & Koen Matthijs

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Paul Puschmann .

Electronic supplementary material

(PDF 177 kb)

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Reprints and permissions

About this article

Puschmann, P., Donrovich, R. & Matthijs, K. Salmon Bias or Red Herring?. Hum Nat 28 , 481–499 (2017). https://doi.org/10.1007/s12110-017-9303-1

Download citation

Published : 17 October 2017

Issue Date : December 2017

DOI : https://doi.org/10.1007/s12110-017-9303-1

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

• Healthy migrant effect
• Salmon bias
• The Netherlands
• Find a journal
• Publish with us
• Track your research

• Search Menu
• Advance Articles
• Editor's Choice
• Supplements
• E-Collections
• Virtual Roundtables
• Author Videos
• Author Guidelines
• Submission Site
• Open Access Options
• About The European Journal of Public Health
• About the European Public Health Association
• Editorial Board
• Advertising and Corporate Services
• Journals Career Network
• Self-Archiving Policy
• Terms and Conditions
• Explore Publishing with EJPH
• Journals on Oxford Academic
• Books on Oxford Academic

Article Contents

Introduction, conclusions, supplementary data, ethics approval, data availability.

• < Previous

Investigating the salmon bias effect among international immigrants in Sweden: a register-based open cohort study

• Article contents
• Figures & tables
• Supplementary Data

Andrea Dunlavy, Agneta Cederström, Srinivasa Vittal Katikireddi, Mikael Rostila, Sol P Juárez, Investigating the salmon bias effect among international immigrants in Sweden: a register-based open cohort study, European Journal of Public Health , Volume 32, Issue 2, April 2022, Pages 226–232, https://doi.org/10.1093/eurpub/ckab222

• Permissions Icon Permissions

Studies of migration and health have hypothesized that immigrants may emigrate when they develop poor health (salmon bias effect), which may partially explain the mortality advantage observed among immigrants in high-income countries. We evaluated the salmon bias effect by comparing the health of immigrants in Sweden who emigrated with those who remained, while also exploring potential variation by macro-economic conditions, duration of residence and region of origin.

A longitudinal, open cohort study design was used to assess risk of emigration between 1992 and 2016 among all adult (18+ years) foreign-born persons who immigrated to Sweden between 1965 and 2012 ( n  = 1 765 459). The Charlson Comorbidity Index was used to measure health status, using information on hospitalizations from the Swedish National Patient Register. Poisson regression models were used to estimate incidence rate ratios (RRs) with 95% confidence intervals (CIs) for emigrating from Sweden.

Immigrants with low (RR = 0.83; 95% CI: 0.76–0.90) moderate (RR = 0.70; 95% CI: 0.62–0.80) and high (RR = 0.62; 95% CI: 0.48–0.82) levels of comorbidities had decreased risk of emigration relative to those with no comorbidities. There was no evidence of variation by health status in emigration during periods of economic recession or by duration of residence. Individuals with low to moderate levels of comorbidities from some regions of origin had an increased risk of emigration relative to those with no comorbidities.

The study results do not support the existence of a salmon bias effect as a universal phenomenon among international immigrants in Sweden.

Migration is a dynamic process that is influenced by a confluence of factors related to the country of origin, the migration experience and the settlement context; yet, the health of immigrants is also intertwined with migration, as it influences both ability and decisions to migrate. Previous research has shown that international immigrants have a mortality advantage relative to native-born majority populations in high-income countries, 1 commonly referred to as the healthy migrant hypothesis 2 or healthy migrant paradox. 3 Different explanations have been put forth for this relative advantage, including positive health selection of immigrants in the country of origin, as well as negative health selection in emigration, whereby it has been hypothesized that as immigrants develop poor health, they may be more likely to emigrate ( salmon bias effect) , 4 in order to receive care or support from family in more familiar or comfortable environments 5 , 6 or health care services, 7 leading to a positive health selection among those who remain.

However, empirical evidence supporting the salmon bias effect is limited, 8–10 especially in settlement contexts outside of the USA, where the salmon bias effect has been the most extensively studied among Hispanic immigrants, but for which evidence supporting this effect has also been mixed. 7 , 11–13 Many previous studies have indirectly tested the salmon bias effect by comparing mortality rates among groups of international immigrants with differing feasibility or probability of emigration. 14 , 15 Others have focussed on internal migrants in order to overcome censoring bias related to unrecorded e-migration, 9 , 16 , 17 whereby individuals who have actually emigrated continue to be registered as residents and included in administrative registers, which can lead to an underestimation of mortality estimates as a result of numerator–denominator mismatch. 18 , 19 Only a paucity of empirical research has evaluated the health status of international immigrants who emigrate. A Danish study 20 showed a lower risk of emigration among refugee and family reunification immigrants with greater disease severity relative to immigrants without disease. However, the generalization of these findings to other immigrant groups is unclear. For instance, immigrants coming from countries in conflict may be less likely to emigrate than labour migrants, thereby also being less likely to contribute to denominator bias, but may have fewer guarantees of receiving medical assistance in their home countries. 21 The likelihood of emigration may also be influenced by macro-economic conditions (e.g. recessionary periods) in the settlement context, as immigrants, especially newcomers, are more likely to face disadvantages in the labour market compared with native-born majority populations. 22 Depending on the feasibility of international mobility as well as education and professional and language skills, some immigrants may have greater employment opportunities elsewhere; as such, recent immigrants and immigrants with free movement rights [e.g. European Union (EU) citizens] may be more likely to emigrate than more established immigrants.

In this study, we aim to evaluate the salmon bias effect among international immigrants in Sweden by assessing if there are systematic differences in health among immigrants who have emigrated compared with those who remain in the country. Effect modification by macro-economic periods of recession, duration of residence, and region of origin will be considered, as these factors may influence associations between health status and emigration.

Study population and study design

Administrative data from multiple population-based registers which were linked via pseudonymized personal identification numbers were used to define the study population. An open cohort study design was utilized to allow for the inclusion of individuals who immigrated to Sweden during the study period. All foreign-born persons who immigrated to Sweden between 1965 and 2012 and were 18+ years old during the follow-up period (1992–2016) were included.

The study population was additionally categorized into nine regions of origin groups, using predefined country and region of origin categories constructed by Statistics Sweden (see table 1 ). Persons from Finland comprised a large proportion of the study population and were analyzed as a separate group. Additional groups included: other Nordic countries (excluding Finland), other EU-28 countries (excluding all Nordic countries but including the UK and Northern Ireland, who were members of the EU during the study follow-up), Eastern Europe and Russia (including former Yugoslavia), North America and Oceania , the Middle East (including Turkey), Asia , Horn of Africa , Rest of Africa and South America . Oceania was comprised largely of individuals from Australia, and as such was grouped with North America due to the similar economic development contexts of these two groups.

Descriptive characteristics of the study population

py, person-years.

Additional covariates, including sex (men and women), age (18–25; 26–35; 36–45; 46–55; 56–65; 66–74; and 75+ years), and duration of residence (under 5 years; 5–10; 10–15; 15–20; and more than 20 years) were also assessed. The follow-up period was sub-divided into five period factors (1992–96; 1997–2002; 2003–07; 2008–11; and 2012–16) to account for macro-economic conditions in Sweden, where periods of economic recession (1992–96; 2008–11), recovery (1997–2002; 2012–16) and stability (2003–07) were considered.

The outcome was defined as the first emigration from Sweden in the follow-up period 1992–2016. Emigration from Sweden was defined in two ways: (i) as the first recorded date of emigration from the Register of the Total Population 23 and (ii) proxy measured using information on sources of income (job-earnings or social benefits) from the Longitudinal Integration Database for Health and Labour Market Studies register, 24 whereby persons who did not have registered sources of income for 2 consecutive years (within each sub-divided follow-up period) were considered as having emigrated at the midpoint of the period factor. Previous studies from the Swedish context have used similar register-based information on income 18 , 25–27 to account for unrecorded emigration, as a consistent lack of income indicates that an individual is not resident in the country. Although a priori health differences between immigrants with recorded and unrecorded emigration are not expected, utilization of the proxy measure allows for an assessment of unrecorded emigration, which will decrease the likelihood of denominator bias. Individuals who emigrated before the start of follow-up were excluded.

Health status

The main exposure of interest was the Charlson Comorbidity Index (CCI), 28 a categorization measure of comorbidities that is calculated using International Classification of Diseases diagnosis codes. Annual information on reason for inpatient care (hospitalizations) from the National Patient Register 29 during the 2 years before the start of each period factor was used to create comorbidity scores based on the number and type of hospitalizations for multiple categories of disease (see Supplementary table S1 ). Each disease category was weighted based on severity (1–6), and the sum of all weights was used as the overall score for each individual within each period factor. The CCI scores were categorized and ranged from 0 to 3+. Scores of zero indicated no comorbidity (no hospitalizations), and scores of 3+ indicated the highest disease severity.

Statistical analyses

Poisson regression models were used to derive incidence rate ratios (RRs) with 95% confidence intervals (CIs), with age, calendar time (period factors), and duration of residence used as the three relevant time-scales. The RR of emigration was assessed using both recorded and unrecorded emigrations. Sensitivity analyses which included only recorded emigrations showed comparable findings, with the exception of emigration during the period 2008–11, during which a slightly lower risk of emigration was observed when only recorded emigrations were considered (see Supplementary figure S2 ). All models were adjusted for sex, age, period factors, duration of residence and region of origin, with robust standard errors. We formally tested for interactions between the CCI and (i) macro-economic period factors, (ii) duration of residence and (iii) region of origin using likelihood-ratio tests between models with and without multiplicative interaction terms. All analyses were conducted in R version 4.0.3 (R Core Team) with the use of Lexis splits from the Epi package 2.42 (Bendix Carsten, Martyn Plummer).

Table 1 describes the key study variables. Emigration rates were higher among men [22.3/1000 person-years (py)] than women (16.4/1000 py) and varied by region of origin. The lowest rates of emigration were observed among those from Eastern Europe and Russia (9.9/1000 py) and the Middle East (10.4/1000 py), whereas the highest rates were seen among those from other Nordic countries (48.0/1000 py) and North America and Oceania (48.1/1000 py). Emigration rates were stable across all period factors, ranging from 17.0/1000 py during the period 1997–2002, and 20.3/1000 py during the period 2008–11. Rates of emigration were lowest among immigrants who had resided in Sweden for 20+ years (5.1/1000 py) and highest among those with 5 years or less of residence (44.2/1000 py).

Figure 1 shows the association between emigration and health status as measured by the CCI during the entire follow-up period. A gradient in the relative risk of emigration was observed whereby emigration risk was significantly lower among immigrants who suffered from comorbidities relative to those who did not. Emigration risk was higher among men (RR = 1.32, 95% CI: 1.31–1.33) relative to women, and was lower for immigrants over age 35 relative to the youngest immigrants in the study population. Supplementary table S2 displays the data from figure 1 in tabular format.

Incidence RRs with 95% CIs for emigration among immigrants to Sweden by CCI score. Adjusted for sex, age, macroeconomic period factors, duration of residence and region of origin

Increased risks of emigration were observed during all macro-economic period factors after 2002, with the largest risk of emigration observed during the period 2008–11 (RR = 1.76, 95% CI 1.74–1.78). Although there were period effects in the risk of emigration, with emigration most likely to occur during the years surrounding the 2008 financial crisis, the interaction analysis indicated that the association between health status and emigration was not modified by macroeconomic period ( P  = 0.63).

A gradient in the relative risk of emigration was also seen by duration of residence, whereby those with longer duration of residence (over 5 years) had a lower risk of emigration than those with established residence in Sweden for 5 years or less. The interaction analysis demonstrated that duration of residence did not modify the association between health status and emigration ( P  = 0.56).

Relative to immigrants from Finland, most region of origin groups showed a lower risk of emigration, except those from other Nordic countries and those from North America and Oceania. The interaction analysis indicated that the association between health status and emigration varied by region of origin ( P  < 0.001). Figure 2 shows the risk of emigration among individuals within each region of origin category, using those with no disease comorbidity (CCI score = 0) from each region of origin as the reference. In most region of origin groups, those with higher disease severity demonstrated an equal or lower risk of emigration than those with no comorbidities. Exceptions to this pattern included immigrants with low disease comorbidity (CCI score = 1) from the Middle East (RR = 1.24; 95% CI: 1.00–1.54), the Horn of Africa (RR = 1.76, 95% CI: 1.09–2.86) and the Rest of Africa (RR = 1.82; 95% CI : 1.19–2.77) as well as immigrants with moderate disease comorbidity (CCI score = 2) from Eastern Europe and Russia (RR = 1.39; 95% CI : 1.01–1.92), all of whom demonstrated an increased risk of emigration relative to those from the same region of origin with no comorbidity. Risk of emigration could not be assessed among immigrants with the highest disease comorbidity (CCI score = 3+) from the Horn of Africa due to insufficient data for analysis among this group. Supplementary table S3 shows the data from figure 2 in tabular form.

Incidence RRs with 95% CIs for emigration by CCI scores and region of origin. Adjusted for sex, age, macroeconomic period factors and duration of residence

This study demonstrated that immigrants in Sweden with low, moderate and high disease comorbidity generally had lower risks of emigration relative to immigrants without comorbidities, which challenges the salmon bias effect. These results were also consistent when considering only recorded emigrations. Furthermore, while periods of economic crisis and length of residence were shown to impact the likelihood of emigration in immigrant populations in Sweden, there was no evidence of effect modification by health status in emigration during times of economic recession or by duration of residence. When interactions between health status and region of origin were investigated, in the majority of region of origin groups individuals with comorbidities showed an equal or lower risk of emigration relative to those without disease comorbidity, although there were variations.

Our overall results were largely in line with those of a Danish study 20 which found a lower risk of emigration among refugee and family reunification immigrants with poorer health status. Our study also used proxy measures to account for unrecorded emigration to minimize the risk of denominator bias. The comparability of our results which included both recorded and unrecorded emigration relative to those which only included recorded emigrations suggested that such bias was not present in our findings, and that there was no systematic variation in health status when comparing immigrants with recorded vs. unrecorded emigration. Taken together, our findings provide additional evidence to suggest that the salmon bias effect is not a key driver of the mortality health advantage observed among the majority of immigrant groups in Sweden. 27 , 30

However, immigrants from Finland are a group which has been shown to deviate from this pattern of advantage, with several studies rather demonstrating elevated mortality risks 27 , 31 , 32 as well as severe morbidities, including increased risk of hospitalizations for alcohol-related disorders 33 and myocardial infarctions. 34 As such, the lower risk of emigration found among persons from Finland with comorbidities (i.e. CCI score of 1–3+) relative to those with no comorbidities, could to some extent partially explain the previously observed elevated mortality risks in this group. Furthermore, the overall finding that immigrants in general with more comorbidities were less likely to emigrate than those without comorbidities, irrespective of their duration of residence, suggests that any indication of worse health with increasing duration of residence (a phenomenon commonly referred to as unhealthy assimilation ) 35 was likely to be a true health effect, and not an artefact due to unhealthy selective emigration.

Interaction analysis of the risk of emigration by health status and region of origin did not show a uniform pattern of findings. Yet, there was an indication pointing towards an association between the level of emigration rates and health status: the groups of immigrants showing the strongest evidence against the salmon bias effect were generally those with higher emigration rates, the exception being immigrants from Finland, with relatively low emigration rates (14%). Health status did not appear to influence risk of emigration in immigrants from South America. However, there was some evidence of a salmon bias effect among persons from the Horn of Africa, the Rest of Africa, the Middle East and Eastern Europe and Russia, whereby those with low to moderate disease comorbidity showed an elevated, rather than a decreased or similar, risk of emigration relative to those with no disease. These groups also had lower rates of emigration in general. However, this effect was not demonstrated amongst those with the highest disease severity (CCI = 3+), and as such evidence supporting the salmon bias effect remains limited. Furthermore, data on the country of emigration was not available in this study, and it remains unknown if individuals who emigrated returned to their countries of origin or moved to a third country. There is existing evidence showing patterns of emigration from Sweden to other countries in Europe, where access to universal health care is often guaranteed, among immigrants who also have EU citizenship or freedom of movement rights. 36–38 As such, emigration may be more likely to occur among some individuals with low to moderate ill-health, for whom migration is feasible and where necessary or desired care in the country of emigration can be guaranteed. Other research has postulated that persons who face economic or social adversities in settlement and subsequently might also be in poorer health may likewise emigrate to seek out alternative employment opportunities. 7 Further research is needed to explore these findings.

Strengths and limitations

This study utilized population-based register data that enabled us to longitudinally assess the risk of emigration in a large study population of international immigrants in Sweden. We were able to use these data to account for unrecorded emigrations, a key limitation of previous studies. We were also able to account for additional factors that can influence decisions to emigrate, including periods of economic crisis and duration of residence. Despite these strengths, our study was also tempered by some limitations. First, we were not able to specifically assess immigrants’ reason for migration to Sweden, but rather assessed region of origin, which has been widely used as a proxy measure of reason for migration in previous research, but also has several limitations. 39 However, a previous Swedish study 40 on risk of adverse birth outcomes among immigrant mothers suggested that administrative information on reason for migration (refugee vs. non-refugee) was not relevant due to its lack of specificity, as family reunification immigrants from countries in conflict may often be categorized as non-refugee immigrants. Relatedly, the generalizability of our findings may be limited to countries which provide universal access to health care services to all legal residents, such as Sweden. Immigrants residing in country contexts that do not provide such access may be more likely to emigrate if they require medical care that could be provided in the country of origin or a third country.

Decisions to migrate are complex and influenced by a number of social, economic and health-related factors. This study has provided evidence which challenges the salmon bias effect, as our findings did not universally support the existence of unhealthy selective emigration among international immigrants in Sweden. Future research is needed to investigate alternative explanations for the mortality advantage often observed among immigrants, including consideration of both protective and risk factors related to the origin and settlement contexts, and should continue to explore potential drivers behind health selection effects in emigration.

Supplementary data are available at EURPUB online.

This study was approved by the Regional Ethical Review Board of Stockholm (decision no. 2017/716-31/5).

The register data underlying this article cannot be shared publicly as it contains sensitive personal information. Access to the register data can be applied for by contacting the relevant Swedish public authorities, including Statistics Sweden and the Swedish National Board for Health and Welfare.

This work was supported by The Swedish Research Council for Health, Working Life and Welfare (Forte) [grant number 2016-071289]. SPJ acknowledges funding from The Swedish Research Council for Health, Working Life and Welfare (Forte) [grant number 2021-00271] and SPJ and AD from the Swedish Research Council (VR) [grant number 2018-01825]. SVK acknowledges funding from an NRS Senior Clinical Fellowship [SCAF/15/02], the UK Medical Research Council [MC_UU_00022/2] and Scottish Government Chief Scientist Office [SPHSU17].

Conflict of interest : None declared.

Key points:

Immigrants in Sweden with low, moderate and high levels of comorbidities had a lower risk of emigration than immigrants with no comorbidities, as measured by the Charlson Comorbidity Index.

Results were comparable when only recorded emigrations were considered.

Periods of economic recession and length of duration of residence did not modify the association between health status and emigration.

There was some evidence of modification in the association between health status and emigration in some region of origin groups, whereby persons with low to moderate levels of comorbidities showed an increased risk of emigration relative to those from the same region of origin without comorbidities.

Aldridge RW , Nellums LB , Bartlett S , et al.  Global patterns of mortality in international migrants: a systematic review and meta-analysis . Lancet 2018 ; 392 : 2553 – 66 .

Google Scholar

Urquia ML , Gagnon AJ. Glossary: migration and health . J Epidemiol Community Health 2011 ; 65 : 467 – 72 .

Palloni A , Morenoff JD. Interpreting the paradoxical in the “Hispanic paradox”: demographic and epidemiological approaches. In: Weinstein M , Hermalin A , Sotot M , editors. Population Health and Aging . New York : New York Academy of Sciences , 2001 : 140 – 74 .

Google Preview

Pablos-Méndez A. Mortality among Hispanics . JAMA 1994 ; 271 : 1237 .

Abraído-Lanza AF , Chao MT , Flórez KR. Do healthy behaviors decline with greater acculturation? Implications for the Latino mortality paradox . Soc Sci Med 2005 ; 61 : 1243 – 55 .

Razum O , Twardella D. Time travel with Oliver Twist–towards an explanation foa a paradoxically low mortality among recent immigrants . Trop Med Int Health 2002 ; 7 : 4 – 10 .

Diaz CJ , Koning SM , Martinez-Donate AP. Moving beyond salmon bias: Mexican return migration and health selection . Demography 2016 ; 53 : 2005 – 30 .

Wallace M , Kulu H. Can the salmon bias effect explain the migrant mortality advantage in England and Wales? Popul Space Place 2018 ; 24 : e2146 .

Lu Y , Qin L. Healthy migrant and salmon bias hypotheses: a study of health and internal migration in China . Soc Sci Med 2014 ; 102 : 41 – 8 .

Bhopal RS , Gruer L , Cezard G , et al.  Mortality, ethnicity, and country of birth on a national scale, 2001–2013: a retrospective cohort (Scottish Health and Ethnicity Linkage Study) . PLoS Med 2018 ; 15 : e1002515 .

Arenas E , Goldman N , Pebley AR , Teruel G. Return migration to Mexico: does health matter? Demography 2015 ; 52 : 1853 – 68 .

Palloni A , Arias E. Paradox lost: explaining the Hispanic adult mortality advantage . Demography 2004 ; 41 : 385 – 415 .

Turra CM , Elo IT. The impact of salmon bias on the hispanic mortality advantage: new evidence from social security data . Popul Res Policy Rev 2008 ; 27 : 515 – 30 .

Abraído-Lanza AF , Dohrenwend BP , Ng-Mak DS , Turner JB. The Latino mortality paradox: a test of the “salmon bias” and healthy migrant hypotheses . Am J Public Health 1999 ; 89 : 1543 – 8 .

Hummer RA , Powers DA , Pullum SG , et al.  Paradox found (again): infant mortality among the Mexican-origin population in the United States . Demography 2007 ; 44 : 441 – 57 .

Andersson G , Drefahl S. Long-Distance Migration and mortality in Sweden: testing the salmon bias and healthy migrant hypotheses . Popul Space Place 2017 ; 23 : e2032 .

Puschmann P , Donrovich R , Matthijs K. Salmon bias or red herring? Hum Nat 2017 ; 28 : 481 – 99 .

Monti A , Drefahl S , Mussino E , Härkönen J. Over-coverage in population registers leads to bias in demographic estimates . Popul Stud (Camb) 2020 ; 74 : 451 – 69 .

Wallace M , Kulu H. Low immigrant mortality in England and Wales: a data artefact? Soc Sci Med 2014 ; 120 : 100 – 9 .

Norredam M , Hansen OH , Petersen JH , et al.  Remigration of migrants with severe disease: myth or reality?—a register-based cohort study . Eur J Public Health 2015 ; 25 : 84 – 9 .

Davies AA , Borland RM , Blake C , West HE. The dynamics of health and return migration . PLoS Med 2011 ; 8 : e1001046 .

Johansson B , Vingård E. Migration, arbetsmiljö, och hälsa . Stockholm: Arbetsmiljöverket, 2012 .

Ludvigsson JF , Almqvist C , Bonamy AK , et al.  Registers of the Swedish total population and their use in medical research . Eur J Epidemiol 2016 ; 31 : 125 – 36 .

Ludvigsson JF , Svedberg P , Olén O , et al.  The longitudinal integrated database for health insurance and labour market studies (LISA) and its use in medical research . Eur J Epidemiol 2019 ; 34 : 423 – 37 .

Weitoft GR , Gullberg A , Hjern A , Rosén M. Mortality statistics in immigrant research: method for adjusting underestimation of mortality . Int J Epidemiol 1999 ; 28 : 756 – 63 .

Dunlavy AC , Juárez S , Rostila M. Employment status and risk of all-cause mortality among native- and foreign-origin persons in Sweden . Eur J Public Health 2018 ; 28 : 891 – 7 .

Juarez SP , Drefahl S , Dunlavy A , Rostila M. All-cause mortality, age at arrival, and duration of residence among adult migrants in Sweden: a population-based longitudinal study . SSM Popul Health 2018 ; 6 : 16 – 25 .

Charlson ME , Pompei P , Ales KL , MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation . J Chronic Dis 1987 ; 40 : 373 – 83 .

Ludvigsson JF , Andersson E , Ekbom A , et al.  External review and validation of the Swedish national inpatient register . BMC Public Health 2011 ; 11 : 450 .

Oksuzyan A , Mussino E , Drefahl S. Sex differences in mortality in migrants and the Swedish-born population: is there a double survival advantage for immigrant women? Int J Public Health 2019 ; 64 : 377 – 86 .

Rostila M , Fritzell J. Mortality differentials by immigrant groups in Sweden: the contribution of socioeconomic position . Am J Public Health 2014 ; 104 : 686 – 95 .

Honkaniemi H , Bacchus-Hertzman J , Fritzell J , Rostila M. Mortality by country of birth in the Nordic countries—a systematic review of the literature . BMC Public Health 2017 ; 17 : 511 .

Hjern A , Allebeck P. Alcohol-related disorders in first- and second-generation immigrants in Sweden: a national cohort study . Addiction 2004 ; 99 : 229 – 36 .

Andersson T , Pikkemaat M , Schiöler L , et al.  Country of birth and mortality risk in hypertension with and without diabetes: the Swedish primary care cardiovascular database . J Hypertens 2021 ; 39 : 1155 – 62 .

Antecol H , Bedard K. Unhealthy assimilation: why do immigrants converge to American health status levels? Demography 2006 ; 43 : 337 – 60 .

Haandrikman K , Hassanen S. Onward Migration of African Europeans: Comparing Attitudes to Migration Motives . Stockholm Research Reports in Demography, 2014 :15. Stockholm: Stockholm University.

Ahrens J , Kelly M , Van Liempt I. Free movement? The onward migration of EU citizens born in Somalia, Iran, and Nigeria . Popul Space Place 2016 ; 22 : 84 – 98 .

Kelly OM. The Transnational Trajectories of Iranians Leaving Sweden . Uppsala : Uppsala Universitet , 2013 .

Gibson-Helm M , Boyle J , Block A , Teede H. Use of country of birth as an indicator of refugee background in health datasets . BMC Med Res Methodol 2014 ; 14 : 27 .

Juárez S , Mussino E , Hjern A. Being a refugee or having a refugee status? Birthweight and gestational age outcomes among offspring of immigrant mothers in Sweden . Scand J Public Health 2019 ; 47 : 730 – 4 .

• comorbidity
• health status

Email alerts

Citing articles via.

• Contact EUPHA
• Recommend to your Library

Affiliations

• Online ISSN 1464-360X
• Print ISSN 1101-1262
• Copyright © 2024 European Public Health Association
• 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.

• Help & FAQ

The Latino mortality paradox: A test of the 'salmon bias' and healthy migrant hypotheses

• Global Public Health

Research output : Contribution to journal › Article › peer-review

Objectives: Relative to non-Latino Whites, Latinos have a worse socioeconomic profile but a lower mortality rate, a finding that presents an epidemiologic paradox. This study tested the salmon bias hypothesis that Latinos engage in return migration to their country of origin and are thereby rendered 'statistically immortal' and the alternative hypothesis that selection of healthier migrants to the United States accounts for the paradox. Methods. National Longitudinal Mortality Study data were used to examine mortality rates of the following groups for whom the salmon hypothesis is not feasible: Cubans, who face barriers against return migration; Puerto Ricans, whose deaths in Puerto Rico are recorded in US national statistics; and US-born individuals, who are not subject to either salmon or healthy migrant effects. Results. The sample included 301 718 non- Latino Whites and 17 375 Latino Whites 25 years or older. Cubans and Puerto Ricans had lower mortality than non-Latino Whites. Moreover, US-born Latinos had lower mortality than US-born non-Latino Whites. Conclusions. Neither the salmon nor the healthy migrant hypothesis explains the pattern of findings. Other factors must be operating to produce the lower mortality.

ASJC Scopus subject areas

• Public Health, Environmental and Occupational Health

Access to Document

• 10.2105/AJPH.89.10.1543

Other files and links

• Link to publication in Scopus
• Link to the citations in Scopus

Fingerprint

• Salmon Medicine & Life Sciences 100%
• mortality Social Sciences 64%
• migrant Social Sciences 56%
• European Continental Ancestry Group Medicine & Life Sciences 50%
• Mortality Medicine & Life Sciences 38%
• trend Social Sciences 37%
• Emigration and Immigration Medicine & Life Sciences 25%
• migration Social Sciences 17%

T1 - The Latino mortality paradox

T2 - A test of the 'salmon bias' and healthy migrant hypotheses

AU - Abraído-Lanza, Ana F.

AU - Dohrenwend, Bruce P.

AU - Ng-Mak, Daisy S.

AU - Turner, J. Blake

PY - 1999/10

Y1 - 1999/10

N2 - Objectives: Relative to non-Latino Whites, Latinos have a worse socioeconomic profile but a lower mortality rate, a finding that presents an epidemiologic paradox. This study tested the salmon bias hypothesis that Latinos engage in return migration to their country of origin and are thereby rendered 'statistically immortal' and the alternative hypothesis that selection of healthier migrants to the United States accounts for the paradox. Methods. National Longitudinal Mortality Study data were used to examine mortality rates of the following groups for whom the salmon hypothesis is not feasible: Cubans, who face barriers against return migration; Puerto Ricans, whose deaths in Puerto Rico are recorded in US national statistics; and US-born individuals, who are not subject to either salmon or healthy migrant effects. Results. The sample included 301 718 non- Latino Whites and 17 375 Latino Whites 25 years or older. Cubans and Puerto Ricans had lower mortality than non-Latino Whites. Moreover, US-born Latinos had lower mortality than US-born non-Latino Whites. Conclusions. Neither the salmon nor the healthy migrant hypothesis explains the pattern of findings. Other factors must be operating to produce the lower mortality.

AB - Objectives: Relative to non-Latino Whites, Latinos have a worse socioeconomic profile but a lower mortality rate, a finding that presents an epidemiologic paradox. This study tested the salmon bias hypothesis that Latinos engage in return migration to their country of origin and are thereby rendered 'statistically immortal' and the alternative hypothesis that selection of healthier migrants to the United States accounts for the paradox. Methods. National Longitudinal Mortality Study data were used to examine mortality rates of the following groups for whom the salmon hypothesis is not feasible: Cubans, who face barriers against return migration; Puerto Ricans, whose deaths in Puerto Rico are recorded in US national statistics; and US-born individuals, who are not subject to either salmon or healthy migrant effects. Results. The sample included 301 718 non- Latino Whites and 17 375 Latino Whites 25 years or older. Cubans and Puerto Ricans had lower mortality than non-Latino Whites. Moreover, US-born Latinos had lower mortality than US-born non-Latino Whites. Conclusions. Neither the salmon nor the healthy migrant hypothesis explains the pattern of findings. Other factors must be operating to produce the lower mortality.

UR - http://www.scopus.com/inward/record.url?scp=2942540654&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=2942540654&partnerID=8YFLogxK

U2 - 10.2105/AJPH.89.10.1543

DO - 10.2105/AJPH.89.10.1543

M3 - Article

C2 - 10511837

AN - SCOPUS:2942540654

SN - 0090-0036

JO - American journal of public health

JF - American journal of public health

Return Migration Selection and Its Impact on the Migrant Mortality Advantage: New Evidence Using French Pension Data

Corresponding author: [email protected]

• Standard View
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• Open the PDF for in another window
• Permissions
• Cite Icon Cite
• Search Site

Michel Guillot , Myriam Khlat , Romeo Gansey , Matthieu Solignac , Irma Elo; Return Migration Selection and Its Impact on the Migrant Mortality Advantage: New Evidence Using French Pension Data. Demography 1 October 2023; 60 (5): 1335–1357. doi: https://doi.org/10.1215/00703370-10938784

Download citation file:

• Reference Manager

The migrant mortality advantage (MMA) has been observed in many immigrant-receiving countries, but its underlying factors remain poorly understood. This article examines the role of return migration selection effects in explaining the MMA among males aged 65+ using a rich, unique dataset from France. This dataset contains information on native-born and foreign-born pensioners who are tracked worldwide until they die, providing a rare opportunity to assess return migration selection effects and their impact on the MMA. Results provide evidence of substantial and systematic negative return migration selection among foreign-born males in France. Old-age returns, in particular, appear particularly affected by such selection; however, they are not frequent enough to explain the MMA at ages 65+. By contrast, returns at younger ages are much more frequent, and the MMA at ages 65+ essentially disappears once these earlier returns are considered. This study extends the literature on negative selection at return and its impact on the MMA by providing evidence that such negative selection may operate not only at older ages but throughout the life course, with impacts on the MMA that are larger than previously suggested.

• Introduction

In the last decades, immigrant populations have increased in both absolute and relative size in many receiving countries. Between 1990 and 2020, the proportion of immigrants in the total population increased from 23.3% to 30.1% in Australia, 15.7% to 21.3% in Canada, 10.4% to 13.1% in France, 7.5% to 18.8% in Germany, 9.2% to 19.8% in Sweden, 6.4% to 13.8% in the United Kingdom, and 9.2% to 15.3% in the United States ( United Nations 2020 ). Given these trends, health and mortality patterns among immigrants are of increasing importance for receiving countries because they affect the demand for health care, health insurance schemes, pension systems, and mortality estimates ( Hendi and Ho 2021 ; Rechel et al. 2013 ; Zallman et al. 2013 ). Understanding health and mortality patterns among immigrants is also relevant for social policies aimed at reducing health disparities.

In the literature on immigrant mortality in high-income countries, the most systematic finding is that immigrants tend to exhibit lower mortality than the native population of their receiving country, a phenomenon termed the migrant mortality advantage (MMA). This pattern has been reported in receiving countries as diverse as Australia ( Kouris-Blazos 2002 ), Belgium ( Deboosere and Gadeyne 2005 ), Canada ( Bourbeau 2002 ), France ( Boulogne et al. 2012 ; Khlat and Courbage 1996 ), Germany ( Razum et al. 1998 ), the Netherlands ( Bos et al. 2004 ), Switzerland ( Tarnutzer et al. 2012 ), the United Kingdom ( Wallace and Kulu 2014 ), and the United States ( Ruiz et al. 2013 ; Zheng and Yu 2022 ). The MMA is often referred to as a mortality paradox because it occurs despite the lower average socioeconomic status (SES) among many immigrant groups relative to the native population of their receiving country.

Several hypotheses have been proposed to explain the MMA, although the relative contribution of each hypothesis in various contexts remains highly debated in the literature ( Agyemang 2019 ; Markides and Eschbach 2011 ). First, the hypothesis regarding in-migration selection effects (sometimes referred to as the healthy migrant effect ) posits that individuals who leave their country of origin may be positively selected on health. Second, cultural factors are hypothesized to produce more favorable health behaviors among immigrants in receiving countries than among natives and perhaps also benefit immigrants through dense immigrant social networks. Third, the MMA is hypothesized to result from data artifacts arising from the difficulty of correctly tracking immigrant populations in receiving countries and their corresponding deaths.

Finally, a hypothesis regarding return migration selection effects, sometimes referred to as the salmon bias hypothesis and most often discussed in reference to older ages, postulates that immigrants who are in poor health (or with other characteristics associated with higher mortality) may be more likely to return to their country of origin than their healthier counterparts ( Abraido-Lanza et al. 1999 ; Arenas et al. 2015 ; Di Napoli et al. 2021 ; Hummer et al. 2007 ; Khlat and Darmon 2003 ; Markides and Eschbach 2011 ; Pablos-Méndez 1994 ; Palloni and Arias 2004 ; Palloni and Morenoff 2001 ; Shai and Rosenwaike 1987 ; Turra and Elo 2008 ). Because of this “unhealthy remigration,” immigrants who remain in the receiving country may be selectively healthier, leading to reported mortality rates that are artificially low and poorly reflect the overall health conditions of immigrants.

In this study, we examine return migration selection effects and their impact on the MMA using a rich, unique dataset from France. This dataset contains information on pensioners from France's largest pension fund who receive lifelong annuities based on contributions accumulated while working in France. These annuities are not contingent on residing in a particular country; pensioners in this dataset are thus tracked worldwide until they die. These data include both native-born and foreign-born pensioners, providing a rare opportunity to assess return migration selection effects and their impact on the MMA. For substantive and methodological reasons (explained later in the article), our analyses focus on males aged 65+.

Return Migration Selection Bias, Censoring Bias, and the Salmon Bias

Return migration selection is often raised in the literature as a factor contributing to the migrant mortality advantage ( Namer and Razum 2018 ). There are several theoretical reasons to believe that return migrants may be subject to negative health selection, defined here as a negative correlation, at a given age, between health status and the likelihood of returning to the country of origin. First, studies have shown that immigrants often express a desire to die in their country of origin ( Razum et al. 2005 ; Tezcan 2019 ). Immigrants may respond to this desire in a way that does not involve health selection—for example, by setting age targets for their return, irrespective of health status. Deterioration of health, however, may trigger or accelerate the decision to return, thereby producing negative health selection. Other reasons for unhealthy remigration include seeking family support ( Pablos-Méndez 1994 ) or more affordable health care in the country of origin ( Arenas et al. 2015 ). Also, for working-age immigrants, a health deterioration can negatively affect employment and income prospects in the receiving country, lowering the financial incentives to remain ( Durand 2006 ).

Negative health selection at return may also operate indirectly if immigrants with low SES, which is associated with higher mortality, are more likely to return than immigrants with high SES, which is associated with lower mortality. Negative health selection could thus occur even when the decision to return is not directly related to health ( Razum et al. 1998 ). Such negative remigration selection is predicted by neoclassical economics theory, which views return migrants as failures or mistaken migrants—that is, migrants whose wages in the receiving country were lower than expected or who experienced higher than anticipated psychological costs of living abroad ( Constant and Massey 2002 ; Duleep 1994 ; Durand 2006 ; Wassink and Hagan 2018 ).

Negative health selection at return, whether operating directly via health or indirectly via SES, mechanically makes the reported mortality of immigrants in a receiving country lower than would be observed in the absence of return migration. This out-migration selection effect is equivalent to the classic problem of informative censoring in event-history analysis, a problem known to produce biased estimates of hazard rates. Because of their connection with health deterioration or, at the extreme, the desire to die in the country of origin, unhealthy remigrations are usually considered particularly salient at older ages, when poor health prevalence increases ( Palloni and Arias 2004 ; Turra and Elo 2008 ).

Despite theoretical support for negative return migration selection, poor health could also be theorized to lower the likelihood of returning, depending on individual and contextual circumstances. For example, the receiving country might have more accessible and higher quality health care than the country of origin, or family support might be more available in the receiving country ( Massey et al. 2015 ; Van Hook and Zhang 2011 ). As for indirect health selection via SES, new economics theory predicts that return migrants are positively selected on earnings; such theory views return migrants as target earners who return home once they have met their earnings targets ( Constant and Massey 2002 ; Wassink and Hagan 2018 ). This theory sees return migrants as successes and as subject to positive selection forces. Therefore, the existence of negative selection at return should not be presumed and ultimately depends on the net effect of various forces operating in opposite directions.

Return migration can also contribute to biased mortality rates among immigrants via censoring bias , which occurs when data sources in the receiving country fail to correctly remove out-migrants from the risk pool. Because immigrants who leave the receiving country are no longer covered by the receiving country's vital registration system, a failure to censor them makes them statistically immortal. In other words, censoring bias artificially inflates the population at risk, generating mortality estimates that are too low. Censoring bias is distinct from return migration selection effects: a failure to censor out-migrants biases migrant mortality rates even in the absence of return migration selection ( Palloni and Arias 2004 ; Wallace and Kulu 2018 ). Censoring bias is best categorized as a data quality explanation for the MMA that could theoretically be addressed by properly tracking and censoring out-migrants.

The salmon bias hypothesis, as originally formulated by Pablos-Méndez (1994) , covers both types of biases. This hypothesis postulates that out-migrations are both selective and not correctly recorded in data sources—two biases that combine to downwardly bias reported mortality rates among immigrants in receiving countries. Although most of the subsequent literature uses this definition of the salmon bias, some studies use a narrower definition that focuses more specifically on selection effects ( Turra and Elo 2008 ). In our study, the availability of information on out-migrations allows us to address censoring bias and thereby to focus specifically on return migration selection and its impact on the MMA.

Evidence for Return Migration Selection Bias

Studies seeking to identify return migration selection effects have produced inconsistent conclusions. Earlier studies relied on indirect evidence, comparing mortality levels among immigrant groups with different return migration patterns. Focusing on Hispanic individuals in the United States, Abraido-Lanza et al. (1999) found that Cuban immigrants had relatively low levels of reported mortality, evidence contradicting the salmon bias hypothesis. By contrast, Palloni and Arias (2004) found that Mexican immigrants in the United States experienced an unusually small slope of mortality at older ages, which was interpreted as evidence of the salmon bias. Other indirect evidence for return migration selection involved analyses of the health status of current versus return migrants using binational surveys. Palloni and Arias (2004) and Riosmena et al. (2013) compared health status among foreign-born Mexicans in the United States versus Mexicans in Mexico with prior U.S. migration experience. Both studies found that the latter group had worse health, evidence consistent with negative return migration selection.

A few studies have sought to capture health selection more directly by examining whether the health status of immigrants in the receiving country was correlated with subsequent out-migration rates. Studies from the United States ( Arenas et al. 2015 ) and the United Kingdom ( Wallace and Kulu 2018 ) found some evidence that immigrants in poor health had higher probabilities of return. By contrast, a study by Diaz et al. (2016) focusing on Mexican return migrants from the United States produced mixed support for negative health selection.

In this literature, a study by Turra and Elo (2008) using U.S. Social Security data stands out for its ability to observe immigrants' mortality after their return to their country of origin. Their results showed that foreign-born Hispanic individuals who returned abroad had higher mortality than those who stayed in the United States, consistent with negative return migration selection. They concluded, however, that the observed amount of negative selection was not sufficient to explain the Hispanic mortality advantage. (For a study focusing on internal migration in Sweden, with results consistent with salmon bias effects, see Andersson and Drefahl (2017) .)

The French pension data we use in our study shares many features with the U.S. Social Security data Turra and Elo (2008) used, including direct information on return migration and subsequent mortality. These shared features provide a unique opportunity to apply an approach similar to that of the Turra and Elo study to the European context. We focus on France, one of the major receiving countries in Europe. Relative to the Turra and Elo study, we have information from a more diverse set of countries of origin of immigrants, and we draw on this diversity to interpret our results. More broadly, we expand the literature on negative selection at return and its impact on the MMA by providing evidence that negative selection operates not only at older ages but throughout the life course, with impacts on the MMA that are larger than previously suggested.

The French Immigration Context

Although not considered a traditional country of immigration (e.g., the United States, Canada, or Australia), France stands out as the oldest European immigrant-receiving country (since the mid-nineteenth century) and the one that has received the largest cumulative number of immigrants ( Noiriel 1988 ; Weil 2005 ). Before 1945, migration flows to France involved primarily migrants from European countries (Italy, Spain, Portugal, Belgium, and Poland). After 1945, large waves of colonial migrants arrived (mostly from North Africa). Despite a labor migration decrease after 1973 ( Therborn 1987 ), immigration to France continued, mostly via family reunification and asylum ( Wihtol de Wenden 2012 ). In addition, the diversity of migrants continued to increase, with larger proportions of migrants from sub-Saharan Africa and Asia. The creation of the borderless Schengen area in 1995 and the European Union (EU) expansion in 2004–2007 also increased immigration from other EU countries.

In 2021, the proportion of immigrants in France's population was 12.8% (10.3% if we remove individuals born abroad with a French nationality at birth, who are not technically immigrants per France's official definition) ( INSEE 2022a ). The two most important regions of origin for France's immigrants were North Africa (Algeria, Morocco, and Tunisia) and southern Europe (Italy, Portugal, and Spain), representing 29.3% and 16.2% of the total immigrant population, respectively ( INSEE 2022a ). The proportion of immigrants from southern Europe is larger among the elderly population, the focus of this study. In 2021, these immigrants represented 33.7% of all immigrants aged 60 or older (vs. 29.3% for immigrants from North Africa), reflecting their preponderance during earlier waves of immigration ( INSEE 2022b ). The presence in France of elderly immigrants from mainly two different regions—southern Europe and North Africa—with different mortality environments and different border regimes (i.e., southern Europe is part of the borderless Schengen area, whereas North Africa is not) offers useful analytical contrasts for this study.

• Data and Methods

This study uses longitudinal data from the Caisse Nationale d'Assurance Vieillesse (CNAV, or National Old-Age Insurance Fund), France's most important pension fund. The CNAV manages pension payments for all individuals who have ever been employed in the private sector in France, regardless of the work contract length. The CNAV maintains a database that tracks pensioners from when they start receiving their pension until their death. This database has high coverage of the elderly male population in France. Approximately 95% of all male French pensioners receive at least a fraction of their pension from the CNAV ( Aubert 2012 ) because it is rare for individuals whose careers were mainly in other sectors (e.g., public sector, self-employed) not to have ever worked in the private sector at some point—for example, via short-term private contracts. Coverage is just as high among foreign-born male pensioners as their native-born counterparts ( Aubert 2012 ; INSEE 2012 ). However, coverage is lower for females, especially foreign-born females. Certain groups of foreign-born women have particularly low rates of labor force participation, including women born in North Africa and Turkey, whose labor force participation rates are 61% and 50%, respectively (vs. 88% for native-born women) ( Perrin-Haynes 2008 ). These low rates of labor force participation raise concerns about the lack of representativeness of foreign-born elderly women in the database. Our study therefore focuses on males.

We use data on a stratified random sample taken from the CNAV's exhaustive pensioners database and provided by the CNAV. This sample includes males who were alive at the baseline date of January 1, 2009, and were receiving at least part of their pension from the CNAV as primary beneficiaries. (We excluded secondary beneficiaries, who derive their entire pension based on their spouse's work history, because they may have never lived in France.) The sample was not restricted by residence; our pensioners may have resided in France or abroad at baseline. Given our focus on immigrants, the CNAV prepared a dataset that oversampled foreign-born individuals, providing us with an original sample of 194,083 foreign-born and 45,561 native-born male pensioners. The dataset contains information on date of birth and country of birth, as well as full residence and mortality follow-up for six calendar years (i.e., January 1, 2009–December 31, 2014). The residence follow-up includes information on the country of residence, updated quarterly based on the pensioner's reported address changes. The mortality follow-up includes death dates for pensioners who died during the follow-up period. The follow-up information also includes pension amounts the pensioners received each year. Because the dataset does not include information on nationality at birth, we define immigrants based on their country of birth.

We excluded all individuals younger than 65 at baseline (45,561 individuals) to address concerns about early retirement–related health selectivity. We also excluded 1,076 individuals (mostly German-born) who derived their pension not from having worked in France but from their prior association with the French military, perhaps without ever having resided in France. Our analysis sample consists of 193,007 individuals: 160,412 foreign-born and 32,595 native-born.

Because the data are administrative, our sample contains little missing information, although quarterly information about the place of residence was sometimes missing. There were 2,523 pensioners (1.31% of the analysis sample) with missing residence information in at least one quarter. Although the number of quarters for which data were missing during the observation window varied across pensioners, approximately 97.6% of all missing cases had three or fewer quarters with missing residence information. We imputed missing residence information according to systematic rules using the sequence of residence data, survival status, and place of death for those who died during the observation window. Appendix Table A1 ( online appendix ) summarizes our procedure for imputing missing residence information.

Adjustment of Deaths Occurring Abroad

For deaths occurring in France, death information is automatically integrated into the CNAV database via the civil registration system. For deaths occurring abroad, except for a few EU countries with data-sharing agreements (e.g., Belgium, Luxembourg, and Germany), no such integration exists. The recording of death information requires the provision of an official death certificate from the country where the death occurred. Because this integration is not automatic, some deaths occurring abroad may be missing. To reduce the likelihood that pension payments will continue after a pensioner's death, pensioners residing abroad are required to submit a “certificate of life” to the CNAV every year. This form, which must be certified by local authorities after the presentation of proper identification, proves that pensioners residing abroad are alive and allows them to continue receiving their pension. The CNAV assumes that pensioners who stop producing this certificate are dead and thus stops sending pension allowances.

Given these concerns about tracking deaths occurring abroad, a sole reliance on official death certificates is insufficient for our study. We thus take advantage of the CNAV's tracking procedure via certificates of life to adjust deaths occurring abroad. Specifically, we examine sequences of annual pension allowances and identify pensioners residing abroad who stopped receiving their pension without official information that they had died. We assume that individuals who stopped receiving their pension allowance at some point during the six-year follow-up period and did not receive it again by the end of the period were no longer alive. To avoid overcorrecting, and given the occurrence of one-year gaps in pension payments in the database, we take the conservative approach of imputing deaths only for individuals with at least two consecutive years without pension allowance. This approach required having at least two consecutive years of pension data to decide about death imputation; we therefore discarded the last two years of data and restricted the analysis window to January 1, 2009—December 31, 2012 (four years). Our imputation strategy adds 959 deaths to the recorded 12,917 deaths occurring abroad during the observation window. Details about the number of imputed deaths corresponding to each pension sequence are summarized in Table A2 ( online appendix ).

Simplifying Residence Information

Most commonly, pensioners did not change residence during the observation period or experienced only one international move. Few pensioners (389, or 0.2% of the sample) experienced two or more changes of residence. To include these individuals in the analysis, we simplified their migration histories by allowing no more than one change of residence during the follow-up period, using the last observed change of residence.

After performing these simplifications, we obtained eight possible configurations of individual migration and mortality histories. These combinations are represented in Figure 1 , with each line representing the lifeline of a pensioner of a given type between January 1, 2009, and January 1, 2013. Lines 1–4 represent pensioners residing in France at baseline. Among them, Lines 1 and 2 represent pensioners who remained in France during the follow-up period: Pensioner 1 was still alive on January 1, 2013, and Pensioner 2 died (in France) before January 1, 2013. Lines 3 and 4 represent pensioners who left France during the follow-up period: Pensioner 3 was still alive on January 1, 2013, and Pensioner 4 died (abroad) before January 1, 2013. Lines 5–8 represent the life lines of pensioners residing abroad at baseline, organized by whether they remained abroad (Lines 5 and 6) or returned to France (Lines 7 and 8) and by whether they remained alive (Lines 5 and 7) or died (Lines 6 and 8). Everyone in our sample falls into one of these eight categories.

In our data, out-migrations from France among the foreign-born almost always (i.e., in 94.7% of cases) correspond to out-migrations to the country of birth rather than to a third country. (See online appendix Table A3 for further details.) In the remainder of the article, we thus do not distinguish between foreign countries of destination and interpret out-migrations from France among the foreign-born as return migrations.

Our assessment of return migration selection and its impact on the migrant mortality advantage is primarily based on models that compare mortality for return migrants versus migrants who remained in France. Similar to Turra and Elo's (2008) study, this approach relies on the prediction that if negative health selection at return is occurring, return migrants will have elevated mortality compared with stayers. However, elevated post-return mortality may arise from factors other than health selection, such as the causal effect of return migration itself and arrival in a place with potentially worse health conditions. To better ascertain the effect of health selection while accounting for the relatively young ages at which many returns occur (as we demonstrate later), we use two analytic approaches.

In the first approach, we focus on foreign-born pensioners residing in France at baseline (Lines 1–4 in Figure 1 ). We follow them prospectively until the end of the follow-up period to determine whether out-migrating from France is associated with higher subsequent mortality relative to remaining in France, as predicted by negative return migration selection. We use a classic, semiparametric Cox proportional hazard model with mortality as the outcome and out-migration as a time-varying explanatory variable:

where µ x ( y ) is the risk of death at age x + y for those age x at baseline (January 1, 2009), µ 0 ( x + y ) is an unspecified baseline hazard, and Z ( y ) is a time-varying variable that switches from 0 to 1 when the pensioner moves abroad from France during the follow-up period. We apply this model to foreign-born pensioners, stratified by country of birth.

We then examine whether patterns of post-return mortality documented via Eq. (1) explain the migrant mortality advantage. We use two models to compare the mortality of foreign-born versus native-born pensioners during 2009–2012, also focusing on individuals residing in France at baseline (Lines 1–4 in Figure 1 ). In Model 1, we focus on the mortality of foreign-born versus native-born pensioners residing in France , using a Cox proportional hazard model in which foreign-born pensioners who out-migrated are censored (as they should be) at the time of their out-migration:

with X representing a vector of dummy variables to identify groups according to their country of birth. This first model allows us to evaluate the extent of the MMA among pensioners in France. Model 2 is similar to Model 1, with one important difference: foreign-born pensioners who out-migrated during the follow-up period are not censored. Instead, they are retained in the exposure pool through the end of the follow-up period, and their deaths occurring abroad are included in the estimation of mortality hazard ratios. This second model allows us to evaluate what the level of the MMA would have been had returning pensioners stayed in France while retaining their observed age at death. The comparison of the relative mortality of foreign-born pensioners in the second versus the first model provides our basis for evaluating the extent to which the MMA may be explained by return migration selection.

In the second approach, our analyses use information about the mortality of all foreign-born pensioners, including those already residing abroad at baseline. We first examine whether foreign-born pensioners residing abroad experienced higher mortality than those residing in France. We use a Cox proportional model that follows the same equation as in Approach 1 ( Eq. (1) ), this time applied to the entire sample (Lines 1–8). We then examine the MMA in France and the extent to which it is explained by return migration selection using Models 3 and 4, which are identical to Models 1 and 2 except that they are applied to the entire sample (Lines 1–8).

The advantage of the first approach is that it uses information on the timing of out-migration and relies on a relatively short time window—no more than four years—between out-migration and mortality. As a result, the potential confounding impact of post-remigration conditions in the country of origin can play only a limited role, taking us as close as possible to ascertaining actual selection effects. The second approach, by contrast, will be less directly interpretable in terms of selection effects because of the longer time window between return migration (which, for Lines 5–8, occurred before the start of the follow-up period) and mortality. Its strength, however, is that it considers the large share of foreign-born pensioners who left France at younger ages and may also have been negatively selected. Both approaches have complementary strengths and limitations; in combination, they provide a more comprehensive picture of return migration selection and its impact on the MMA.

Table 1 shows information about the subsample of individuals who resided in France at baseline (January 1, 2009): those illustrated with Lines 1–4 in Figure 1 . This subsample is used in Approach 1. The first column of Table 1 shows the counts of foreign-born pensioners at baseline by region/country of birth, as well as all foreign-born and native-born pensioners. Consistent with France's immigration history, the regions most represented among these pensioners are North Africa (Algeria, Morocco, and Tunisia) and southern Europe (Italy, Portugal, and Spain). For these two regions, we show details by individual country of birth. The remaining columns show counts of deaths and out-migrations among these pensioners during the follow-up period (2009–2012). Of the 83,658 foreign-born pensioners residing in France at baseline, 12,592 (15.1%) died in France, and 2,435 (2.9%) out-migrated during the follow-up period. Among those who out-migrated, 437 (17.9%) died between the time of the relocation abroad and the end of the follow-up period.

These data are used in the prospective model for Approach 1, with hazard ratios representing whether foreign-born pensioners who out-migrated had higher subsequent mortality than those who remained in France. Results presented in Table 2 show that return migration is associated with substantial excess mortality. For all countries of birth combined, the mortality hazard ratio is 2.406 ( p < .001). This excess mortality is highly consistent among migrant groups. The only foreign-born pensioners with a statistically nonsignificant hazard ratio are those born in the residual “other foreign countries” category, a small group of pensioners with very few (11) returnees during the follow-up period. All other groups display statistically significant hazard ratios ranging from 1.556 to 3.617, with no clear pattern by region/country of birth. Pensioners born in and returning to low-mortality regions/countries (e.g., Italy, Portugal, and Spain) do not have lower hazard ratios associated with return migration.

Is this excess mortality among returnees strong enough to explain the migrant mortality advantage? We answer this question by comparing mortality among foreign-born versus native-born pensioners in Models 1 and 2; results are shown in Table 3 . Model 1, which censors pensioners who out-migrate at the time of out-migration, confirms a consistent migrant mortality advantage at ages 65+ in France. With a hazard ratio of 0.927 ( p < .001), foreign-born pensioners residing in France have mortality risks at ages 65+ that are 7.3% lower, on average, than those of their native-born counterparts. This mortality advantage is particularly pronounced among pensioners from Morocco (hazard ratio = 0.845, p < .001), Tunisia (0.880, p < .001), and Portugal (0.883, p < .001). The MMA was not shared by all groups of foreign-born pensioners, however. Pensioners born in Italy, other countries in Europe, and other countries in Africa did not have statistically significant hazard ratios.

Model 2, which does not censor foreign-born pensioners who out-migrated during the follow-up period but instead keeps them in the exposure pool and takes their deaths abroad into account, seeks to measure the effect of return migration selection on the MMA. Results show that the hazard ratios are higher in Model 2 than in Model 1. This finding is expected because Model 2 includes the mortality of returnees, who had higher mortality than those who stayed. The effects are relatively minor, however. When we examine all foreign-born pensioners combined, the hazard ratio increases only slightly, from 0.927 to 0.946, and retains its significance. Effects vary by country of birth, with two countries (Algeria and Portugal) experiencing larger increases in the hazard ratio than others. Overall, though, the effects are modest. Among countries with a statistically significant MMA, only one (Spain) loses significance once mortality among returnees is considered. Regions/countries with no statistically significant hazard ratio in Model 1 remain the same in Model 2. The overall lesson of this comparison is that even though we detect strong excess mortality among foreign-born pensioners who resided abroad, this excess mortality appears to explain only a small portion of the overall MMA. This result is due to the relative rarity of return migration among these pensioners. In our sample, only 2.9% of those residing in France at baseline out-migrated during the follow-up period. Even though these returnees experienced substantial excess mortality, their demographic weight is too small to reverse the advantage experienced by the overwhelming majority of those who stayed.

Approach 1 excluded pensioners who had returned before 2009 (Lines 5–8 in Figure 1 ). Table 4 and Figure 2 show that this group of early returnees represents a large share of foreign-born pensioners, with almost half (47.8%) residing abroad in 2009. This reality illustrates the importance of return migration across the life course in a receiving country like France: by the time they reach retirement age, almost half of foreign-born males who had worked in France at some point had already left France and were receiving their French pension abroad. The CNAV data offer a rare opportunity to capture this combined stock of current and former immigrants. Table 4 and Figure 2 also show variations in the proportion of returnees by region/country of birth. The proportions tend to be higher among pensioners from neighboring EU countries, where return migration is easier. For example, 64.4% of Spain-born pensioners and 59.3% of Portugal-born pensioners resided abroad at baseline. At the other end of the spectrum, countries such as Morocco and Tunisia, for which return migration may be more costly, have smaller proportions of returnees (25.2% and 25.1%, respectively). An analysis of sequences of pension contributions in the CNAV data suggests that a large share of returns occur in midlife, at around ages 45–55, with a smaller peak around retirement ( Gansey et al. 2019 ).

Approach 2 accounts for this large stock of individuals already residing abroad at baseline. Table 5 compares the mortality risks of foreign-born pensioners residing abroad versus those residing in France. Results are consistent with the patterns found in Table 2 (for Approach 1): mortality rates are higher among foreign-born pensioners residing abroad than among those residing in France. Hazard ratios are systematically greater than 1, with statistical significance reached for all regions except the small “other foreign countries” residual category, all but one of the top six countries of origin (Tunisia), and all foreign-born pensioners combined (hazard ratio = 1.159, p < .001). Including early returnees does not modify our earlier conclusion that residing abroad versus in France is associated with higher mortality at ages 65+. The hazard ratios in Table 5 are not as high as those generated with Approach 1. The consistency of the pattern of excess mortality across foreign-born pensioners with countries of origin as diverse as Italy, Morocco, Portugal, and Spain is nonetheless striking. 1

As with Approach 1, we examine the extent to which the excess mortality associated with residing abroad may explain the MMA. Model 3 in Table 6 compares the mortality of foreign-born versus native-born pensioners residing in France. This model is similar to Model 1 in Table 3 but considers all deaths and exposures associated with a French residence, regardless of whether pensioners resided in France or abroad at baseline. Results are very similar to those from Model 1, which is expected because the main difference between the models arises from a small number of pensioners who resided abroad at baseline but returned to France during the follow-up period. The hazard ratio of 0.926 for all foreign-born pensioners combined in Model 3 reflects the same overall mortality advantage as in Model 1 (0.927). Hazard ratios for individual groups of pensioners by region/country of residence are virtually identical to those estimated in Model 1.

The main difference from Approach 1 appears in Model 4, which examines the relative mortality of foreign-born pensioners (vs. their native-born counterparts) regardless of their country of residence at baseline or during the follow-up period. Including the large stock of foreign-born pensioners already residing abroad at baseline provides a different picture. For all countries of birth combined, the hazard ratio rises from 0.926 to 0.998 and loses significance, indicating that the MMA disappears once we include foreign-born pensioners residing abroad in the comparison. We find this pattern of lost advantage among pensioners from several regions/countries of birth, including Algeria and Portugal. In addition, certain groups of foreign-born pensioners that did not have an MMA in Model 3 (i.e., those born in Italy and other European countries) exhibit statistically significant excess mortality once the mortality of all foreign-born pensioners is considered in Model 4. The overall lesson of comparing Model 4 with Model 3 is that the MMA disappears (or is sometimes reversed) once the mortality of foreign-born pensioners who lived abroad at baseline is considered. This finding results from the combined effect of large proportions of foreign-born pensioners living abroad and their higher mortality relative to those residing in France—a systematic pattern across a wide range of countries of origin. Although the excess mortality associated with foreign residence is larger in Approach 1, effects on the MMA are stronger in Approach 2.

During their employment in a receiving country, foreign-born workers accumulate rights to a pension they can receive upon reaching retirement age, regardless of where they reside. Some of these workers return to their country of origin while relatively young, after only a few years of employment in the receiving country. Some wait until or after retirement to return. Some never return, staying in the receiving country until the end of their life. Comparing the mortality of resident foreign-born versus native-born individuals in receiving countries ignores these returns, even though in a country like France, roughly half the initial stock of foreign-born male workers return to their country of origin by retirement age. Those still residing in France at older ages experience a consistent migrant mortality advantage relative to natives. This advantage, however, is difficult to interpret in a context where mortality rates are calculated based on only a portion of the initial stock of foreign-born workers, ignoring the experience of those who returned. If returns were selective, those remaining in France might represent the healthiest of the initial stock of immigrants.

In this study, we take advantage of a unique, longitudinal source of information that allows us to observe the mortality of males aged 65+ who were immigrant workers in France and returned to their country of origin before or after retirement. We find that foreign-born pensioners residing abroad experience higher mortality than their peers staying in France. This excess mortality is particularly strong among recent returnees.

We interpret the excess mortality of recent returnees, shown in Approach 1, as indicating a strong and pervasive process of negative return migration selection. For these individuals, return migration occurred after age 65, and this excess mortality was observed over a short time window following return migration. This short time window attenuates the possible confounding effect of exposures to conditions in the country of origin after the return. Moreover, the excess mortality associated with returning to the country of origin is observed across diverse countries of origin, including countries with mortality conditions and health care systems comparable to those in France (e.g., Italy and Spain). This finding further supports our interpretation that excess mortality results from negative selection rather than differences in mortality conditions or health care systems between France and the country of origin.

Conditions in the country of origin after return are likely more influential for the mortality of those who returned before retirement. Indeed, some of the pensioners examined in Approach 2 may have returned years before retiring, and thus their excess mortality observed at older ages is, a priori , less directly interpretable in terms of negative return selection. Nonetheless, like in the case of Approach 1, our results are strikingly consistent across diverse countries of origin. We find excess mortality even among immigrants returning to Italy, Portugal, and Spain—countries with mortality conditions that are not particularly different from those in France. In 2010–2014, male life expectancy at age 65 in Italy and Spain was 18.62 years and 18.64 years, respectively, versus 18.92 years in France ( Human Mortality Database 2021 ). These life expectancies translate to mortality ratios of 1.016 for Italy and 1.015 for Spain (vs. France), which are substantially lower than those documented in Table 5 for these two countries of birth. The mortality gap is slightly larger in Portugal, with a life expectancy at age 65 of 17.60 years. Here also, this difference in life expectancy translates to a much lower mortality ratio than that found in Table 5 : 1.075 versus 1.227 in Table 5 . The consistency of the pattern of excess mortality shown in Table 5 , even for countries of origin that do not exhibit important mortality differences from France, supports the conclusion that the excess mortality arises from negative return selection.

The higher mortality hazard ratios produced from Approach 1 relative to Approach 2 further support this interpretation of our results. In Approach 1, post-return exposure to conditions in the country of origin is shorter than in Approach 2, yet returnees experience higher relative mortality. If exposure to worse mortality conditions in the country of origin were the dominant explanation, we would observe lower hazard ratios in Approach 1 than in Approach 2. It is possible that negative return migration selection is just as high in Approach 1 as in Approach 2 but that the impact of negative selection in Approach 2 is attenuated by some positive consequences of return migration on health. Our data do not allow us to make this distinction, but if this were the case, then Approach 2 would underestimate the true scale of negative selection return, and our results would be conservative.

Our results extend the literature in several ways. First, we expand the findings of Turra and Elo (2008) via a similar approach applied to the European context. Like Turra and Elo, we find that return migrants experience higher post-remigration mortality than comparable migrants who stayed, with particularly high excess mortality among recent returnees (Approach 1). A key difference is that once we account for mortality among all return migrants regardless of age at return (Approach 2), the MMA essentially disappears in our study; by contrast, in the Turra and Elo study, the MMA remained unaffected. This difference arises primarily from the much more frequent return in the French context. In the U.S. Social Security data Turra and Elo used, 9.3% of foreign-born Hispanic males and 11.0% of non-Hispanic White males resided abroad as pensioners, compared with 47.8% of foreign-born males in the French CNAV data. Taking the excess mortality of return migrants into account is therefore much more consequential in the French context.

Overall, our results from Approach 2 emphasize the importance of negative return migration selection as a mechanism that may operate not only at older ages, as suggested in much of the salmon bias literature, but across the life course. Our results do support the operation of negative return selection at older ages, but they also suggest that negative return selection is likely to occur over the life course, with mortality impacts that remain visible years later. Although return migration selection across the life course was previously theorized and documented, our results show that the dominant selection process appears to be negative rather than positive. This finding is consistent with those of several studies focusing on health outcomes among working-age return migrants, with results indicating the presence of negative health selection ( Arenas et al. 2015 ; Lu and Qin 2014 ; Martinez-Cardoso and Geronimus 2021 ). It is also consistent with the hypothesis initially proposed by Razum et al. (1998) , who emphasized negative selection via SES; they postulated that immigrants who fail to cope well socially and economically in the receiving country might remigrate “even before becoming manifestly ill” ( Razum et al. 1998 :302).

Regardless of whether return migration is motivated by health or other factors, our study stresses the importance of considering returns occurring at relatively young ages when examining return migration selection and its impact on the MMA or the immigrant health advantage ( Arenas et al. 2015 ; Riosmena et al. 2017 ). The impact of return migration selection effects on the MMA may be larger than previously suggested, especially in contexts with a large volume of return migration, such as France.

Our results show that the migrant mortality advantage at ages 65+ is reduced or eliminated once return migration selection is considered. From a life course perspective, the migrant mortality advantage in receiving countries tends to be largest around age 45 and then gradually diminishes with age via a process of mortality convergence ( Guillot et al. 2018 ; Namer and Razum 2018 ). By the time immigrants reach retirement age, little advantage remains, as illustrated by the foreign-born versus native-born mortality hazard ratio of 0.926 shown in Table 6 . By contrast, mortality ratios observed around age 45 in several receiving countries, including France, are 0.6–0.7 ( Guillot et al. 2018 ). Our results for mortality at ages 65+ suggest that return migration selection slows the process of mortality convergence with age. Were returns not selective, the process of mortality convergence with age would likely occur faster than observed using reported mortality data.

Our study has limitations. First, we defined immigrants using country of birth information. This definition is problematic for pensioners born in Algeria because it does not distinguish between former European colonists in Algeria who returned to France when Algeria became independent in 1962 versus Algerian immigrants. In 2019, only 61.3% of the Algeria-born population residing in France had a foreign nationality at birth and were thus considered immigrants. This issue may explain why the migrant mortality advantage is not as large for Algeria as for the other two North African countries (Morocco and Tunisia). Using country of birth information as the sole criteria for defining immigrants is likely to have a negligible impact on the results for the other countries of birth, for which the proportion of true immigrants among the foreign-born population is 82% to 86% ( INSEE 2022a ).

Second, our data do not include information on returnees who never claimed their right to a pension. We do not have information about how frequently this occurs, but it is likely to be more common among returnees who spent short amounts of time in France, for whom pension amounts may be trivial. To address potential selection biases associated with pension-claiming, we conducted a sensitivity analysis in which we removed from the sample those returnees with fewer than five years of salary contributions in France. The results are virtually unchanged (see Tables A6 and A7, online appendix ), suggesting that such selection biases are likely to be small.

Third, our data do not include information on pensioners who never worked in the private sector. This omission is likely to have only a negligible impact on our results because of the very high coverage among males in the CNAV database, as discussed in the Data section. Our data also omit information on workers who spent their entire time in France without a regular work contract. However, undocumented immigrants in France can and often do work with official contracts because proof of employment and tax returns can be used as positive evidence for legalization. This path to legalization for undocumented immigrants in France also implies that very few immigrants who still reside in France at retirement age can be expected to have remained undocumented. Workers who spent their entire time in France without a work contract are likely to have resided in France for only short durations. As shown in our sensitivity analyses discussed earlier (Tables A6 and A7), excluding individuals with short durations of stay has virtually no effect on the results.

Fourth, deaths abroad may be underreported or reported late. Our study addressed this issue by considering information on pension payments, which are conditional on producing a certificate of life. However, mortality abroad remains likely to be underreported, especially at very old ages, at which observed slopes of age-specific mortality rates appear implausibly low even after our mortality adjustment (results not shown). This pattern implies that our estimates of excess mortality associated with residing abroad remain underestimated. Our results are thus likely conservative and underestimate the true level of negative return migration selection.

Fifth, the residence information we use to distinguish between pensioners residing in France and those living abroad is based on declared rather than actual residence. Some pensioners who declare French residence might spend most (if not all) of their time abroad, given that retaining an official residence in France brings some advantages in terms of pension amounts and health care coverage. This issue would affect our conclusion only if the pensioners who declared residing in France but actually resided abroad had lower mortality risks than those who actually resided in France. There is no reason to believe that such positive selection occurs. On the contrary, it is likely that negative return selection also operates for unofficial returns, in which case our results would be conservative, underestimating the amount of negative return selection.

Finally, our analytic approach does not allow us to rule out the possibility that at least some excess mortality occurring after returning to the country of origin may be caused by return migration rather than solely arising from negative selection. For a country of origin with worse health conditions and poorer health care, return migration may lead to excess mortality even without selection. Because our data do not include information on health or SES before remigration, we cannot evaluate the respective roles of selection and causal effects in explaining the excess mortality of return migrants. As discussed earlier, we nonetheless conclude that our results are best explained by selection because excess mortality is present across a wide range of countries of origin, including countries with mortality conditions similar to those of France. Moreover, excess mortality is particularly strong among recent migrants (Approach 1), for whom post–return migration conditions are less likely to play a substantial role. For a more direct ascertainment of selection versus causation in explaining the excess mortality of returnees, future research would benefit from using longitudinal data that would not only track the mortality of immigrants after they return to their country of birth but also provide sufficient information on their characteristics during the period preceding their return.

• Acknowledgments

Research reported in this manuscript was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) of the National Institutes of Health (NIH) under award number R01HD079475. We would like to thank the CNAV, in particular the former directors of its Statistics, Forecasting and Research department, Vincent Poubelle and Pascale Breuil, for providing us with the data used in this study. The findings and conclusions presented here are solely those of the authors. Matthieu Solignac passed away in December 2020. He contributed to the analyses and early versions of this manuscript and is thus included as a coauthor with his affiliations at the time of his passing.

Results in Tables 2 and 5 remain virtually identical when we focus only on foreign-born out-migrants who returned to their country of birth (i.e., excluding those who out-migrated to other foreign countries), confirming that our results also apply specifically to the process of return migration. See Tables A4 and A5 ( online appendix ) for details.

Supplementary data

Data & figures.

Configuration of life lines of pensioners in the CNAV dataset according to their residence and mortality history. Brackets for Approach 1 and Approach 2 show the types of pensioners included in each analytic approach used in the article.

Percentage of CNAV foreign-born male pensioners aged 65+ residing abroad as of January 1, 2009, by region/country of birth

Distribution of CNAV foreign-born male pensioners aged 65+ residing in France on January 1, 2009, and their worldwide deaths and out-migrations until December 31, 2012, by country of birth

Typologies of residence and mortality follow-up are depicted in Figure 1 .

Effect of out-migration (ref. = remaining in France) on subsequent mortality among CNAV foreign-born male pensioners aged 65+ residing in France on January 1, 2009, and followed up until December 31, 2012, by country of birth: Cox regression models with associated 95% confidence intervals (CIs)

Note: Models are stratified by country of birth, with country of residence (abroad vs. France) treated as a time-varying variable.

p < .05; *** p < .001

Mortality hazard ratios of CNAV foreign-born versus native-born male pensioners aged 65+ residing in France on January 1, 2009, and followed up until December 31, 2012, by country of birth: Cox regression models with associated 95% confidence intervals (CIs)

Notes: Model 1: Foreign-born pensioners who out-migrated from France in 2009–2012 are censored at out-migration. Their exposures and deaths after out-migration are not included in the estimation of mortality hazard ratios. Model 2: Foreign-born pensioners who out-migrated from France in 2009–2012 remain in the exposure pool through the end of the observation period. Their worldwide deaths are included in the estimation of mortality hazard ratios.

The reference category in both models is native-born pensioners residing in France. The 15 native-born pensioners who out-migrated in 2009–2012 are censored at out-migration.

p < .05; ** p < .01; *** p < .001

Distribution of CNAV male pensioners aged 65+ by country of birth and place of residence on January 1, 2009, and their subsequent worldwide deaths until December 31, 2012, by country of birth and place of residence at the time of death

Mortality hazard ratios of CNAV foreign-born male pensioners aged 65+ residing abroad (i.e., out-migrated) versus in France (the reference category), 2009–2012, by country of birth: Cox regression models with associated 95% confidence intervals (CIs)

Mortality hazard ratios of CNAV foreign-born versus native-born male pensioners aged 65+, 2009–2012, by country of birth: Cox regression models with associated 95% confidence intervals (CIs)

Notes: Model 3 accounts for exposures and deaths occurring while residing in France only. Model 4 accounts for exposures and deaths among all CNAV pensioners, regardless of place of residence (in France or abroad).

The reference category in both models involves native-born pensioners residing in France.

• Previous Issue
• Previous Article
• Next Article

Advertisement

Supplements

Citing articles via, email alerts, related articles, related topics, related book chapters, affiliations.

• About Demography
• Editorial Board
• For Authors
• Rights and Permissions Inquiry
• Online ISSN 1533-7790
• Print ISSN 0070-3370
• Copyright © 2024
• Duke University Press
• 905 W. Main St. Ste. 18-B
• Durham, NC 27701
• (888) 651-0122
• International
• +1 (919) 688-5134
• Information For
• Advertisers
• Book Authors
• Booksellers/Media
• Journal Authors/Editors
• Journal Subscribers
• Prospective Journals
• Licensing and Subsidiary Rights
• View Open Positions
• email Join our Mailing List
• catalog Current Catalog
• Accessibility
• Get Adobe Reader

This Feature Is Available To Subscribers Only

Sign In or Create an Account

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
• Am J Public Health
• v.89(10); Oct 1999

The Latino mortality paradox: a test of the "salmon bias" and healthy migrant hypotheses.

OBJECTIVES: Relative to non-Latino Whites, Latinos have a worse socioeconomic profile but a lower mortality rate, a finding that presents an epidemiologic paradox. This study tested the salmon bias hypothesis that Latinos engage in return migration to their country of origin and are thereby rendered "statistically immortal" and the alternative hypothesis that selection of healthier migrants to the United States accounts for the paradox. METHODS: National Longitudinal Mortality Study data were used to examine mortality rates of the following groups for whom the salmon hypothesis is not feasible: Cubans, who face barriers against return migration; Puerto Ricans, whose deaths in Puerto Rico are recorded in US national statistics; and US-born individuals, who are not subject to either salmon or healthy migrant effects. RESULTS: The sample included 301,718 non-Latino Whites and 17,375 Latino Whites 25 years or older. Cubans and Puerto Ricans had lower mortality than non-Latino Whites. Moreover, US-born Latinos had lower mortality than US-born non-Latino Whites. CONCLUSIONS: Neither the salmon nor the healthy migrant hypothesis explains the pattern of findings. Other factors must be operating to produce the lower mortality.

Full text is available as a scanned copy of the original print version. Get a printable copy (PDF file) of the complete article (1.6M), or click on a page image below to browse page by page. Links to PubMed are also available for Selected References .

Selected References

These references are in PubMed. This may not be the complete list of references from this article.

• Adler NE, Boyce T, Chesney MA, Cohen S, Folkman S, Kahn RL, Syme SL. Socioeconomic status and health. The challenge of the gradient. Am Psychol. 1994 Jan; 49 (1):15–24. [ PubMed ] [ Google Scholar ]
• Flack JM, Amaro H, Jenkins W, Kunitz S, Levy J, Mixon M, Yu E. Epidemiology of minority health. Health Psychol. 1995 Dec; 14 (7):592–600. [ PubMed ] [ Google Scholar ]
• Markides KS, Coreil J. The health of Hispanics in the southwestern United States: an epidemiologic paradox. Public Health Rep. 1986 May-Jun; 101 (3):253–265. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Sorlie PD, Backlund E, Johnson NJ, Rogot E. Mortality by Hispanic status in the United States. JAMA. 1993 Nov 24; 270 (20):2464–2468. [ PubMed ] [ Google Scholar ]
• Liao Y, Cooper RS, Cao G, Durazo-Arvizu R, Kaufman JS, Luke A, McGee DL. Mortality patterns among adult Hispanics: findings from the NHIS, 1986 to 1990. Am J Public Health. 1998 Feb; 88 (2):227–232. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Sorlie P, Rogot E, Anderson R, Johnson NJ, Backlund E. Black-white mortality differences by family income. Lancet. 1992 Aug 8; 340 (8815):346–350. [ PubMed ] [ Google Scholar ]
• Scribner R. Mortality among Hispanics. JAMA. 1994 Apr 27; 271 (16):1238–1239. [ PubMed ] [ Google Scholar ]
• Scribner R. Paradox as paradigm--the health outcomes of Mexican Americans. Am J Public Health. 1996 Mar; 86 (3):303–305. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Shai D, Rosenwaike I. Mortality among Hispanics in metropolitan Chicago: an examination based on vital statistics data. J Chronic Dis. 1987; 40 (5):445–451. [ PubMed ] [ Google Scholar ]
• Marmot MG, Adelstein AM, Bulusu L. Lessons from the study of immigrant mortality. Lancet. 1984 Jun 30; 1 (8392):1455–1457. [ PubMed ] [ Google Scholar ]
• Pablos-Méndez A. Mortality among Hispanics. JAMA. 1994 Apr 27; 271 (16):1237–1237. [ PubMed ] [ Google Scholar ]
• Jasso G, Rosenzweig MR. Estimating the emigration rates of legal immigrants using administrative and survey data: the 1971 cohort of immigrants to the United States. Demography. 1982 Aug; 19 (3):279–290. [ PubMed ] [ Google Scholar ]
• Reichert J, Massey DS. Patterns of U.S. migration from a Mexican sending community: a comparison of legal and illegal migrants. Int Migr Rev. 1979; 13 (4):599–623. [ PubMed ] [ Google Scholar ]
• Bernal G, Enchautegui-de-Jesús N. Latinos and Latinas in community psychology: a review of the literature. Am J Community Psychol. 1994 Aug; 22 (4):531–557. [ PubMed ] [ Google Scholar ]
• Sorlie PD, Rogot E, Johnson NJ. Validity of demographic characteristics on the death certificate. Epidemiology. 1992 Mar; 3 (2):181–184. [ PubMed ] [ Google Scholar ]
• Rosenwaike I. Mortality differentials among persons born in Cuba, Mexico, and Puerto Rico residing in the United States, 1979-81. Am J Public Health. 1987 May; 77 (5):603–606. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Thiel de Bocanegra H, Gany F, Fruchter R. Available epidemiologic data on New York's Latino population: a critical review of the literature. Ethn Dis. 1993 Fall; 3 (4):413–426. [ PubMed ] [ Google Scholar ]
• Pérez-Stable EJ, Marín G, Marín BV. Behavioral risk factors: a comparison of Latinos and non-Latino whites in San Francisco. Am J Public Health. 1994 Jun; 84 (6):971–976. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Vega WA, Amaro H. Latino outlook: good health, uncertain prognosis. Annu Rev Public Health. 1994; 15 :39–67. [ PubMed ] [ Google Scholar ]
• Haynes SG, Harvey C, Montes H, Nickens H, Cohen BH. Patterns of cigarette smoking among Hispanics in the United States: results from HHANES 1982-84. Am J Public Health. 1990 Dec; 80 (Suppl):47–53. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Marin G, Perez-Stable EJ, Marin BV. Cigarette smoking among San Francisco Hispanics: the role of acculturation and gender. Am J Public Health. 1989 Feb; 79 (2):196–198. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Shai D. Cancer mortality, ethnicity, and socioeconomic status: two New York City groups. Public Health Rep. 1986 Sep-Oct; 101 (5):547–552. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Scribner R, Dwyer JH. Acculturation and low birthweight among Latinos in the Hispanic HANES. Am J Public Health. 1989 Sep; 79 (9):1263–1267. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cobas JA, Balcazar H, Benin MB, Keith VM, Chong Y. Acculturation and low-birthweight infants among Latino women: a reanalysis of HHANES data with structural equation models. Am J Public Health. 1996 Mar; 86 (3):394–396. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Notzon FC, Bobadilla JL, Coria I. Birthweight distributions in Mexico City and among US Southwest Mexican Americans: the effect of altitude. Am J Public Health. 1992 Jul; 82 (7):1014–1017. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Collins JW, Jr, Shay DK. Prevalence of low birth weight among Hispanic infants with United States-born and foreign-born mothers: the effect of urban poverty. Am J Epidemiol. 1994 Jan 15; 139 (2):184–192. [ PubMed ] [ Google Scholar ]
• James SA. Racial and ethnic differences in infant mortality and low birth weight. A psychosocial critique. Ann Epidemiol. 1993 Mar; 3 (2):130–136. [ PubMed ] [ Google Scholar ]

Salmon Bias or Red Herring? : Comparing Adult Mortality Risks (Ages 30-90) between Natives and Internal Migrants: Stayers, Returnees and Movers in Rotterdam, the Netherlands, 1850-1940

Affiliations.

• 1 Radboud Group for Historical Demography and Family History, Department of History, Radboud University, Erasmusplein 1, 6525, HT, Nijmegen, the Netherlands. [email protected].
• 2 Family and Population Studies, Centre for Sociological Research, KU Leuven, Parkstraat 45, 3000, Leuven, Belgium. [email protected].
• 3 Family and Population Studies, Centre for Sociological Research, KU Leuven, Parkstraat 45, 3000, Leuven, Belgium.
• PMID: 29043501
• PMCID: PMC5662680
• DOI: 10.1007/s12110-017-9303-1

The purpose of this research is to empirically test the salmon bias hypothesis, which states that the "healthy migrant" effect-referring to a situation in which migrants enjoy lower mortality risks than natives-is caused by selective return-migration of the weak, sick, and elderly. Using a unique longitudinal micro-level database-the Historical Sample of the Netherlands-we tracked the life courses of internal migrants after they had left the city of Rotterdam, which allowed us to compare mortality risks of stayers, returnees, and movers using survival analysis for the study group as a whole, and also for men and women separately. Although migrants who stayed in the receiving society had significantly higher mortality risks than natives, no significant difference was found for migrants who returned to their municipality of birth (returnees). By contrast, migrants who left for another destination (movers) had much lower mortality risks than natives. Natives who left Rotterdam also had significantly lower mortality risks than natives who stayed in Rotterdam. Female migrants, in particular, who stayed in the receiving urban society paid a long-term health price. In the case of Rotterdam, the salmon bias hypothesis can be rejected because the lower mortality effect among migrants was not caused by selective return-migration. The healthy migrant effect is real and due to a positive selection effect: Healthier people are more likely to migrate.

Keywords: Healthy migrant effect; Migration; Mortality; Rotterdam; Salmon bias; The Netherlands.

Publication types

• Comparative Study
• Aged, 80 and over
• Databases, Factual / statistics & numerical data
• Emigration and Immigration / statistics & numerical data*
• Longitudinal Studies
• Middle Aged
• Netherlands / epidemiology
• Transients and Migrants / statistics & numerical data*