Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Spatial variation of migrant‐native mortality differentials by duration of residence in Belgium: A story of partial convergence

Spatial variation of migrant‐native mortality differentials by duration of residence in Belgium:... INTRODUCTIONSignificant mortality advantages have repeatedly been documented for migrants compared to native born populations, although migrants in many cases face socio‐economic deprivation, show higher morbidity levels, and experience barriers in health care use relative to natives (Jervelund et al., 2017; Urquia et al., 2012; Wallace & Darlington‐Pollock, 2020). A systematic review and meta‐analysis of 96 studies, predominantly pertaining to international migrants in high income countries, finds a migrant all‐cause mortality advantage in 75% of the reported standardised mortality ratios. This advantage persists for the majority of cause groups, with the exception of infectious diseases (viral hepatitis, tuberculosis, and HIV) and external causes (assaults and deaths of undetermined cause) where migrants show higher mortality compared to the general population in destination countries (Aldridge et al., 2018). The size and direction of this healthy migrant effect are subject to considerable variation, depending on the migrant groups included, the definitions used, and the extent to which duration of residence, socio‐economic position in the destination country, or level of acculturation are taken into account (Gimeno‐Feliu et al., 2019; Hamilton, 2015; Kennedy et al., 2015).Several mechanisms have been suggested to account for the mortality advantage in various contexts. The data artefact hypothesis suggests that the mortality advantage can be entirely accounted for by unregistered (r)emigration leading to an under‐registration of deaths and an overestimation of the population at risk. Although research suggests that the impact of unregistered migration may be sizeable in specific contexts (Kibele et al., 2008), it seems unlikely that unregistered and/or selective emigration can fully account for the mortality advantage across different contexts (Deboosere & Gadeyne, 2005). An alternative mechanism that has been suggested to account for the mortality advantage is that migrants are positively selected in terms of health status, giving rise to an initial mortality advantage compared to natives which is largest shortly after immigration (healthy migrant hypothesis). The strength of the initial health selection has been shown to vary by migration motive and to be stronger in case of labour and study migration compared to family reunification (Guillot et al., 2018). Convergence between the mortality patterns of foreign‐born and native‐born populations is expected to take place with increasing duration of residence due to the waning of the initial selection effects and the dynamic frailty composition of both groups. Convergence may accelerate as a result of exposure to adverse conditions that migrants often face in destination countries. Conversely, convergence may be (partially) offset by the preservation of favourable health behaviours characteristic for the country of origin or selective emigration taking place in case of deteriorating health (salmon bias hypothesis).Acculturation of mortality patterns has frequently been studied by considering convergence of mortality risks with increasing duration of residence (Guillot et al., 2018; Vandenheede et al., 2015; Wallace et al., 2019). The spatial dimension of this convergence process has hitherto received limited attention, although several arguments suggest that it may be particularly relevant to understand the origin of the mortality advantage. First, several studies have shown that all‐cause mortality is subject to significant, and often persistent, spatial variation within countries (Deboosere & Gadeyne, 2002; Eggerickx et al., 2018; Wilson et al., 2020). Second, specific immigrant groups tend to settle in specific areas (Reniers, 1999), where local conditions may differentially affect the acculturation process and the evolution of mortality risks over time. This study contributes to the available body of literature that has overwhelmingly compared migrant mortality to the average of the entire native‐born population by taking a subnational spatial approach and comparing migrants to natives living in the same area, because their health and mortality risks are influenced by these local conditions. We consider whether migrant‐native differentials are subject to spatial variation and document whether, and to what extent, such differentials fade with increasing duration of residence in Belgium, allowing for variation in terms of migrant generation. This paper is part of a special issue on ‘Social Inequalities in Health and Mortality: The Analysis of Longitudinal Register Data from Selected European Countries’. Belgium is characterised by substantial spatial variation in all‐cause mortality and has a sizeable migrant population for which population‐wide register data allow us to address spatial variation in the mortality advantage. More specifically, this paper uses longitudinal microdata from the 2011 Belgian census which have been linked to longitudinal microdata drawn from the population register and tax return registers for the period 2001–2015.MIGRANT MORTALITY ADVANTAGE: UNDERLYING MECHANISMS AND PREVIOUS FINDINGSSeveral explanations have been suggested for the migrant mortality advantage, mainly focusing on five mechanisms: (i) selective immigration in terms of health status (healthy migrant effect), (ii) unregistered (r)emigration leading to an underestimation of death rates in migrant populations (data artefacts), (iii) selective emigration in terms of health status leading to an under‐registration of deaths (salmon bias), (iv) migration as a rapid health transition, and (v) foreign born retaining behaviours from their country of origin that positively influence their health (cultural effects) (Guillot et al., 2018; Keenan et al., 2021).Concerns about the reliability of demographic data on mortality of migrants suggests that data artefacts may (partially) account for lower mortality in migrant populations (Kibele et al., 2008). To the extent that departures from the host country are unregistered and deaths subsequently take place abroad, migrants become statistically immortal and migrant mortality is underestimated. Although studies for specific contexts have shown that data artefacts may contribute substantially to the migrant mortality advantage (Weitoft et al., 1999), other studies that have conducted sensitivity analyses on the impact of entry and exit uncertainty show that the mortality advantage cannot in general be attributed to data artefacts across contexts (Wallace & Kulu, 2014). Considering spatial variation within countries, data artefacts are unlikely to contribute to spatial variation in the mortality advantage unless vital registration is strongly decentralised and subject to differential bias across regions.Underestimation of migrant mortality may result from selective out‐migration in terms of health status and mortality risks. This mechanism is frequently referred to as the salmon bias, assuming that migrants return to their country of origin in anticipation of subsequent death (Andersson & Drefahl, 2017; Arenas et al., 2015; Bostean, 2013; Clark et al., 2007). The contribution of selective out‐migration to the mortality advantage has been notoriously difficult to assess, as it requires the longitudinal follow‐up of migrants after having resettled in their respective countries of origin. As register data are typically confined to one specific country context, this option can rarely be pursued in practice. Focusing on long‐distance migration within Sweden, Andersson and Drefahl (2017) find no evidence for the selection of healthy migrants moving from the North to the South, but do find evidence for a salmon bias in terms of an elevated mortality risk among return migrants to Northern Sweden. In contrast, studies in other settings have shown that the probability of (r)emigration is effectively lower among migrants suffering chronic disease (Norredam et al., 2015), that mortality levels would have to be unrealistically high among (r)emigrants to account for the mortality advantage (Deboosere & Gadeyne, 2005; Vandenheede et al., 2015), and that adjusting age‐specific mortality rates in groups where a salmon bias has been established cannot account for the mortality advantage (Wallace & Kulu, 2018). A further argument against the salmon bias hypothesis is that the quality and accessibility of health care in receiving countries in most cases outperforms systems and provisions in sending countries, particularly for groups for which large mortality advantages have been established.Migrants are not representative of the population in the country of origin but are self‐selected in terms of having a better health status (depending on their motive for migration), frequently referred to as the healthy migrant effect. Immigration into the destination country may in some cases involve explicit screening in terms of health status, and the profile of immigrants may be highly selective in terms of characteristics that have a moderating effect on mortality risks (Bostean, 2013; Brazil, 2017; Castaneda et al., 2015; Chiswick et al., 2008; Fuller‐Thomson et al., 2015; Hamilton, 2015; Helgesson et al., 2019; Hofmann, 2012; Ichou & Wallace, 2019; Kennedy et al., 2015; Martinez et al., 2015; Mehta & Elo, 2012). In the case of health selection, the all‐cause mortality advantage is expected to be largest for migrants with shorter durations of residence, because the contrast with the health status of the native population in the destination country is expected to be largest shortly after the health selection has taken place. With increasing duration of residence, the migrant‐native differential is expected to wear off. The selection in terms of health status and the size of the mortality advantage have been suggested to vary with the migration motive, being stronger among individuals immigrating for education and work, and less so among refugees and individuals immigrating in view of family formation or family reunification. This results in variation of the health selection effect by gender (less articulated selection among women than men, particularly in older migration cohorts), age (less selection among children and elderly persons), and country of origin (to the extent that migration motives are correlated with origin; Guillot et al., 2018). Considering disease incidence among refugees and family reunited migrants, Norredam et al. (2014) analyse disease incidence by duration of residence and find lower hazard ratios of stroke and breast cancer within 5 years after arrival, but hazard ratios increase with longer follow‐up. In contrast, hazard ratios for ischemic heart disease and diabetes were higher within 5 years after arrival and increased further with duration of residence. Results therefore suggest that the healthy migrant effect (positive health selection) should be used with caution when explaining migrants' advantageous health outcomes (Norredam et al., 2014).The conditions that immigrants are exposed to in receiving countries are often more precarious than those faced by the native population, suggesting acculturation or assimilation of mortality risks. A considerable amount of empirical evidence using different markers of socio‐economic status—education, income, activity status, or housing characteristics—shows that mortality levels are consistently higher among individuals having a more precarious socio‐economic position (Gadeyne, 2006; Vallin et al., 2001). Various pathways connect socio‐economic position to health outcomes and mortality, including life styles, environmental factors, health behaviours (e.g., preventive health care), and access to health care systems. As migrants on average have a more precarious socio‐economic position than natives, the exposure to adverse living conditions is expected to gradually override their initial health advantage and foster convergence to the mortality pattern of natives or even a reversal of the migrant‐native differential in all‐cause mortality with increasing duration of residence. The speed at which convergence, and potentially a cross‐over, takes place is determined by several other factors which may be subject to variation between local settings.Finally, the theory of migration as a rapid health transition suggests that migrants from low‐income countries immediately experience a drop in mortality from infectious disease after migration as a result of access to better health care in the destination country compared with the origin country. As an increase from chronic disease mortality is not expected to follow immediately, the mortality advantage may persist over an extended period of time (Vandenheede et al., 2015; Vanthomme & Vandenheede, 2019). This suggests that convergence between spatial mortality patterns of migrants and natives can only be expected to take place after extended durations of residence. The speed of convergence in mortality patterns will depend on the extent to which initial beneficial health behaviours are preserved, or whether rapid assimilation takes place with respect to (local) risk factors for the main causes of death (cardiovascular disease, neoplasms, and cerebrovascular disease).Whereas several mechanisms have been suggested to account for the migrant mortality advantage, the implications of these mechanisms for spatial variation in migrant‐native mortality differentials have hitherto received limited attention. Literature suggests that health selection is strongly related to age and migration motive, but it is unclear whether health selection is regionally differentiated in such a way that it would give rise to spatial variation of all‐cause mortality in migrant populations that is similar to spatial patterns found in natives. To the extent that regional variation in health selection is limited or different from spatial variation of all‐cause mortality in natives, the migrant mortality advantage may be subject to strong variation between regions. Local variation in the migrant mortality advantage is expected to diminish with duration of residence due to waning of selection effects and exposure to local conditions, but the theory of migration as a rapid health transition suggests that convergence to the spatial pattern of all‐cause mortality in natives may only emerge gradually in migrant groups with extended durations of residence. In sum, several mechanisms that have been suggested to account for the migrant mortality advantage suggest that migrant‐native differentials may well be subject to variation between regions, suggesting that it would be relevant to adopt a spatial dimension in the analysis of the mortality advantage.SPATIAL VARIATION IN MORTALITYSubstantive spatial variation in all‐cause and cause‐specific mortality has been documented in various contexts, which may represent regional variation in risk factors, regional variation in the composition of the resident population (e.g., in terms of socio‐economic position), and selective internal migration between regions in terms of health status (Keenan et al., 2021; Wilson et al., 2020). In Belgium, substantive spatial variation at both the regional and district level has been documented for all‐cause as well as cause‐specific mortality (Deboosere & Fiszman, 2009; Deboosere & Gadeyne, 2002; Duchene & Thiltgès, 1993; Eggerickx et al., 2018; Grimmeau et al., 2015). Controlling for composition in terms of socio‐economic status, pronounced regional variation persists in Belgium with lower mortality in Flemish districts than Walloon districts, and Brussels taking an intermediate position (Eggerickx et al., 2018; Grimmeau et al., 2015). This spatial pattern of all‐cause mortality is largely shaped by spatial variation in cardiovascular disease, cerebrovascular disease, and cancer which constitute the major causes of death (Grimmeau et al., 2015), although spatial patterns also vary by cause of death. A north–south divide emerges for all‐cause mortality, as well as for mortality from cardiovascular disease in both sexes, cerebrovascular disease in men, diabetes in both sexes, mental and neurological disease in both sexes, chronic obstructive pulmonary disease in men, alcohol‐related deaths in men, and non‐transport related deaths in men. In contrast, a northwest–southeast gradient parallel to the French border emerges for lip, oral cavity, pharynx, larynx, and oesophageal cancers (Renard et al., 2015).The impact of internal migration on spatial variation in all‐cause mortality has been documented in the literature. Using the Office of National Statistics Longitudinal Study, Riva et al. (2011) show that residential mobility between 1981 and 2001 accounts for about 30% of the urban–rural difference in all‐cause mortality observed in 2001–2005; individuals moving between urban and rural areas are found to be relatively healthier than long‐term urban residents (Riva et al., 2011). Using register data for the Norwegian population aged 60–89 in 1991–2002, Kravdal (2009) finds that for individuals who had moved between municipalities once in the preceding 10 years, the current socio‐economic context was not important for their mortality, suggesting that neighbourhood socio‐economic effects need some time to build up and do not dissipate soon after immigration to a new environment. Similarly, using longitudinal data for Turin between 1975 and 2005, Rasulo et al. (2012) show that internal migrants initially face lower mortality risks than locals but that migrant‐native differentials gradually reduce as internal migrants not only accumulate exposure to environmental health hazards similar to natives but also increasingly face the health consequences of worse socio‐economic conditions compared to local individuals.Four groups of factors have been suggested to account for spatial variation mortality patterns: environmental factors, contextual socio‐economic factors, effects of different public health policies and amenities, and behavioural and cultural factors which are unrelated to the individual socio‐economic position (Deboosere & Gadeyne, 2002). The physical, social, family, and institutional environment can also interact with individual characteristics to generate inequality in health and mortality outcomes (Eggerickx et al., 2018).MIGRANT GROUPS IN BELGIUMThroughout the twentieth century, Belgium has increasingly become an immigration country. Whereas the foreign population was limited to 2.1% of the population after the First World War (1920), it increased to 4% in 1938, predominantly as a result of immigration from neighbouring countries—but also Poland and Italy—for the mining, metal, and textile industries (Stols, 1985). Following World War II, migration resumed immediately with active recruitment of foreign workers from Italy (1946–1956), other Southern European countries (predominantly Spain and Greece in 1955–1961), and Turkey and Morocco (1962–1969). Immigration was particularly high in the early 1960s. Due to articulated labour shortages, migration in that period was no longer checked through work permits and migrant workers were allowed to settle in Belgium under tourist visas, with regularisation typically following after medical screening (Martens, 1985). Labour migration decreased substantially following the economic stall in 1967, after which migration was strongly controlled through the selective granting of work permits for predominantly highly educated profiles (Martens, 1985; Reniers, 1999). Following the migration stop in 1974, family reunification of spouses and children who were still in the country of origin gained importance (Reniers, 1999) as well as non‐European marriage migration when the intermediate and second generation reached adolescence in the 1980s and 1990s (Lievens, 2000). Following a period of low immigration from the 1980s up to the mid‐1990s, the late 1990s witnessed an articulated increase of immigration and a diversification of migration motives and origin countries resulting from the free movement of individuals within the European Union (Schengen treaty), continued marriage migration and family reunification, but also student and labour migration, accompanied by period fluctuations in the number of asylum seekers and refugees (Corijn & Lodewijckx, 2009; Myria, 2019).Several studies have documented migrant mortality in Belgium. Using linked microdata from the 1991 census and the population register for the period 1991–1995, Deboosere and Gadeyne (2005) compare all‐cause mortality in natives and different origin groups. With the exception of immigrants from France (significantly higher mortality), Germany (no significant difference), and Luxemburg (no significant difference), migrant men in the age group 25–54 originating from other neighbouring countries, Southern Europe as well as Turkey and Morocco, have significantly lower mortality than natives. Differentials typically increase in favour of migrant men when additional adjustments are made for age, level of education, and housing characteristics (Deboosere & Gadeyne, 2005). Results for women were found to be similar to those for men. Using linked census and register data for 2001–2010, Vandenheede et al. (2015) find that first generation immigrants have lower all‐cause and chronic disease mortality than natives: a mortality advantage that wears off with increasing duration of residence, in line with health selection and acculturation theories. Consistent with the migration as rapid health transition theory, infectious‐disease mortality is higher in specific non‐European migrant groups, but not in other origin groups. Second generation migrants on the other hand suffer a mortality disadvantage relative to natives which disappears when controlling for age, educational attainment, employment status, and housing characteristics. Considering approximately the same period and controlling for age, socio‐economic position, and urban typology, Reus‐Pons et al. (2016) find that the mortality advantage persists after age 50 for most origin groups and that the mortality disadvantage of specific origin groups can be attributed partially to their lower socio‐economic status compared to natives. Similar to other studies considering migrant‐native mortality differentials in Belgium (Bauwelinck et al., 2017; Deboosere & Gadeyne, 2005; Van Hemelrijck et al., 2016; Vandenheede et al., 2015), the study by Reus‐Pons et al. does not consider spatial variation in migrant‐native differentials in all‐cause or cause‐specific mortality.RESEARCH QUESTIONS AND HYPOTHESESPrevious studies have documented extensively that both all‐cause and cause‐specific mortality are subject to considerable regional variation and that acculturation of migrant mortality to native patterns takes place only gradually with increasing duration of residence. This suggests that the spatial pattern of all‐cause mortality is likely to differ between migrants and natives and that the migrant‐mortality advantage is therefore subject to variation between local settings. However, this spatial dimension to the migrant mortality advantage has not been explored in previous research. This paper introduces a spatial dimension into the analysis of the migrant mortality advantage and has three aims:First, we assess whether the migrant mortality advantage documented previously using the 1991 and 2001 Belgian censuses persists up to 2011–2015, controlling for socio‐demographic and socio‐economic characteristics. We distinguish between migrant generations and differentiate by duration of residence as we expect that the mortality risk in migrant populations will converge to natives with length of stay, assuming that exposure to local conditions and adoption of related behaviours play an important role in the acculturation of mortality (Bos et al., 2007; Vandenheede et al., 2015; Hypothesis 1).Second, controlling for socio‐demographic and socio‐economic characteristics, we consider spatial variation in migrant mortality. In line with health selection and acculturation theories, we expect that spatial variation of all‐cause mortality in recent migrants will be dissimilar from natives. We expect that migrant mortality will converge to the spatial pattern of natives among first generation migrants with extended durations of residence and groups where health selection can be largely presumed absent such as the second generation (Hypothesis 2).Third, we consider spatial variation in the migrant mortality advantage and test whether the overall mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population and their concentration in local settings characterised by high native mortality and large migrant‐native differentials in all‐cause mortality (Hypothesis 3).DATA AND METHODSDataThe analyses of all‐cause mortality use microdata from the 2011 Belgian census which have been linked to the population register both retrospectively for the period 2001–2010 and prospectively for the period 2011–2015 using a coded individual identifier. The 2011 census is register‐based and provides detailed information on individual and household characteristics on January 1, 2011, drawn from the population register, social security registers, the address and dwelling register, and tax return data. We focus on all‐cause mortality among individuals aged 25 to 75 between January 1, 2011, and December 31, 2015. With respect to mortality, the exact date of death is available from the population register. With respect to emigration and accurately detecting events that cause truncation, the situation is somewhat more complex. Registration of emigration can be considered (quasi‐)universal, but the timing of the registration may in certain cases be late, thus overestimating exposure to some degree. Individuals may in principle leave the country to return to their country of origin or move abroad without notifying authorities of their emigration (Deboosere & Gadeyne, 2005). However, a large share of emigrants and their household members are estimated to communicate their new address abroad in order to be able to benefit from social security entitlements and other rights that they themselves or their relatives have accumulated in Belgium (through labour, social security, or the health system). Those who do not actively report their emigration will with some delay be detected through a variety of mechanisms, such as failed contact attempts by administrations or utility companies, address conflicts in the population register when new occupants move into the former dwelling of emigrated households, and follow‐up of unpaid bills. After the unreported leave is detected, the municipality administration will make the necessary corrections to the population register, but it is unclear to what degree the newly recorded emigration will bear a correct date. Any deviations from reality can be expected to be selectively late. However, late recording of emigrations is unlikely to contribute substantially to the migrant mortality advantage in Belgium, as proportions (r)emigrating—and within this group proportions doing so without reporting it—are unlikely to be large enough to explain (significant parts of) the observed mortality advantage. As the data only cover the population officially residing in Belgium, we cannot test directly whether migrants are positively selected for health status. Individuals are censored on December 31, 2015, in case they survived throughout the observation period, or in case of emigration, censored on June 30 of the year in which they left the country as the exact date of emigration was not recorded.ModelConsistent with the construction of synthetic or period life tables, both the descriptive results and the multivariate analyses use a late entry design where exposure‐to‐risk and mortality occurring in the observation window from January 1, 2011, until December 31, 2015, are appropriately incorporated into the estimation to generate the synthetic or period hazard function of all‐cause mortality by age (Singer & Willett, 2003). Cox regression models were used to estimate the effect of covariates on the continuous‐time hazard of all‐cause mortality (Singer & Willett, 2003). The analyses of migrant‐native differentials in all‐cause mortality incrementally control for age (baseline) (Model 1), living arrangements (Model 2), socio‐economic position (Model 3), and housing characteristics (Model 4). Subsequently, we introduce district as a covariate (Models 5a–5c) to discern spatial variation in all‐cause mortality at the district‐level net of spatial variation in the composition of the resident population in terms of characteristics considered in Models 1 through 4 which have been shown to affect mortality (Gadeyne, 2006). Model 5a re‐estimates Model 4 using a collapsed version of the typology of migrant background as a reference model for subsequent models including spatial variation. Model 5b introduces district as categorical covariate into an additive model with the typology of migrant background, constraining spatial variation in all‐cause mortality to be identical across natives and migrant groups. Finally, Model 5c includes the interaction between the typology of migrant groups and district, to assess whether spatial variation in all‐cause mortality differs between migrants and natives and whether spatial variation of all‐cause mortality in migrant groups converges to the pattern found in the native population with increasing duration of residence:lnhtij=lnh0tj+∑βk.Mki+∑βk.Dki+∑βk.LAkij+∑βk.SEPkij+∑βk.HCki,where h0(tj) is the unspecified baseline hazard function of all‐cause mortality by age, Mki is a set of dummy variables distinguishing migrant groups from natives (reference category) by migrant generation and duration of residence, Dki is a set of dummy variables comparing the mortality risk between districts, LAkij is a set of dummy‐variables denoting the living arrangement on January 1 in each observation year t, SEPkij is a vector of time‐constant and time‐varying indicators of socio‐economic position, and HCki denotes a vector of housing characteristics as measured in the 2011 Census. All models were estimated separately for men and women. Missing values were deleted listwise in view of model comparison using Likelihood Ratios tests. All models were estimated using the stcox and contrast commands in Stata 14MP.CovariatesMigrant background and duration of residence constitute the main variables of interest in the study. Migrant background was defined as not having Belgian nationality at birth, which is considered more accurate to identify migrants than country of birth, as this would include natives with Belgian parents who were born abroad. Duration of residence is based on information available from the population register which provides the year of immigration since 1980. The exact duration of residence cannot be ascertained for migrants who settled in Belgium prior to 1980 and have an extended duration of residence. As a result, the typology of migrant background distinguishes seven groups: (i) Belgian born of native origin, (ii) Belgian born of foreign origin, (iii) foreign born who immigrated prior to the age of 18, and finally foreign born who immigrated after the age of 18 and who had attained (iv) 0 to 9 years, (v) 10 to 19 years, (vi) 20 to 29 years, and (vii) 30 years of residence or more in Belgium by the time of the 2011 census (see Tables 1, A1, and A2).1TableRegional distribution of native and migrant populations aged 25–75 by sex, migrant generation, and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):Belgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7]Men:Flemish region67.130.944.543.946.538.729.1‐ Province Antwerp18.610.119.818.118.314.19.3‐ Province Limburg8.710.38.68.59.98.39.3‐ Province Flemish‐Brabant11.13.65.36.46.48.24.4‐ Province East‐Flanders15.74.57.37.27.45.04.3‐ Province West‐Flanders13.02.43.53.84.53.11.8Brussels capital region4.114.927.034.929.328.922.3Walloon region28.854.328.521.224.232.448.7‐ Province Hainaut10.227.89.57.57.512.724.5‐ Province Liège8.818.812.88.410.611.116.7‐ Province Walloon‐Brabant3.22.82.62.22.64.53.5‐ Province Namur4.33.52.11.51.92.42.8‐ Province Luxembourg2.41.31.61.71.61.91.2100.0100.0100.0100.0100.0100.0100.0Women:Flemish region68.331.944.644.547.140.630.9‐ Province Antwerp18.810.220.619.019.716.59.6‐ Province Limburg9.111.17.88.910.16.510.0‐ Province Flemish‐Brabant11.13.65.45.75.88.14.6‐ Province East‐Flanders16.14.67.07.37.46.04.2‐ Province West‐Flanders13.32.53.73.64.03.52.6Brussels capital region3.714.225.934.930.429.619.6Walloon region28.054.029.520.622.529.949.5‐ Province Hainaut9.727.49.57.47.111.724.4‐ Province Liège8.718.713.58.310.310.817.4‐ Province Walloon‐Brabant3.02.93.01.82.23.73.4‐ Province Namur4.33.62.11.41.61.93.0‐ Province Luxembourg2.41.31.41.71.41.71.4Total100.0100.0100.0100.0100.0100.0100.0Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors.To determine whether spatial variation in migrants' mortality risks converges to that of natives as their duration of residence increases, mortality risks are differentiated by district (n = 43), which have been shown to exhibit significant variation in mortality levels in the general population (Deboosere & Gadeyne, 2002; Eggerickx et al., 2018). The level of districts is not ideal as it reflects a purely administrative division and districts vary considerably in size. Although lowering the geographical level of analysis would be relevant to understand spatial variation in mortality risks (Deboosere & Gadeyne, 2002), the level of municipalities (n = 589) did not allow robust estimates for migrant groups differentiated by migrant generation and duration of residence. To avoid that estimates of spatial convergence are contaminated by internal migration, the analyses are restricted to the subset of individuals whose district of residence on January 1 has remained constant between 2001 and 2015 or since immigration into the country. Previous studies suggest that internal migrants differ in terms of both mortality levels and spatial variation of all‐cause mortality compared to the group who lived in the same district throughout the observation period (Kravdal, 2009; Riva et al., 2011). The subset of individuals whose district of residence has remained constant throughout the observation period or since immigration represents 81.02% and 81.81% of native men and women and 78.81 and 78.84% of migrant men and women, respectively. To avoid issues of quasi‐complete separation in the model including interaction between district and migrant typology (Model 5c), the districts Diksmuide and Veurne (province of West‐Flanders) were collapsed for men, whereas the districts Ath and Tournai (province of Hainaut), the districts Diksmuide and Roeselare (province of West‐Flanders), and the districts Kortrijk and Tielt (province of West‐Flanders) were collapsed for women.The history of migration into Belgium has resulted in migrant populations being strongly concentrated in the former industrial belt in Wallonia, the former mining region in Limburg, smaller industrial areas, and central regions characterised by large secondary labour markets (Kesteloot, 1985). This results in a settlement pattern of the migrant population that is altogether different from that of the native‐born population (Table 1). The larger part of the native population lives in the Flemish region (~67%), followed by the Walloon region (28%), with only a minority living in the Brussels capital region (~4%). In contrast, both first generation migrants who immigrated prior to 1980 and second generation migrants are primarily concentrated in the Walloon region (50% to 55%), followed by Flanders (approximately 30%) and a more sizeable portion living in the capital region of Brussels (15 to 20%). Among migrants who immigrated after 1980, Flanders has gained importance with 40% to 45% of first‐generation migrants living in Flemish districts at the time of the 2011 Census, and equal shares of around 30% living in Wallonia and Brussels, although Brussels has clearly gained prominence over Wallonia between 2000 and 2010.Variation in living arrangements is captured through individuals' household positions using the LIPRO household typology (Van Imhoff & Keilman, 1991), distinguishing following positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child in a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL). Household position is included in the analysis as a time‐varying covariate as the information on living arrangements is updated annually (referring to the situation on January 1 of every year of observation).The socio‐economic position is measured using the activity status at the time of the 2011 census in tandem with a time‐varying indicator of income derived from the tax registers. The indicator of activity status distinguishes (i) students, (ii) employed individuals, (iii) housekeepers and other inactive persons, (iv) pensioners, (v) unemployed individuals with no prior work experience, and (vi) individuals with prior work experience who were unemployed at the time of the 2011 census. Income reflects individuals' income recorded in the annual tax returns between 2011 and 2015. Income is collapsed into deciles and included in the analysis as a time‐varying covariate. Education was not included as an indicator of socio‐economic position as information on educational attainment is frequently incomplete or missing for migrants, particularly for recent immigrants. We do not consider the lack of information on educational attainment to be problematic for the aims of this study as previous research for Belgium has shown that education is predominantly a relevant indicator of mortality risks in the youngest age groups, whereas professional status and particularly housing characteristics are more discriminate of mortality among individuals aged 25 and older (Gadeyne, 2006). Lacking reliable information on level of education, the analyses are restricted to men and women aged 25 and older.The 2011 Census provides a rich set of indicators of housing quality which serve as an indicator of accumulated wealth and have been shown to be strongly associated with mortality risks (Gadeyne, 2006). The analyses control for the following housing characteristics: (i) ownership status (renter versus owner), (ii) presence of a bath or shower in the dwelling, (iii) presence of central heating in the dwelling, (iv) presence of multiple dwellings in the housing unit (multiple vs. single dwelling), (v) number of rooms per inhabitant in the dwelling (less than 0.5 rooms, 0.5 to 1.5 rooms, 1.5 to 2.99 rooms, and 3 or more rooms), and (vi) period of construction (before 1919, 1919–1945, 1946–1960, 1961–1970, 1971–1980, 1981–1990, 1991–2000, and 2001–2010).Tables A1 and A2 provide the distribution of person‐years by covariates for both natives and migrants by migrant generation and duration of residence. As could be expected, migrants with durations of residence under 20 years are on average younger than natives, whereas migrants with extended durations of residence of 30 years and longer are somewhat older than natives on average. With respect to household positions, women who recently settled in Belgium are somewhat more likely to be in a union and have somewhat higher proportions of being heads of single person households. The latter household position is also more prevalent among migrant women with longer durations of residence than among native women. Also, the proportion of singles is higher among migrant women with a duration of residence of 30 years and longer, than among natives. The household positions of migrant men residing in the country for at least 30 years are similar to natives, although the proportion of singles is higher among recent immigrants. With respect to activity status, the proportions employed or retired reflect the differences in age distribution between groups, but the proportion unemployed is consistently higher in migrant groups than among natives. With respect to income, migrants are more concentrated in the middle‐income groups than natives, which is also reflected by the housing characteristics. Ownership is less prevalent in migrant groups, and the proportion of migrants living in dwellings without a bathroom or central heating is substantially higher. Similarly, migrants more frequently live in older housing units that often have multiple dwellings with less rooms available per inhabitant on average.RESULTSWe first estimate the mortality advantage of migrants compared to natives by duration of residence, controlling for age, living arrangement, socio‐economic position, and housing characteristics. Second, we document spatial variation of all‐cause mortality in the native population and test whether the spatial variation of all‐cause mortality in migrants converges to the pattern of natives with increasing duration of residence. Finally, we consider spatial variation in the migrant mortality advantage and test whether the overall mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population and their concentration in local settings characterised by high native mortality and large migrant‐native differentials in all‐cause mortality.Migrant‐native differentials in all‐cause mortality by duration of residenceTable 2 reports hazard ratios of all‐cause mortality in migrant women and men compared to natives. Mortality hazards are approximately 13% lower in migrant men and women compared to natives (Table 2, Model 1) and the mortality advantage of migrants relative to natives increases to 35% controlling for living arrangements, socio‐economic position, and housing characteristics (Table 2, Model 4). Further disaggregating by migrant generation and duration of residence indicates that mortality hazards of the intermediate and second generation are similar to natives, and even significantly higher among women who immigrated as children (Table 2, Model 1). The mortality hazard being similar or higher is due to the disadvantaged socio‐economic position of the intermediate and second generations as mortality hazards are 17% to 26% lower than natives controlling for age, living arrangement, socio‐economic position, and dwelling characteristics (Table 2, Model 4). In contrast, mortality hazards are significantly lower than natives among first generation men and women who immigrated after the age of 18, with differentials being largest among men and women with durations of residence between 0 and 9 years, and differentials being smaller in migrant groups with 10–19 and 20–29 years of residence, and particularly 30 or more years of residence (Table 2, Model 1). Similar to the results for the intermediate and second generation, the mortality advantage increases when controlling for living arrangements, socio‐economic position and housing characteristics. Mortality hazards are more than 60% lower than natives among migrants with 0 to 9 years of residence in Belgium, 45 to 50% lower among migrants with durations of residence ranging from 10 to 29 years, and approximately 27% lower among migrants with durations of residence of 30 years or longer (Table 2, Model 4). The mortality advantage is similar for men and women allowing for migrant generation and duration of residence and is consistent with health selection and acculturation mechanisms, thus largely confirming Hypothesis 1.2TableHazard ratios (HR) comparing all‐cause mortality in migrant men and women to natives by migrant generation and duration of residence, ages 25–75, Belgium, 2011–2015Model 1Model 2Model 3Model 4HRSig.HRSig.HRSig.HRSig.A. Women (native women reference)Belgian born, Belgian origin1.000Ref.1.000Ref.1.000Ref.1.000Ref.Migrant & migrant background0,869***0,834***0,690***0,634***Belgian born, Belgian origin1.000Ref.1.000Ref.1,000Ref.1,000Ref.Belgian born, foreign origin0,966‐0,918**0,841‐0,805***Agemig<18, foreign origin1,325*1,280*0,947***0,874‐Agemig>18, 00–09 yrs residence0,755***0,654***0,439***0,374***Agemig>18, 10–19 yrs residence0,819***0,797***0,609***0,548***Agemig>18, 20–29 yrs residence0,790***0,781***0,608***0,574***Agemig>18, 30 + yrs residence0,898***0,887***0,780***0,720***B. Men (native men reference)Belgian born, Belgian origin1.000Ref.1.000Ref.1.000Ref.1.000Ref.Migrant & migrant background0,871***0,871***0,694***0,650***Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1,000Ref.Belgian born, foreign origin1,028‐0,975‐0,860***0,834***Agemig < 18, foreign origin1,105‐1,059‐0,797*0,740**Agemig > 18, 00–09 yrs residence0,740***0,686***0,448***0,387***Agemig > 18, 10–19 yrs residence0,751***0,762***0,557***0,509***Agemig > 18, 20–29 yrs residence0,751***0,771***0,576***0,549***Agemig > 18, 30+ yrs residence0,899***0,940***0,779***0,731***Model legend: models control incrementally for age (Model 1), living arrangements (Model 2), activity status and income (Model 3) and housing characteristics (Model 4).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), Calculations by authors.*Significance levels: ‐ not significant, p < .050,**p < .010,***p < .001.Spatial variation in all‐cause mortality in migrant and native populationsTo compare spatial variation in all‐cause mortality between natives and migrants, the intermediate and second generation were collapsed into a single category, and also, first‐generation migrants with 10–19 and 20–29 years of residence were combined. The collapsed typology allows to compare migrants to natives in all districts. Collapsing the migrant typology did not result in a significant deterioration of model fit for either men or women. Model 5a (Table 3) replicates Model 4 (Table 2) using the collapsed migration typology. In Model 5b, district is added to Model 5a, whereas Model 5c additionally tests the interaction between the migration typology and district. Comparing models 5c and 5b indicates that the spatial variation in all‐cause mortality differs significantly between migrants and natives (∆‐2LL = 347.48; ∆ df = 148; p  < .001 for women, and ∆ ‐2LL = 512.22; ∆ df = 164; p  < .001 for men). The results of Model 5c are shown in Figure 1 (women) and Figure 2 (men). In both figures, districts were ranked by the mortality hazards in the native population, resulting in the familiar pattern that has been documented extensively in the literature: Flemish districts show the lowest mortality, with mortality being lowest in districts located in the provinces of East and West Flanders, followed by districts located in the provinces of Antwerp, Flemish‐Brabant and Limburg. With the exception of Mouscron, the Walloon districts show higher mortality than Flemish districts, with mortality levels being somewhat more favourable in districts located in the provinces Walloon Brabant and Luxembourg and overall less favourable for districts in the province of Namur. Districts in the provinces of Hainaut and Liège show somewhat more heterogeneity compared to the districts in the other Walloon provinces. Finally, the capital region of Brussel takes an intermediate position, with mortality being higher than in all of the Flemish districts, as well as a limited group of Walloon districts.3TableHazard ratios (HR) comparing all‐cause mortality in migrant men and women to natives by migrant generation and duration of residence, ages 25–75, Belgium, 2011–2015Model 5aModel 5bModel 5cHRSig.HRSig.HRSig.A. Women (native women reference)Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1.5 and 2nd generation0.809***0.753***0.840***Agemig>18, 00–09 yrs residence0.374***0.375***0.447***Agemig>18, 10–29 yrs residence0.559***0.556***0.679***Agemig>18, 30 + yrs residence0.720***0.682***0.805***B. Men (native men reference)Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1.5 and 2nd generation0.830***0.775***0.852**Agemig>18, 00–09 yrs residence0.387***0.408***0.444***Agemig>18, 10–29 yrs residence0.525***0.546***0.658***Agemig>18, 30 + yrs residence0.731***0.700***0.745***Model legend: Model 5a reports the migrant mortality advantage controlling for age, living arrangements, activity status and income. Model 5b incrementally includes district, whereas Model 5c additionally includes the interaction between migrant typology and district. The estimates of the migrant mortality advantage in Model 5c reflect the average migrant mortality advantage across districts.Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), Calculations by authors.*Significance levels: ‐ not significant. p < .050,**p < .010,***p < .001.1FigureDistrict‐level variation in all‐cause mortality hazards of native versus migrant women by migrant generation and duration of residence controlling for household position, socio‐economic and housing characteristics (Model 5c), Belgium, 2011–20152FigureDistrict‐level variation in all‐cause mortality hazards of native versus migrant men by migrant generation and duration of residence controlling for household position, socio‐economic, and housing characteristics (Model 5c), Belgium, 2011–2015Turning to spatial variation in migrant mortality, the familiar spatial variation of all‐cause mortality in native women does not emerge among migrant women who settled in Belgium as adults and had a duration of residence between 0 and 9 years at the time of the 2011 Census (Figure 1a). Allowing for differences in socio‐demographic profile and socio‐economic position, the mortality advantage relative to natives is subject to substantial variation between districts, with differentials generally being smallest in the Flemish districts and considerably larger in Brussels and the majority of Walloon districts, where migrant groups are predominantly located (Table 1). With increasing duration of residence, the mortality advantage becomes smaller (Figure 1b) and the spatial gradient in all‐cause mortality among migrants with a duration of residence of 30 years or longer converges to the pattern of native women, albeit that the mortality advantage remains somewhat larger in the Walloon districts (Figure 1c). Among women of migrant origin who were born in Belgium or immigrated under the age of 18 (Figure 1d), the mortality advantage relative to natives is virtually absent in most of the Flemish districts and much smaller in Walloon districts compared with the other migrant groups. This suggests that health selection has only carried over to the intermediate and second generations to a limited extent. Alternatively, these groups may continue to enjoy a limited mortality advantage compared to natives as a result of group‐specific behaviours, but this advantage is nullified by the unfavourable socio‐economic position of the groups considered. Similar patterns emerge with respect to the mortality advantage of migrant men compared to natives (Figures 2a–d). For both men and women, the results largely confirm hypothesis 2 stating that health selection and acculturation mechanisms imply a gradual convergence of mortality risks between migrants and natives with increasing duration of residence, and also a gradual convergence of spatial variation in all‐cause mortality, leading to significant between‐district variation in the migrant mortality advantage, particularly among migrant groups with short durations of residence.Finally, considering the between‐district variation of the migrant mortality advantage, Model 5c calculates the average mortality advantage of each migrant group as the geometric mean of the mortality advantages within districts (Table 3). Allowing for living arrangements, socio‐economic position, and housing characteristics, mortality hazards in men and women of the intermediate and second generation are on average approximately 15% lower than is the case for natives. Among men and women with a duration of residence between 0 and 9 years, the mortality advantage amounts to 55% on average across districts, compared with 35% among men and women with durations of residence between 10 and 29 years, and 20% to 25% among women and men with durations of residence exceeding 30 years, respectively (Model 5c, Table 3). The comparison of Model 5c to the results of Model 5a provides clues as to how settlement patterns and spatial variation in the mortality advantage can partially explain the overall migrant mortality advantage in Belgium. Allowing for living arrangements, socio‐economic position and housing characteristics, the average mortality advantage within districts (Model 5c) is smaller for all migrant groups than the country‐level mortality advantage (Model 5a). In line with the acculturation and migration as rapid health transition arguments, mortality in several migrant groups is less subject to spatial variation than mortality in natives (Figures 1 and 2). In regions with low native mortality such as the majority of Flemish districts, this gives rise to a smaller mortality advantage for migrants compared to districts with higher mortality in the native population such as Brussels and the majority of Walloon districts. As the migrant population is strongly concentrated in districts with high native mortality and large migrant‐native differentials in all‐cause mortality (Table 1), the country‐level mortality advantage largely reflects the sizeable migrant‐native mortality differentials in these districts.DISCUSSION AND CONCLUSIONThis paper builds on two separate bodies of literature that consider spatial variation in mortality and migrant‐native differentials in mortality, respectively. The first body of literature has documented for various contexts that all‐cause and cause‐specific mortality are typically subject to spatial variation net of compositional effects in terms of socio‐economic position and selective internal migration between regions by health status. The second body of literature has documented for various contexts that recent migrants typically have lower mortality than natives as a result of health selection and that the mortality advantage gradually diminishes with increasing duration of residence and among the second generation in line with acculturation theories. The latter arguments suggest, however, that research on migrant‐native differentials in all‐cause and cause‐specific mortality may benefit from explicitly introducing a spatial dimension. First, arguments referring to health selection and acculturation suggest that spatial variation in migrant mortality may largely differ from the pattern found in the native population, implying that the migrant‐native differentials may be subject to considerable regional variation allowing for migrant generation and duration of residence. Second, migrant groups often have specific settlement patterns, implying that consideration of these spatial aspects may provide additional insight into the nature of the country‐level mortality advantage.This paper contributes to the literature by introducing a spatial dimension in the analysis of migrant‐native differentials controlling for socio‐demographic characteristics, socio‐economic position, and housing characteristics. Considering variation in the mortality advantage migrant generation and duration of residence, our findings are consistent with earlier findings on migrant‐native differentials in all‐cause mortality in Belgium. Similar to the study by Vandenheede et al. (2015) for the period 2001–2010, we find that the mortality advantage compared to natives is largest among first generation migrants with short durations of residence between 0 and 9 years and is progressively smaller among first generation migrants with durations of residence between 10 and 19 years, 20–29 years and 30 years and over, in line with health selection and acculturation arguments (Hypothesis 1). Migrants who immigrated as children (intermediate generation) and the second generation, where health selection can be largely presumed absent, have mortality risks similar to natives. For all groups, migrant‐native differentials in all‐cause mortality increase when controlling for living arrangements, socio‐economic position and housing characteristics, indicating that the socio‐economic profile of migrant groups is less favourable than the profile of natives and partially conceals the mortality advantage.Health selection and acculturation theories suggest that the gradual convergence of mortality patterns may also apply to spatial variation of all‐cause mortality, with spatial variation in migrant mortality being dissimilar from natives among first‐generation migrants with short durations of residence and being more similar among first generation migrants with long durations of residence, as well as the intermediate and second generation (Hypothesis 2). Controlling for socio‐demographic profile, socio‐economic position, and housing characteristics, we find that spatial variation of all‐cause mortality differs significantly between migrant groups and natives, with spatial variation in mortality becoming increasingly similar to natives with increasing duration of residence. Although migrant‐native differentials in all‐cause mortality are virtually absent in the Flemish districts among first‐generation migrants with an extended duration of residence, as well as the intermediate and second generation, the mortality advantage persists in Brussels and the majority of Walloon districts and predominantly reflects the high mortality level of natives in these districts.The finding that spatial variation in all‐cause mortality differs between natives and a sizeable part of the migrant population implies that the mortality advantage is subject to variation between districts. The country‐level mortality advantage is obtained by comparing the nation‐wide mortality of migrants to the nation‐wide mortality rate of natives and presents a weighted average of the mortality advantage within districts which partially reflects the spatial distribution of both migrant and native populations. As a result, the migrant mortality advantage at the country level largely reflects the mortality advantage in districts where migrants are primarily located, which may not necessarily reflect the mortality advantage in other districts. Given these mechanisms, the last part of the paper explored whether and to what extent the country‐level mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population (Hypothesis 3). In the Belgian case, a substantial part of the migrant population is located in Brussels and Wallonia, characterised by higher mortality among natives, which gives rise to a large mortality advantage in these districts. In contrast, the mortality advantage is substantially smaller in the majority of Flemish districts. Whereas a high level of convergence in mortality risks is reached between migrants and natives in Flemish districts, the convergence is lower in Brussels and the majority of Walloon districts. This is also true in groups with extended durations of residence exceeding 30 years as well as the intermediate and second generations, indicating that only partial exposure or acculturation with local risk factors takes place in migrant populations in these districts (Grimmeau et al., 2015). The partial convergence of spatial variation in all‐cause mortality of migrants and natives is likely to be driven by selective exposure to or adoption of risk behaviours by migrants, resulting in a partial adoption of the spatial patterns of cause‐specific mortality found in natives. The country‐level mortality advantage partially reflects the concentration of the migrant population in districts characterised by high mortality among natives and is therefore somewhat larger than the average mortality advantage within districts.Despite their strengths, the analyses reported in this paper also suffer a number of limitations. First, although the gender differences in mortality advantage reported in this paper are consistent with the healthy migrant hypothesis, the analyses could not explicitly control for migration motive, although this has been shown to affect the health selection at immigration (Syse et al., 2018). As the Belgian Census and population register only cover the resident population, the data at hand do not allow to determine directly whether immigrants are selective in terms of health status compared to the resident population in the country of origin. Second, given that the observation window covers the period from 2011 to 2015, the different groups in terms of duration of residence pertain to different migration cohorts. Hence, the finding that migrant‐native mortality differentials are smaller in migrant groups with longer duration of residence may suggest convergence of mortality risks or may indicate that health selection was less articulate in older cohorts than is the case in more recent migrant cohorts. Particularly for men, we consider it unlikely that health selection would have increased over time as labour migration was the prime motive in older migrant cohorts (suggesting strong health selection and even explicit health screening for some groups), whereas migration motives have become more diverse in recent cohorts (suggesting weaker health selection). Third, the analysis has not used spatial lag models, which are less straightforward to implement in the context of hazard models. We feel that future work could further explore the clustering of mortality risks in neighbouring districts which may violate the assumption of independence.Despite the limitations, we feel that our analysis illustrates the relevance of introducing a spatial component into the analysis of the migrant mortality advantage, and several avenues for future research can be identified. First, the heterogeneity in the migrant population could be further explored; migration motives and socio‐economic position are known to vary between EU and non‐EU migrants, but also by the specific country of origin, particularly in the case of non‐EU migrants (UNIA & Fod WASO, 2017). An obvious extension of the analyses presented in this paper would therefore be to explore how selective exposure to local conditions affects the mortality advantage found in different migrant groups given their specific settlement patterns, which may enhance our understanding of the factors driving between‐group variation in mortality outcomes. Although we consider such differentiation to be a potentially fruitful avenue for future research, analyses for specific origin groups will have to deal with substantially smaller sample sizes and specific patterns of residence that preclude a nation‐wide comparison of spatial variation in mortality patterns, as was the case in this paper. As a result, analyses for specific subgroups are likely to require more specific research designs. Second, the analyses have focused on individuals who have been consistently exposed to local conditions for a period of 15 years in line with previous research that suggests that change in context does not affect mortality risks immediately (Kravdal, 2009; Riva et al., 2011). The results in this paper refer to a (large) subset of the population, but more work is needed to quantify how exposure to different spatial contexts over the life course affects mortality risks, in both migrant and native populations. Third, although the possibility of unregistered emigration in the population register was deemed unlikely, it is clear that a fair amount of registered emigration took place throughout the observation period, with emigration being considerably more prevalent than mortality among migrants with shorter duration of residence. To the extent that emigration among recent migrants is selective in terms of health status (e.g., as a result of limited access to social security and/or health care), the impact on the mortality advantage found among recent immigrants may be substantial. We consider more detailed work on selective emigration among recent immigrants and its impact on the mortality advantage found in this group to be a relevant avenue for future research. The decline of the mortality advantage with increasing duration of residence and the convergence of spatial variation in all‐cause mortality between natives and migrants with an extended duration of residence suggest that acculturation of mortality risks is taking place that is not offset by selective (r)emigration. Fourth and finally, a logical extension would be to consider regional patterns in cause‐specific mortality given that regional variation in all‐cause mortality is the outcome of regional variation in patterns of cause‐specific mortality.ACKNOWLEDGEMENTSWe thank Patrick Lusyne, Cloë Ost, and Lien Co at Statistics Belgium for making the coded longitudinal microdata from the 2011 Belgian Census, the Population Register (2000–2015), and the tax return register (2000–2015) available.DATA AVAILABILITY STATEMENTThe data that support the findings of this study are available from Statistics Belgium. Restrictions apply to the availability of these data, which were used under licence for this study.REFERENCESAldridge, R. W., Nellums, L. B., Bartlett, S., Barr, A. L., Patel, P., Burns, R., Hargreaves, S., Miranda, J. J., Tollman, S., Friedland, J. S., & Abubakar, I. (2018). Global patterns of mortality in international migrants: A systematic review and meta‐analysis. Lancet, 392(10164), 2553–2566. https://doi.org/10.1016/s0140-6736(18)32781-8Andersson, G., & Drefahl, S. (2017). Long‐distance migration and mortality in Sweden: Testing the salmon bias and healthy migrant hypotheses. Population Space and Place, 23(4), 11. https://doi.org/10.1002/psp.2032Arenas, E., Goldman, N., Pebley, A. R., & Teruel, G. (2015). Return migration to Mexico: Does health matter? Demography, 52(6), 1853–1868. https://doi.org/10.1007/s13524-015-0429-7Bauwelinck, M., Deboosere, P., Willaert, D., & Vandenheede, H. (2017). Suicide mortality in Belgium at the beginning of the 21st century: Differences according to migrant background. European Journal of Public Health, 27(1), 111–116. https://doi.org/10.1093/eurpub/ckw159Bos, V., Kunst, A. E., Garssen, J., & Mackenbach, J. P. (2007). Duration of residence was not consistently related to immigrant mortality. Journal of Clinical Epidemiology, 60(6), 585–592. https://doi.org/10.1016/j.jclinepi.2006.08.010Bostean, G. (2013). Does selective migration explain the Hispanic paradox? A comparative analysis of Mexicans in the US and Mexico. Journal of Immigrant and Minority Health, 15(3), 624–635. https://doi.org/10.1007/s10903-012-9646-yBrazil, N. (2017). Spatial variation in the Hispanic paradox: Mortality rates in new and established Hispanic US destinations. Population Space and Place, 23(1), 17. https://doi.org/10.1002/psp.1968Castaneda, H., Holmes, S. M., Madrigal, D. S., Young, M. E. D., Beyeler, N., & Quesada, J. (2015). Immigration as a Social Determinant of Health. In J. E. Fielding (Ed.), Annual review of public health, Vol 36 (Vol. 36) (pp. 375–392). Palo Alto: Annual Reviews.Chiswick, B. R., Lee, Y. L., & Miller, P. W. (2008). Immigrant selection systems and immigrant health. Contemporary Economic Policy, 26(4), 555–578. https://doi.org/10.1111/j.1465-7287.2008.00099.xClark, S. J., Collinson, M. A., Kahn, K., Drullinger, K., & Tollman, S. M. (2007). Returning home to die: Circular labour migration and mortality in South Africa. Scandinavian Journal of Public Health, 35, 35–44. https://doi.org/10.1080/14034950701355619Corijn, M., & Lodewijckx, E. (2009). De start van de gezinsvorming bij de Turkse en Marokkaanse tweede generatie in het Vlaams gewest. Een analyse op basis van Rijksregistergegevens. Studiedienst van de Vlaamse Regering.Deboosere, P., & Fiszman, P. (2009). De la persistance des inégalités socio‐spatiales de santé. Le cas Belge. Espace, Populations, Sociétés, 2009(1), 149–158.Deboosere, P., & Gadeyne, S. (2002). Can regional patterns of mortality in Belgium be explained by individual socio‐economic characteristics? Reflets et perspectives de la vie économique, XLI(4), 87–103.Deboosere, P., & Gadeyne, S. (2005). La sous‐mortalité des immigrés adultes en Belgique: une réalité attestée par les recensements et les registres. Population, 60(5), 765–811.Duchene, J., & Thiltgès, E. (1993). La mortalité des plus de 15 ans en Belgique: les disparités régionales en 1985–1987. Espace, Populations, Sociétés, 1, 61–74.Eggerickx, T., Sanderson, J. P., & Vandeschrick, C. (2018). Les inégalités sociales et spatiales de mortalité en Belgique: 1991–2016. Espace Populations Sociétés, 2018(1–2), 1–37.Fuller‐Thomson, E., Brennenstuhl, S., Cooper, R., & Kuh, D. (2015). An investigation of the healthy migrant hypothesis: Pre‐emigration characteristics of those in the British 1946 birth cohort study. Canadian Journal of Public Health‐Revue Canadienne De Sante Publique, 106(8), E502–E508. https://doi.org/10.17269/cjph.106.5218Gadeyne, S. (2006). The ultimate inequality: Socio‐economic differences in all‐cause and cause‐specific mortality in Belgium in the first part of the 1990s (Vol. 39). Brussels & The Hague: NIDI‐CBGS Publications.Gimeno‐Feliu, L. A., Calderon‐Larranaga, A., Diaz, E., Laguna‐Berna, C., Poblador‐Plou, B., Coscollar‐Santaliestra, C., & Prados‐Torres, A. (2019). The definition of immigrant status matters: Impact of nationality, country of origin, and length of stay in host country on mortality estimates. BMC Public Health, 19, 8. https://doi.org/10.1186/s12889-019-6555-1Grimmeau, J.‐P., Deboosere, P., Eggerickx, T., Gadeyne, S., Hermia, J.‐P., Marissal, P., … Willaert, D. (2015). Atlas van België: Bevolking. Gent: Academia Press.Guillot, M., Khlat, M., Elo, I., Solignac, M., & Wallace, M. (2018). Understanding age variations in the migrant mortality advantage: An international comparative perspective. PLoS ONE, 13(6), 17. https://doi.org/10.1371/journal.pone.0199669Hamilton, T. G. (2015). The healthy immigrant (migrant) effect: In search of a better native‐born comparison group. Social Science Research, 54, 353–365. https://doi.org/10.1016/j.ssresearch.2015.08.008Helgesson, M., Johansson, B., Nordquist, T., Vingard, E., & Svartengren, M. (2019). Healthy migrant effect in the Swedish context: A register‐based, longitudinal cohort study. BMJ Open, 9(3), 10. https://doi.org/10.1136/bmjopen-2018-026972Hofmann, E. T. (2012). The burden of culture? Health outcomes among immigrants from the former Soviet Union in the United States. Journal of Immigrant and Minority Health, 14(2), 315–322. https://doi.org/10.1007/s10903-010-9436-3Ichou, M., & Wallace, M. (2019). The healthy immigrant effect: The role of educational selectivity in the good health of migrants. Demographic Research, 40, 61–93. https://doi.org/10.4054/DemRes.2019.40.4Jervelund, S. S., Malik, S., Ahlmark, N., Villadsen, S. F., Nielsen, A., & Vitus, K. (2017). Morbidity, self‐perceived health and mortality among non‐Western immigrants and their descendants in Denmark in a life phase perspective. Journal of Immigrant and Minority Health, 19(2), 448–476. https://doi.org/10.1007/s10903-016-0347-9Keenan, K., Kulu, H., & Cox, F. (2021). Editorial introduction: Social and spatial inequalities in health and mortality: The analysis of longitudinal register data from selected European countries. Population, Space and Place.Kennedy, S., Kidd, M. P., McDonald, J. T., & Biddle, N. (2015). The healthy immigrant effect: Patterns and evidence from four countries. Journal of International Migration and Integration, 16(2), 317–332. https://doi.org/10.1007/s12134-014-0340-xKesteloot, C. (1985). De geografische spreiding van buitenlandse minderheden in België. Beschrijvingen en verklaringen. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen & Amsterdam: De Nederlandsche Boekhandel.Kibele, E., Scholz, R., & Shkolnikov, V. M. (2008). Low migrant mortality in Germany for men aged 65 and older: Fact or artifact? European Journal of Epidemiology, 23(6), 389–393. https://doi.org/10.1007/s10654-008-9247-1Kravdal, O. (2009). Mortality effects of average education in current and earlier municipality of residence among internal migrants, net of their own education. Social Science & Medicine, 69(10), 1484–1492. https://doi.org/10.1016/j.socscimed.2009.08.035Lievens, J. (2000). The third wave of immigration from Turkey and Morocco: Determinants and characteristics. In R. Lesthaeghe (Ed.), Communities and generations. Turkish and Moroccan populations in Belgium (Vol. 36). Brussels & The Hague: NIDI/CBGS Publications.Martens, A. (1985). Het na‐oorlogs immigratiebeleid. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen & Amsterdam: De Nederlandsche Boekhandel.Martinez, J. N., Aguayo‐Tellez, E., & Rangel‐Gonzalez, E. (2015). Explaining the Mexican‐American health paradox using selectivity effects. International Migration Review, 49(4), 878–906. https://doi.org/10.1111/imre.12112Mehta, N. K., & Elo, I. T. (2012). Migrant selection and the health of US immigrants from the former Soviet Union. Demography, 49(2), 425–447. https://doi.org/10.1007/s13524-012-0099-7Myria. (2019). Migratie in cijfers en in rechten. Brussel: Federaal Migratiecentrum.Norredam, M., Agyemang, C., Hansen, O. K. H., Petersen, J. H., Byberg, S., Krasnik, A., & Kunst, A. E. (2014). Duration of residence and disease occurrence among refugees and family reunited immigrants: Test of the 'healthy migrant effect' hypothesis. Tropical Medicine & International Health, 19(8), 958–967. https://doi.org/10.1111/tmi.12340Norredam, M., Hansen, O. H., Petersen, J. H., Kunst, A. E., Kristiansen, M., Krasnik, A., & Agyemang, C. (2015). Remigration of migrants with severe disease: Myth or reality?—A register‐based cohort study. European Journal of Public Health, 25(1), 84–89. https://doi.org/10.1093/eurpub/cku138Rasulo, D., Spadea, T., Onorati, R., & Costa, G. (2012). The impact of migration in all‐cause mortality: The Turin longitudinal study, 1971‐2005. Social Science & Medicine, 74(6), 897–906. https://doi.org/10.1016/j.socscimed.2011.10.045Renard, F., Tafforeau, J., & Deboosere, P. (2015). Mapping the cause‐specific premature mortality reveals large between‐districts disparity in Belgium, 2003‐2009. Archives of Public Health, 73, 18. https://doi.org/10.1186/s13690-015-0060-5Reniers, G. (1999). On the history and selectivity of Turkish and Moroccan migration to Belgium. International Migration, 37(4), 679–713. https://doi.org/10.1111/1468-2435.00090Reus‐Pons, M., Vandenheede, H., Janssen, F., & Kibele, E. U. B. (2016). Differences in mortality between groups of older migrants and older non‐migrants in Belgium, 2001‐09. European Journal of Public Health, 26(6), 992–1000. https://doi.org/10.1093/eurpub/ckw076Riva, M., Curtis, S., & Norman, P. (2011). Residential mobility within England and urban‐rural inequalities in mortality. Social Science & Medicine, 73(12), 1698–1706. https://doi.org/10.1016/j.socscimed.2011.09.030Singer, J., & Willett, J. (2003). Applied longitudinal data analysis. Modeling change and event occurrence. Oxford: Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195152968.001.0001Stols, E. (1985). Emigratie en immigratie in historich perspectief. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen/Amsterdam: Uitgeverij De Nederlandsche Boekhandel.Syse, A., Dzamarija, M. T., Kumar, B. N., & Diaz, E. (2018). An observational study of immigrant mortality differences in Norway by reason for migration, length of stay and characteristics of sending countries. BMC Public Health, 18, 12. https://doi.org/10.1186/s12889-018-5435-4UNIA, & Fod WASO. (2017). Socio‐economische monitoring 2017: Arbeidsmarkt en origine. Brussel: Interfederaal Gelijkekansencentrum en Federale Overheidsdienst Werkgelegenheid en Sociaal Overleg.Urquia, M. L., O'Campo, P. J., & Heaman, M. I. (2012). Revisiting the immigrant paradox in reproductive health: The roles of duration of residence and ethnicity. Social Science & Medicine, 74(10), 1610–1621. https://doi.org/10.1016/j.socscimed.2012.02.013Vallin, J., Meslé, F., & Valkonen, T. (2001). Trends in mortality and differential mortality. Strassbourg: Council of Europe Publishing.Van Hemelrijck, W. M. J., de Valk, H. A. G., & Vandenheede, H. (2016). Cancer mortality by migrant background in the 2000s in Belgium: Patterns and determinants. European Journal of Public Health, 26, 383–383.Van Imhoff, E., & Keilman, N. (1991). Lipro 2.0: An application of a dynamic demographic projection model to household structure in the Netherlands (Vol. 23). The Hague & Brussels: NIDI‐CBGS.Vandenheede, H., Willaert, D., De Grande, H., Simoens, S., & Vanroelen, C. (2015). Mortality in adult immigrants in the 2000s in Belgium: A test of the ‘healthy‐migrant’ and the ‘migration‐as‐rapid‐health‐transition’ hypotheses. Tropical Medicine & International Health, 20(12), 1832–1845. https://doi.org/10.1111/tmi.12610Vanthomme, K., & Vandenheede, H. (2019). Trends in Belgian cause‐specific mortality by migrant origin between the 1990s and the 2000s. BMC Public Health, 19, 16. https://doi.org/10.1186/s12889-019-6724-2Wallace, M., & Darlington‐Pollock, F. (2020). Poor health, low mortality? Paradox found among immigrants in England and Wales. Population, Space and Place, e2360. https://doi.org/10.1002/psp.2360Wallace, M., Khlat, M., & Guillot, M. (2019). Mortality advantage among migrants according to duration of stay in France, 2004‐2014. BMC Public Health, 19, 9. https://doi.org/10.1186/s12889-019-6652-1Wallace, M., & Kulu, H. (2014). Low immigrant mortality in England and Wales: A data artefact? Social Science & Medicine, 120, 100–109. https://doi.org/10.1016/j.socscimed.2014.08.032Wallace, M., & Kulu, H. (2018). Can the salmon bias effect explain the migrant mortality advantage in England and Wales? Population Space and Place, 24(8), 18. https://doi.org/10.1002/psp.2146Weitoft, G. R., Gullberg, A., Hjern, A., & Rosén, M. (1999). Mortality statistics in immigrant research: Method for adjusting underestimation of mortality. International Journal of Epidemiology, 28, 756–763. https://doi.org/10.1093/ije/28.4.756Wilson, B., Drefahl, S., Sasson, I., Henery, P. M., & Uggla, C. (2020). Regional trajectories in life expectancy and lifespan variation: Persistent inequality in two Nordic welfare states. Population, Space and Place, 26(e2378). https://doi.org/10.1002/psp.2378AAPPENDIXA1TableSocio‐demographic and socio‐economic profile of native and migrant women aged 25–75 by migrant generation and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):DeathsBelgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7][N]Age (time‐varying)‐ Mean age51.4940.5432.2540.3347.3854.5457.9645,565Household position [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ SING14.3511,079.2013.9510.1113.4116.4612,734‐ MAR028.4410.193.9712.9512.5917.6628.3417,341‐ MAR+30.7435.6634.8340.9950.7947.0735.046,507‐ CMAR2,839.2715.440.660.230.120.53353‐ UNM05,134.925.155.822.101.941.981,882‐ UNM+6,829.389.007.195.042.772.30767‐ CUNM0,180.420.930.110.020.020.0219‐ H1PA8.1712.8611.109.6816.0014.9612.163,528‐ C1PA1.593.956.690.430.200.170.72579‐ NFR0.580.941.835.951.820.801.02699‐ OTHER1.111.331.832.241.081.041.40886‐ COLL0.060.020.030.030.020.030.05270Activity status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Student1.484.3510.472.060.230.070.0442‐ Employed58.3761.3247.4140.2746.0440.2630.549,680‐ Homekeepers20.1320.4629.3852.6742.7546.2971.7416,412‐ Pensioner16.082.540.220.983.296.6019.9618,117‐ Never worked0.151.181.730.040.030.030.2939‐ Unemployed9.7910.1610.783.987.676.767.431,275Income decile [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ 1st decile (low)0.831.924.269.945.375.461.42983‐ 2nd decile0.821.934.569.985.405.361.35964‐ 3rd decile1.683.328.3710.796.325.832.301,185‐ 4th decile16.1317.7425.7027.1627.2328.6533.5211,073‐ 5th decile16.9619.4916.8815.0720.2220.5022.289,822‐ 6th decile12.3611.9010.468.119.969.6013.658,280‐ 7th decile12.0511.7110.367.478.917.989.655,109‐ 8th decile13.0513.069.934.637.016.736.623,203‐ 9th decile14.0511.276.223.414.454.614.802,850‐ 10th decile (high)12.067.653.263.445.145.294.412,096N deaths40,4221,134667936284572,06545,565N individuals1,990,659143,10816,858155,47766,43130,75386,3732,489,659N person‐years9,702,778765,33999,637667,120314,227145,723401,68912,096,513Ownership status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Home owner81.2669.9754.7836.7358.5068.8372.6331,480‐ Renter18.7430.0345.2263.2741.5031.1727.3714,085Dwelling with bath or shower [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available0.801.170.971.240.790.710.90665‐ Bath/shower99.2098.8399.0398.7699.2199.2999.1044,900Dwelling with central heating [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available12.8414.4912.3310.3310.2310.0714.248,040‐ Central heating87.1687.5187.6789.6789.7789.9385.7637,525Type of building [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ 1 dwelling84.9775.2758.2645.6860.3768.5472.5234,325‐ 2 dwellings2.724.325.875.924.634.675.071,607‐ 3+ dwellings12.0520.1535.5748.0934.6726.5122.169,516‐ Nonresidential0.260.260.300.320.340.290.25117Number of rooms per inhabitant [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ less than 0.500.070.371.031.520.830.460.3346‐ 0.50–0.991.215.1313.1111.8111.036.304.70537‐ 1.00–1.245.2811.5718.7116.5816.5112.488.901,439‐ 1.25–1.498.1612.8814.4812.7413.2210.798.331,722‐ 1.50–1.9918.7422.5819.4416.2918.6419.5415.724,300‐ 2.00–2.4919.0319.1015.5216.0015.0516.9517.257,430‐ 2.50–2.9913.278.915.646.996.888.6012.077,018‐ 3.00 or more34.2519.4612.0618.0617.8424.8832.7123,073Construction period of dwelling [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Before 191917.1828.8525.7826.2321.7524.4531.759,452‐ 1919–194512.3714.7616.5515.1415.3715.7015.415,890‐ 1946–196010.4012.0114.1013.0514.0012.7311.374,897‐ 1961–197010.808.8611.8912.8512.1910.668.846,570‐ 1971–198017.0810.5111.8812.2513.3112.6514.109,435‐ 1981–199011.507.075.424.555.969.267.203,925‐ 1991–200012.569.307.026.7710.4710.047.203,459‐ 2001–20054.074.023.314.413.942.632.271,070‐ 2006 or later4.054.634.064.763.011.881.86867N deaths40,4221,13466793628457206545,565N individuals1,990,659143,10816,858155,47766,43130,75386,3732,489,659N person‐years9,702,778765,33999,637667,120314,227145,723401,68912,096,513Note: Legend Household Positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child of a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors.A2TableSocio‐demographic and socio‐economic profile of native and migrant men aged 25–75 by migrant generation and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):DeathsBelgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7][N]Age (time‐varying)‐ Mean age50.7440.1631.3341.4548.4355.5958.99Household position [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ SING14.5916.7517.5323.0515.6514.3712.5819,371‐ MAR027.309.553.4112.2812.0117.9632.0331,837‐ MAR+31.7530.5923.3538.9258.1656.3943.2612,510‐ CMAR5.6515.5625.641.140.360.191.081,049‐ UNM05.675.474.955.772.632.512.493,174‐ UNM+7.149.387.087.625.703.382.941,503‐ CUNM0.320.701.320.160.040.010.0373‐ H1PA2.182.000.801.672.313.122.481,765‐ C1PA3.196.7110.550.670.330.231.201,813‐ NFR0.671.051.973.950.980.690.83866‐ OTHER1.472.193.334.731.811.111.031,470‐ COLL0.060.050.080.060.030.040.04312Activity status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ student1.564.099.681.730.100.060.0189‐ employed69.2070.5159.2357.2465.2760.3844.0021,998‐ Homekeepers6.4711.6119.1733.0519.9821.4815.5812,545‐ pensioner19.573.710.071.513.697.8332.8738,295‐ never worked0.110.610.720.030.010.000.0641‐ unemployed3.089.4711.136.4410.9610.247.472,775Income decile [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ 1st decile (low)0.882.325.469.294.444.300.931,186‐ 2nd decile0.882.345.359.274.354.440.921,186‐ 3rd decile1.343.728.559.424.393.981.091,204‐ 4th decile3.388.2013.1813.1012.7112.616.663,148‐ 5th decile6.829.7610.9310.7613.6015.4116.7511,708‐ 6th decile10.609.169.349.4010.5910.9921.1618,270‐ 7th decile13.9012.8213.4712.0213.5611.5919.9315,577‐ 8th decile16.8118.3316.5511.4314.6412.7212.269,139‐ 9th decile19.9817.9611.257.0910.4711.0810.237,199‐ 10th decile (high)25.4115.385.918.2211.2412.8810.087,126N deaths66,5712,206971,2699216644,01575,743N individuals1,939,833149,09615,684146,66459,07925,50189,0952,424,952N person‐years9,441,797794,87494,502597,194275,607119,067407,61711,730,658Ownership status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Home owner82.8372.9655.0232.4557.4470.5576.7253,316‐ Renter17.1727.0444.9867.5542.5629.4523.2822,427Dwelling with bath or shower [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available1.051.281.111.341.020.861.021,627‐ Bath/shower98.9598.7298.8998.6698.9899.1498.9874,116Dwelling with central heating [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available13.6715.1311.8610.5310.7010.4314.3015,055‐ Central heating86.3384.8788.1489.4789.3089.5785.7060,688Type of building [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ 1 dwelling85.8676.1958.4343.9057.6166.9675.0058,716‐ 2 dwellings2.764.626.196.305.235.205.152,629‐ 3 + dwellings11.0618.9235.0549.4136.8627.5119.6014,133‐ Nonresidential0.320.270.330.400.290.320.26265Number of rooms per inhabitant [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ less than 0.500.070.401.211.851.040.650.39119‐ 0.50–0.991.244.8012.1211.7613.099.275.66989‐ 1.00–1.245.6111.3817.6816.1818.7614.6910.432,868‐ 1.25–1.498.4512.2013.3011.8212.4810.899.072,978‐ 1.50–1.9919.3822.1518.0215.1516.4018.2316.657,486‐ 2.00–2.4919.3818.6315.9015.2212.6215.2617.0812,623‐ 2.50–2.9912.638.135.616.215.627.4812.0712,013‐ 3.00 or more33.2422.3116.1721.8119.9823.5228.6536,667Construction period of dwelling [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Before 191918.0130.9927.0128.6524.3925.8631.7517,669‐ 1919–194512.6914.8016.4815.0415.9716.4515.6010,456‐ 1946–196010.6411.4513.9512.4313.9412.4711.018,036‐ 1961–197010.078.4111.2812.3011.8810.448.239,843‐ 1971–198016.3710.1611.9211.5312.4412.1314.2115,372‐ 1981–199011.746.975.604.335.108.177.406,283‐ 1991–200012.549.156.996.539.569.757.575,110‐ 2001–20053.973.882.994.373.712.672.301,591‐ 2006 or later3.984.203.774.823.012.071.921,383N deaths66,5712,206971,2699216644,01575,743N individuals1,939,833149,09615,684146,66459,07925,50189,0952,424,952N person‐years9,441,797794,87494,502597,194275,607119,067407,61711,730,658Note: Legend Household Positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child of a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png "Population, Space and Place" Wiley

Spatial variation of migrant‐native mortality differentials by duration of residence in Belgium: A story of partial convergence

Download PDF
20 pages

Loading next page...
 
/lp/wiley/spatial-variation-of-migrant-native-mortality-differentials-by-LZDfyIH3OE

References (55)

Publisher
Wiley
Copyright
© 2022 John Wiley & Sons, Ltd.
ISSN
1544-8444
eISSN
1544-8452
DOI
10.1002/psp.2498
Publisher site
See Article on Publisher Site

Abstract

INTRODUCTIONSignificant mortality advantages have repeatedly been documented for migrants compared to native born populations, although migrants in many cases face socio‐economic deprivation, show higher morbidity levels, and experience barriers in health care use relative to natives (Jervelund et al., 2017; Urquia et al., 2012; Wallace & Darlington‐Pollock, 2020). A systematic review and meta‐analysis of 96 studies, predominantly pertaining to international migrants in high income countries, finds a migrant all‐cause mortality advantage in 75% of the reported standardised mortality ratios. This advantage persists for the majority of cause groups, with the exception of infectious diseases (viral hepatitis, tuberculosis, and HIV) and external causes (assaults and deaths of undetermined cause) where migrants show higher mortality compared to the general population in destination countries (Aldridge et al., 2018). The size and direction of this healthy migrant effect are subject to considerable variation, depending on the migrant groups included, the definitions used, and the extent to which duration of residence, socio‐economic position in the destination country, or level of acculturation are taken into account (Gimeno‐Feliu et al., 2019; Hamilton, 2015; Kennedy et al., 2015).Several mechanisms have been suggested to account for the mortality advantage in various contexts. The data artefact hypothesis suggests that the mortality advantage can be entirely accounted for by unregistered (r)emigration leading to an under‐registration of deaths and an overestimation of the population at risk. Although research suggests that the impact of unregistered migration may be sizeable in specific contexts (Kibele et al., 2008), it seems unlikely that unregistered and/or selective emigration can fully account for the mortality advantage across different contexts (Deboosere & Gadeyne, 2005). An alternative mechanism that has been suggested to account for the mortality advantage is that migrants are positively selected in terms of health status, giving rise to an initial mortality advantage compared to natives which is largest shortly after immigration (healthy migrant hypothesis). The strength of the initial health selection has been shown to vary by migration motive and to be stronger in case of labour and study migration compared to family reunification (Guillot et al., 2018). Convergence between the mortality patterns of foreign‐born and native‐born populations is expected to take place with increasing duration of residence due to the waning of the initial selection effects and the dynamic frailty composition of both groups. Convergence may accelerate as a result of exposure to adverse conditions that migrants often face in destination countries. Conversely, convergence may be (partially) offset by the preservation of favourable health behaviours characteristic for the country of origin or selective emigration taking place in case of deteriorating health (salmon bias hypothesis).Acculturation of mortality patterns has frequently been studied by considering convergence of mortality risks with increasing duration of residence (Guillot et al., 2018; Vandenheede et al., 2015; Wallace et al., 2019). The spatial dimension of this convergence process has hitherto received limited attention, although several arguments suggest that it may be particularly relevant to understand the origin of the mortality advantage. First, several studies have shown that all‐cause mortality is subject to significant, and often persistent, spatial variation within countries (Deboosere & Gadeyne, 2002; Eggerickx et al., 2018; Wilson et al., 2020). Second, specific immigrant groups tend to settle in specific areas (Reniers, 1999), where local conditions may differentially affect the acculturation process and the evolution of mortality risks over time. This study contributes to the available body of literature that has overwhelmingly compared migrant mortality to the average of the entire native‐born population by taking a subnational spatial approach and comparing migrants to natives living in the same area, because their health and mortality risks are influenced by these local conditions. We consider whether migrant‐native differentials are subject to spatial variation and document whether, and to what extent, such differentials fade with increasing duration of residence in Belgium, allowing for variation in terms of migrant generation. This paper is part of a special issue on ‘Social Inequalities in Health and Mortality: The Analysis of Longitudinal Register Data from Selected European Countries’. Belgium is characterised by substantial spatial variation in all‐cause mortality and has a sizeable migrant population for which population‐wide register data allow us to address spatial variation in the mortality advantage. More specifically, this paper uses longitudinal microdata from the 2011 Belgian census which have been linked to longitudinal microdata drawn from the population register and tax return registers for the period 2001–2015.MIGRANT MORTALITY ADVANTAGE: UNDERLYING MECHANISMS AND PREVIOUS FINDINGSSeveral explanations have been suggested for the migrant mortality advantage, mainly focusing on five mechanisms: (i) selective immigration in terms of health status (healthy migrant effect), (ii) unregistered (r)emigration leading to an underestimation of death rates in migrant populations (data artefacts), (iii) selective emigration in terms of health status leading to an under‐registration of deaths (salmon bias), (iv) migration as a rapid health transition, and (v) foreign born retaining behaviours from their country of origin that positively influence their health (cultural effects) (Guillot et al., 2018; Keenan et al., 2021).Concerns about the reliability of demographic data on mortality of migrants suggests that data artefacts may (partially) account for lower mortality in migrant populations (Kibele et al., 2008). To the extent that departures from the host country are unregistered and deaths subsequently take place abroad, migrants become statistically immortal and migrant mortality is underestimated. Although studies for specific contexts have shown that data artefacts may contribute substantially to the migrant mortality advantage (Weitoft et al., 1999), other studies that have conducted sensitivity analyses on the impact of entry and exit uncertainty show that the mortality advantage cannot in general be attributed to data artefacts across contexts (Wallace & Kulu, 2014). Considering spatial variation within countries, data artefacts are unlikely to contribute to spatial variation in the mortality advantage unless vital registration is strongly decentralised and subject to differential bias across regions.Underestimation of migrant mortality may result from selective out‐migration in terms of health status and mortality risks. This mechanism is frequently referred to as the salmon bias, assuming that migrants return to their country of origin in anticipation of subsequent death (Andersson & Drefahl, 2017; Arenas et al., 2015; Bostean, 2013; Clark et al., 2007). The contribution of selective out‐migration to the mortality advantage has been notoriously difficult to assess, as it requires the longitudinal follow‐up of migrants after having resettled in their respective countries of origin. As register data are typically confined to one specific country context, this option can rarely be pursued in practice. Focusing on long‐distance migration within Sweden, Andersson and Drefahl (2017) find no evidence for the selection of healthy migrants moving from the North to the South, but do find evidence for a salmon bias in terms of an elevated mortality risk among return migrants to Northern Sweden. In contrast, studies in other settings have shown that the probability of (r)emigration is effectively lower among migrants suffering chronic disease (Norredam et al., 2015), that mortality levels would have to be unrealistically high among (r)emigrants to account for the mortality advantage (Deboosere & Gadeyne, 2005; Vandenheede et al., 2015), and that adjusting age‐specific mortality rates in groups where a salmon bias has been established cannot account for the mortality advantage (Wallace & Kulu, 2018). A further argument against the salmon bias hypothesis is that the quality and accessibility of health care in receiving countries in most cases outperforms systems and provisions in sending countries, particularly for groups for which large mortality advantages have been established.Migrants are not representative of the population in the country of origin but are self‐selected in terms of having a better health status (depending on their motive for migration), frequently referred to as the healthy migrant effect. Immigration into the destination country may in some cases involve explicit screening in terms of health status, and the profile of immigrants may be highly selective in terms of characteristics that have a moderating effect on mortality risks (Bostean, 2013; Brazil, 2017; Castaneda et al., 2015; Chiswick et al., 2008; Fuller‐Thomson et al., 2015; Hamilton, 2015; Helgesson et al., 2019; Hofmann, 2012; Ichou & Wallace, 2019; Kennedy et al., 2015; Martinez et al., 2015; Mehta & Elo, 2012). In the case of health selection, the all‐cause mortality advantage is expected to be largest for migrants with shorter durations of residence, because the contrast with the health status of the native population in the destination country is expected to be largest shortly after the health selection has taken place. With increasing duration of residence, the migrant‐native differential is expected to wear off. The selection in terms of health status and the size of the mortality advantage have been suggested to vary with the migration motive, being stronger among individuals immigrating for education and work, and less so among refugees and individuals immigrating in view of family formation or family reunification. This results in variation of the health selection effect by gender (less articulated selection among women than men, particularly in older migration cohorts), age (less selection among children and elderly persons), and country of origin (to the extent that migration motives are correlated with origin; Guillot et al., 2018). Considering disease incidence among refugees and family reunited migrants, Norredam et al. (2014) analyse disease incidence by duration of residence and find lower hazard ratios of stroke and breast cancer within 5 years after arrival, but hazard ratios increase with longer follow‐up. In contrast, hazard ratios for ischemic heart disease and diabetes were higher within 5 years after arrival and increased further with duration of residence. Results therefore suggest that the healthy migrant effect (positive health selection) should be used with caution when explaining migrants' advantageous health outcomes (Norredam et al., 2014).The conditions that immigrants are exposed to in receiving countries are often more precarious than those faced by the native population, suggesting acculturation or assimilation of mortality risks. A considerable amount of empirical evidence using different markers of socio‐economic status—education, income, activity status, or housing characteristics—shows that mortality levels are consistently higher among individuals having a more precarious socio‐economic position (Gadeyne, 2006; Vallin et al., 2001). Various pathways connect socio‐economic position to health outcomes and mortality, including life styles, environmental factors, health behaviours (e.g., preventive health care), and access to health care systems. As migrants on average have a more precarious socio‐economic position than natives, the exposure to adverse living conditions is expected to gradually override their initial health advantage and foster convergence to the mortality pattern of natives or even a reversal of the migrant‐native differential in all‐cause mortality with increasing duration of residence. The speed at which convergence, and potentially a cross‐over, takes place is determined by several other factors which may be subject to variation between local settings.Finally, the theory of migration as a rapid health transition suggests that migrants from low‐income countries immediately experience a drop in mortality from infectious disease after migration as a result of access to better health care in the destination country compared with the origin country. As an increase from chronic disease mortality is not expected to follow immediately, the mortality advantage may persist over an extended period of time (Vandenheede et al., 2015; Vanthomme & Vandenheede, 2019). This suggests that convergence between spatial mortality patterns of migrants and natives can only be expected to take place after extended durations of residence. The speed of convergence in mortality patterns will depend on the extent to which initial beneficial health behaviours are preserved, or whether rapid assimilation takes place with respect to (local) risk factors for the main causes of death (cardiovascular disease, neoplasms, and cerebrovascular disease).Whereas several mechanisms have been suggested to account for the migrant mortality advantage, the implications of these mechanisms for spatial variation in migrant‐native mortality differentials have hitherto received limited attention. Literature suggests that health selection is strongly related to age and migration motive, but it is unclear whether health selection is regionally differentiated in such a way that it would give rise to spatial variation of all‐cause mortality in migrant populations that is similar to spatial patterns found in natives. To the extent that regional variation in health selection is limited or different from spatial variation of all‐cause mortality in natives, the migrant mortality advantage may be subject to strong variation between regions. Local variation in the migrant mortality advantage is expected to diminish with duration of residence due to waning of selection effects and exposure to local conditions, but the theory of migration as a rapid health transition suggests that convergence to the spatial pattern of all‐cause mortality in natives may only emerge gradually in migrant groups with extended durations of residence. In sum, several mechanisms that have been suggested to account for the migrant mortality advantage suggest that migrant‐native differentials may well be subject to variation between regions, suggesting that it would be relevant to adopt a spatial dimension in the analysis of the mortality advantage.SPATIAL VARIATION IN MORTALITYSubstantive spatial variation in all‐cause and cause‐specific mortality has been documented in various contexts, which may represent regional variation in risk factors, regional variation in the composition of the resident population (e.g., in terms of socio‐economic position), and selective internal migration between regions in terms of health status (Keenan et al., 2021; Wilson et al., 2020). In Belgium, substantive spatial variation at both the regional and district level has been documented for all‐cause as well as cause‐specific mortality (Deboosere & Fiszman, 2009; Deboosere & Gadeyne, 2002; Duchene & Thiltgès, 1993; Eggerickx et al., 2018; Grimmeau et al., 2015). Controlling for composition in terms of socio‐economic status, pronounced regional variation persists in Belgium with lower mortality in Flemish districts than Walloon districts, and Brussels taking an intermediate position (Eggerickx et al., 2018; Grimmeau et al., 2015). This spatial pattern of all‐cause mortality is largely shaped by spatial variation in cardiovascular disease, cerebrovascular disease, and cancer which constitute the major causes of death (Grimmeau et al., 2015), although spatial patterns also vary by cause of death. A north–south divide emerges for all‐cause mortality, as well as for mortality from cardiovascular disease in both sexes, cerebrovascular disease in men, diabetes in both sexes, mental and neurological disease in both sexes, chronic obstructive pulmonary disease in men, alcohol‐related deaths in men, and non‐transport related deaths in men. In contrast, a northwest–southeast gradient parallel to the French border emerges for lip, oral cavity, pharynx, larynx, and oesophageal cancers (Renard et al., 2015).The impact of internal migration on spatial variation in all‐cause mortality has been documented in the literature. Using the Office of National Statistics Longitudinal Study, Riva et al. (2011) show that residential mobility between 1981 and 2001 accounts for about 30% of the urban–rural difference in all‐cause mortality observed in 2001–2005; individuals moving between urban and rural areas are found to be relatively healthier than long‐term urban residents (Riva et al., 2011). Using register data for the Norwegian population aged 60–89 in 1991–2002, Kravdal (2009) finds that for individuals who had moved between municipalities once in the preceding 10 years, the current socio‐economic context was not important for their mortality, suggesting that neighbourhood socio‐economic effects need some time to build up and do not dissipate soon after immigration to a new environment. Similarly, using longitudinal data for Turin between 1975 and 2005, Rasulo et al. (2012) show that internal migrants initially face lower mortality risks than locals but that migrant‐native differentials gradually reduce as internal migrants not only accumulate exposure to environmental health hazards similar to natives but also increasingly face the health consequences of worse socio‐economic conditions compared to local individuals.Four groups of factors have been suggested to account for spatial variation mortality patterns: environmental factors, contextual socio‐economic factors, effects of different public health policies and amenities, and behavioural and cultural factors which are unrelated to the individual socio‐economic position (Deboosere & Gadeyne, 2002). The physical, social, family, and institutional environment can also interact with individual characteristics to generate inequality in health and mortality outcomes (Eggerickx et al., 2018).MIGRANT GROUPS IN BELGIUMThroughout the twentieth century, Belgium has increasingly become an immigration country. Whereas the foreign population was limited to 2.1% of the population after the First World War (1920), it increased to 4% in 1938, predominantly as a result of immigration from neighbouring countries—but also Poland and Italy—for the mining, metal, and textile industries (Stols, 1985). Following World War II, migration resumed immediately with active recruitment of foreign workers from Italy (1946–1956), other Southern European countries (predominantly Spain and Greece in 1955–1961), and Turkey and Morocco (1962–1969). Immigration was particularly high in the early 1960s. Due to articulated labour shortages, migration in that period was no longer checked through work permits and migrant workers were allowed to settle in Belgium under tourist visas, with regularisation typically following after medical screening (Martens, 1985). Labour migration decreased substantially following the economic stall in 1967, after which migration was strongly controlled through the selective granting of work permits for predominantly highly educated profiles (Martens, 1985; Reniers, 1999). Following the migration stop in 1974, family reunification of spouses and children who were still in the country of origin gained importance (Reniers, 1999) as well as non‐European marriage migration when the intermediate and second generation reached adolescence in the 1980s and 1990s (Lievens, 2000). Following a period of low immigration from the 1980s up to the mid‐1990s, the late 1990s witnessed an articulated increase of immigration and a diversification of migration motives and origin countries resulting from the free movement of individuals within the European Union (Schengen treaty), continued marriage migration and family reunification, but also student and labour migration, accompanied by period fluctuations in the number of asylum seekers and refugees (Corijn & Lodewijckx, 2009; Myria, 2019).Several studies have documented migrant mortality in Belgium. Using linked microdata from the 1991 census and the population register for the period 1991–1995, Deboosere and Gadeyne (2005) compare all‐cause mortality in natives and different origin groups. With the exception of immigrants from France (significantly higher mortality), Germany (no significant difference), and Luxemburg (no significant difference), migrant men in the age group 25–54 originating from other neighbouring countries, Southern Europe as well as Turkey and Morocco, have significantly lower mortality than natives. Differentials typically increase in favour of migrant men when additional adjustments are made for age, level of education, and housing characteristics (Deboosere & Gadeyne, 2005). Results for women were found to be similar to those for men. Using linked census and register data for 2001–2010, Vandenheede et al. (2015) find that first generation immigrants have lower all‐cause and chronic disease mortality than natives: a mortality advantage that wears off with increasing duration of residence, in line with health selection and acculturation theories. Consistent with the migration as rapid health transition theory, infectious‐disease mortality is higher in specific non‐European migrant groups, but not in other origin groups. Second generation migrants on the other hand suffer a mortality disadvantage relative to natives which disappears when controlling for age, educational attainment, employment status, and housing characteristics. Considering approximately the same period and controlling for age, socio‐economic position, and urban typology, Reus‐Pons et al. (2016) find that the mortality advantage persists after age 50 for most origin groups and that the mortality disadvantage of specific origin groups can be attributed partially to their lower socio‐economic status compared to natives. Similar to other studies considering migrant‐native mortality differentials in Belgium (Bauwelinck et al., 2017; Deboosere & Gadeyne, 2005; Van Hemelrijck et al., 2016; Vandenheede et al., 2015), the study by Reus‐Pons et al. does not consider spatial variation in migrant‐native differentials in all‐cause or cause‐specific mortality.RESEARCH QUESTIONS AND HYPOTHESESPrevious studies have documented extensively that both all‐cause and cause‐specific mortality are subject to considerable regional variation and that acculturation of migrant mortality to native patterns takes place only gradually with increasing duration of residence. This suggests that the spatial pattern of all‐cause mortality is likely to differ between migrants and natives and that the migrant‐mortality advantage is therefore subject to variation between local settings. However, this spatial dimension to the migrant mortality advantage has not been explored in previous research. This paper introduces a spatial dimension into the analysis of the migrant mortality advantage and has three aims:First, we assess whether the migrant mortality advantage documented previously using the 1991 and 2001 Belgian censuses persists up to 2011–2015, controlling for socio‐demographic and socio‐economic characteristics. We distinguish between migrant generations and differentiate by duration of residence as we expect that the mortality risk in migrant populations will converge to natives with length of stay, assuming that exposure to local conditions and adoption of related behaviours play an important role in the acculturation of mortality (Bos et al., 2007; Vandenheede et al., 2015; Hypothesis 1).Second, controlling for socio‐demographic and socio‐economic characteristics, we consider spatial variation in migrant mortality. In line with health selection and acculturation theories, we expect that spatial variation of all‐cause mortality in recent migrants will be dissimilar from natives. We expect that migrant mortality will converge to the spatial pattern of natives among first generation migrants with extended durations of residence and groups where health selection can be largely presumed absent such as the second generation (Hypothesis 2).Third, we consider spatial variation in the migrant mortality advantage and test whether the overall mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population and their concentration in local settings characterised by high native mortality and large migrant‐native differentials in all‐cause mortality (Hypothesis 3).DATA AND METHODSDataThe analyses of all‐cause mortality use microdata from the 2011 Belgian census which have been linked to the population register both retrospectively for the period 2001–2010 and prospectively for the period 2011–2015 using a coded individual identifier. The 2011 census is register‐based and provides detailed information on individual and household characteristics on January 1, 2011, drawn from the population register, social security registers, the address and dwelling register, and tax return data. We focus on all‐cause mortality among individuals aged 25 to 75 between January 1, 2011, and December 31, 2015. With respect to mortality, the exact date of death is available from the population register. With respect to emigration and accurately detecting events that cause truncation, the situation is somewhat more complex. Registration of emigration can be considered (quasi‐)universal, but the timing of the registration may in certain cases be late, thus overestimating exposure to some degree. Individuals may in principle leave the country to return to their country of origin or move abroad without notifying authorities of their emigration (Deboosere & Gadeyne, 2005). However, a large share of emigrants and their household members are estimated to communicate their new address abroad in order to be able to benefit from social security entitlements and other rights that they themselves or their relatives have accumulated in Belgium (through labour, social security, or the health system). Those who do not actively report their emigration will with some delay be detected through a variety of mechanisms, such as failed contact attempts by administrations or utility companies, address conflicts in the population register when new occupants move into the former dwelling of emigrated households, and follow‐up of unpaid bills. After the unreported leave is detected, the municipality administration will make the necessary corrections to the population register, but it is unclear to what degree the newly recorded emigration will bear a correct date. Any deviations from reality can be expected to be selectively late. However, late recording of emigrations is unlikely to contribute substantially to the migrant mortality advantage in Belgium, as proportions (r)emigrating—and within this group proportions doing so without reporting it—are unlikely to be large enough to explain (significant parts of) the observed mortality advantage. As the data only cover the population officially residing in Belgium, we cannot test directly whether migrants are positively selected for health status. Individuals are censored on December 31, 2015, in case they survived throughout the observation period, or in case of emigration, censored on June 30 of the year in which they left the country as the exact date of emigration was not recorded.ModelConsistent with the construction of synthetic or period life tables, both the descriptive results and the multivariate analyses use a late entry design where exposure‐to‐risk and mortality occurring in the observation window from January 1, 2011, until December 31, 2015, are appropriately incorporated into the estimation to generate the synthetic or period hazard function of all‐cause mortality by age (Singer & Willett, 2003). Cox regression models were used to estimate the effect of covariates on the continuous‐time hazard of all‐cause mortality (Singer & Willett, 2003). The analyses of migrant‐native differentials in all‐cause mortality incrementally control for age (baseline) (Model 1), living arrangements (Model 2), socio‐economic position (Model 3), and housing characteristics (Model 4). Subsequently, we introduce district as a covariate (Models 5a–5c) to discern spatial variation in all‐cause mortality at the district‐level net of spatial variation in the composition of the resident population in terms of characteristics considered in Models 1 through 4 which have been shown to affect mortality (Gadeyne, 2006). Model 5a re‐estimates Model 4 using a collapsed version of the typology of migrant background as a reference model for subsequent models including spatial variation. Model 5b introduces district as categorical covariate into an additive model with the typology of migrant background, constraining spatial variation in all‐cause mortality to be identical across natives and migrant groups. Finally, Model 5c includes the interaction between the typology of migrant groups and district, to assess whether spatial variation in all‐cause mortality differs between migrants and natives and whether spatial variation of all‐cause mortality in migrant groups converges to the pattern found in the native population with increasing duration of residence:lnhtij=lnh0tj+∑βk.Mki+∑βk.Dki+∑βk.LAkij+∑βk.SEPkij+∑βk.HCki,where h0(tj) is the unspecified baseline hazard function of all‐cause mortality by age, Mki is a set of dummy variables distinguishing migrant groups from natives (reference category) by migrant generation and duration of residence, Dki is a set of dummy variables comparing the mortality risk between districts, LAkij is a set of dummy‐variables denoting the living arrangement on January 1 in each observation year t, SEPkij is a vector of time‐constant and time‐varying indicators of socio‐economic position, and HCki denotes a vector of housing characteristics as measured in the 2011 Census. All models were estimated separately for men and women. Missing values were deleted listwise in view of model comparison using Likelihood Ratios tests. All models were estimated using the stcox and contrast commands in Stata 14MP.CovariatesMigrant background and duration of residence constitute the main variables of interest in the study. Migrant background was defined as not having Belgian nationality at birth, which is considered more accurate to identify migrants than country of birth, as this would include natives with Belgian parents who were born abroad. Duration of residence is based on information available from the population register which provides the year of immigration since 1980. The exact duration of residence cannot be ascertained for migrants who settled in Belgium prior to 1980 and have an extended duration of residence. As a result, the typology of migrant background distinguishes seven groups: (i) Belgian born of native origin, (ii) Belgian born of foreign origin, (iii) foreign born who immigrated prior to the age of 18, and finally foreign born who immigrated after the age of 18 and who had attained (iv) 0 to 9 years, (v) 10 to 19 years, (vi) 20 to 29 years, and (vii) 30 years of residence or more in Belgium by the time of the 2011 census (see Tables 1, A1, and A2).1TableRegional distribution of native and migrant populations aged 25–75 by sex, migrant generation, and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):Belgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7]Men:Flemish region67.130.944.543.946.538.729.1‐ Province Antwerp18.610.119.818.118.314.19.3‐ Province Limburg8.710.38.68.59.98.39.3‐ Province Flemish‐Brabant11.13.65.36.46.48.24.4‐ Province East‐Flanders15.74.57.37.27.45.04.3‐ Province West‐Flanders13.02.43.53.84.53.11.8Brussels capital region4.114.927.034.929.328.922.3Walloon region28.854.328.521.224.232.448.7‐ Province Hainaut10.227.89.57.57.512.724.5‐ Province Liège8.818.812.88.410.611.116.7‐ Province Walloon‐Brabant3.22.82.62.22.64.53.5‐ Province Namur4.33.52.11.51.92.42.8‐ Province Luxembourg2.41.31.61.71.61.91.2100.0100.0100.0100.0100.0100.0100.0Women:Flemish region68.331.944.644.547.140.630.9‐ Province Antwerp18.810.220.619.019.716.59.6‐ Province Limburg9.111.17.88.910.16.510.0‐ Province Flemish‐Brabant11.13.65.45.75.88.14.6‐ Province East‐Flanders16.14.67.07.37.46.04.2‐ Province West‐Flanders13.32.53.73.64.03.52.6Brussels capital region3.714.225.934.930.429.619.6Walloon region28.054.029.520.622.529.949.5‐ Province Hainaut9.727.49.57.47.111.724.4‐ Province Liège8.718.713.58.310.310.817.4‐ Province Walloon‐Brabant3.02.93.01.82.23.73.4‐ Province Namur4.33.62.11.41.61.93.0‐ Province Luxembourg2.41.31.41.71.41.71.4Total100.0100.0100.0100.0100.0100.0100.0Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors.To determine whether spatial variation in migrants' mortality risks converges to that of natives as their duration of residence increases, mortality risks are differentiated by district (n = 43), which have been shown to exhibit significant variation in mortality levels in the general population (Deboosere & Gadeyne, 2002; Eggerickx et al., 2018). The level of districts is not ideal as it reflects a purely administrative division and districts vary considerably in size. Although lowering the geographical level of analysis would be relevant to understand spatial variation in mortality risks (Deboosere & Gadeyne, 2002), the level of municipalities (n = 589) did not allow robust estimates for migrant groups differentiated by migrant generation and duration of residence. To avoid that estimates of spatial convergence are contaminated by internal migration, the analyses are restricted to the subset of individuals whose district of residence on January 1 has remained constant between 2001 and 2015 or since immigration into the country. Previous studies suggest that internal migrants differ in terms of both mortality levels and spatial variation of all‐cause mortality compared to the group who lived in the same district throughout the observation period (Kravdal, 2009; Riva et al., 2011). The subset of individuals whose district of residence has remained constant throughout the observation period or since immigration represents 81.02% and 81.81% of native men and women and 78.81 and 78.84% of migrant men and women, respectively. To avoid issues of quasi‐complete separation in the model including interaction between district and migrant typology (Model 5c), the districts Diksmuide and Veurne (province of West‐Flanders) were collapsed for men, whereas the districts Ath and Tournai (province of Hainaut), the districts Diksmuide and Roeselare (province of West‐Flanders), and the districts Kortrijk and Tielt (province of West‐Flanders) were collapsed for women.The history of migration into Belgium has resulted in migrant populations being strongly concentrated in the former industrial belt in Wallonia, the former mining region in Limburg, smaller industrial areas, and central regions characterised by large secondary labour markets (Kesteloot, 1985). This results in a settlement pattern of the migrant population that is altogether different from that of the native‐born population (Table 1). The larger part of the native population lives in the Flemish region (~67%), followed by the Walloon region (28%), with only a minority living in the Brussels capital region (~4%). In contrast, both first generation migrants who immigrated prior to 1980 and second generation migrants are primarily concentrated in the Walloon region (50% to 55%), followed by Flanders (approximately 30%) and a more sizeable portion living in the capital region of Brussels (15 to 20%). Among migrants who immigrated after 1980, Flanders has gained importance with 40% to 45% of first‐generation migrants living in Flemish districts at the time of the 2011 Census, and equal shares of around 30% living in Wallonia and Brussels, although Brussels has clearly gained prominence over Wallonia between 2000 and 2010.Variation in living arrangements is captured through individuals' household positions using the LIPRO household typology (Van Imhoff & Keilman, 1991), distinguishing following positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child in a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL). Household position is included in the analysis as a time‐varying covariate as the information on living arrangements is updated annually (referring to the situation on January 1 of every year of observation).The socio‐economic position is measured using the activity status at the time of the 2011 census in tandem with a time‐varying indicator of income derived from the tax registers. The indicator of activity status distinguishes (i) students, (ii) employed individuals, (iii) housekeepers and other inactive persons, (iv) pensioners, (v) unemployed individuals with no prior work experience, and (vi) individuals with prior work experience who were unemployed at the time of the 2011 census. Income reflects individuals' income recorded in the annual tax returns between 2011 and 2015. Income is collapsed into deciles and included in the analysis as a time‐varying covariate. Education was not included as an indicator of socio‐economic position as information on educational attainment is frequently incomplete or missing for migrants, particularly for recent immigrants. We do not consider the lack of information on educational attainment to be problematic for the aims of this study as previous research for Belgium has shown that education is predominantly a relevant indicator of mortality risks in the youngest age groups, whereas professional status and particularly housing characteristics are more discriminate of mortality among individuals aged 25 and older (Gadeyne, 2006). Lacking reliable information on level of education, the analyses are restricted to men and women aged 25 and older.The 2011 Census provides a rich set of indicators of housing quality which serve as an indicator of accumulated wealth and have been shown to be strongly associated with mortality risks (Gadeyne, 2006). The analyses control for the following housing characteristics: (i) ownership status (renter versus owner), (ii) presence of a bath or shower in the dwelling, (iii) presence of central heating in the dwelling, (iv) presence of multiple dwellings in the housing unit (multiple vs. single dwelling), (v) number of rooms per inhabitant in the dwelling (less than 0.5 rooms, 0.5 to 1.5 rooms, 1.5 to 2.99 rooms, and 3 or more rooms), and (vi) period of construction (before 1919, 1919–1945, 1946–1960, 1961–1970, 1971–1980, 1981–1990, 1991–2000, and 2001–2010).Tables A1 and A2 provide the distribution of person‐years by covariates for both natives and migrants by migrant generation and duration of residence. As could be expected, migrants with durations of residence under 20 years are on average younger than natives, whereas migrants with extended durations of residence of 30 years and longer are somewhat older than natives on average. With respect to household positions, women who recently settled in Belgium are somewhat more likely to be in a union and have somewhat higher proportions of being heads of single person households. The latter household position is also more prevalent among migrant women with longer durations of residence than among native women. Also, the proportion of singles is higher among migrant women with a duration of residence of 30 years and longer, than among natives. The household positions of migrant men residing in the country for at least 30 years are similar to natives, although the proportion of singles is higher among recent immigrants. With respect to activity status, the proportions employed or retired reflect the differences in age distribution between groups, but the proportion unemployed is consistently higher in migrant groups than among natives. With respect to income, migrants are more concentrated in the middle‐income groups than natives, which is also reflected by the housing characteristics. Ownership is less prevalent in migrant groups, and the proportion of migrants living in dwellings without a bathroom or central heating is substantially higher. Similarly, migrants more frequently live in older housing units that often have multiple dwellings with less rooms available per inhabitant on average.RESULTSWe first estimate the mortality advantage of migrants compared to natives by duration of residence, controlling for age, living arrangement, socio‐economic position, and housing characteristics. Second, we document spatial variation of all‐cause mortality in the native population and test whether the spatial variation of all‐cause mortality in migrants converges to the pattern of natives with increasing duration of residence. Finally, we consider spatial variation in the migrant mortality advantage and test whether the overall mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population and their concentration in local settings characterised by high native mortality and large migrant‐native differentials in all‐cause mortality.Migrant‐native differentials in all‐cause mortality by duration of residenceTable 2 reports hazard ratios of all‐cause mortality in migrant women and men compared to natives. Mortality hazards are approximately 13% lower in migrant men and women compared to natives (Table 2, Model 1) and the mortality advantage of migrants relative to natives increases to 35% controlling for living arrangements, socio‐economic position, and housing characteristics (Table 2, Model 4). Further disaggregating by migrant generation and duration of residence indicates that mortality hazards of the intermediate and second generation are similar to natives, and even significantly higher among women who immigrated as children (Table 2, Model 1). The mortality hazard being similar or higher is due to the disadvantaged socio‐economic position of the intermediate and second generations as mortality hazards are 17% to 26% lower than natives controlling for age, living arrangement, socio‐economic position, and dwelling characteristics (Table 2, Model 4). In contrast, mortality hazards are significantly lower than natives among first generation men and women who immigrated after the age of 18, with differentials being largest among men and women with durations of residence between 0 and 9 years, and differentials being smaller in migrant groups with 10–19 and 20–29 years of residence, and particularly 30 or more years of residence (Table 2, Model 1). Similar to the results for the intermediate and second generation, the mortality advantage increases when controlling for living arrangements, socio‐economic position and housing characteristics. Mortality hazards are more than 60% lower than natives among migrants with 0 to 9 years of residence in Belgium, 45 to 50% lower among migrants with durations of residence ranging from 10 to 29 years, and approximately 27% lower among migrants with durations of residence of 30 years or longer (Table 2, Model 4). The mortality advantage is similar for men and women allowing for migrant generation and duration of residence and is consistent with health selection and acculturation mechanisms, thus largely confirming Hypothesis 1.2TableHazard ratios (HR) comparing all‐cause mortality in migrant men and women to natives by migrant generation and duration of residence, ages 25–75, Belgium, 2011–2015Model 1Model 2Model 3Model 4HRSig.HRSig.HRSig.HRSig.A. Women (native women reference)Belgian born, Belgian origin1.000Ref.1.000Ref.1.000Ref.1.000Ref.Migrant & migrant background0,869***0,834***0,690***0,634***Belgian born, Belgian origin1.000Ref.1.000Ref.1,000Ref.1,000Ref.Belgian born, foreign origin0,966‐0,918**0,841‐0,805***Agemig<18, foreign origin1,325*1,280*0,947***0,874‐Agemig>18, 00–09 yrs residence0,755***0,654***0,439***0,374***Agemig>18, 10–19 yrs residence0,819***0,797***0,609***0,548***Agemig>18, 20–29 yrs residence0,790***0,781***0,608***0,574***Agemig>18, 30 + yrs residence0,898***0,887***0,780***0,720***B. Men (native men reference)Belgian born, Belgian origin1.000Ref.1.000Ref.1.000Ref.1.000Ref.Migrant & migrant background0,871***0,871***0,694***0,650***Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1,000Ref.Belgian born, foreign origin1,028‐0,975‐0,860***0,834***Agemig < 18, foreign origin1,105‐1,059‐0,797*0,740**Agemig > 18, 00–09 yrs residence0,740***0,686***0,448***0,387***Agemig > 18, 10–19 yrs residence0,751***0,762***0,557***0,509***Agemig > 18, 20–29 yrs residence0,751***0,771***0,576***0,549***Agemig > 18, 30+ yrs residence0,899***0,940***0,779***0,731***Model legend: models control incrementally for age (Model 1), living arrangements (Model 2), activity status and income (Model 3) and housing characteristics (Model 4).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), Calculations by authors.*Significance levels: ‐ not significant, p < .050,**p < .010,***p < .001.Spatial variation in all‐cause mortality in migrant and native populationsTo compare spatial variation in all‐cause mortality between natives and migrants, the intermediate and second generation were collapsed into a single category, and also, first‐generation migrants with 10–19 and 20–29 years of residence were combined. The collapsed typology allows to compare migrants to natives in all districts. Collapsing the migrant typology did not result in a significant deterioration of model fit for either men or women. Model 5a (Table 3) replicates Model 4 (Table 2) using the collapsed migration typology. In Model 5b, district is added to Model 5a, whereas Model 5c additionally tests the interaction between the migration typology and district. Comparing models 5c and 5b indicates that the spatial variation in all‐cause mortality differs significantly between migrants and natives (∆‐2LL = 347.48; ∆ df = 148; p  < .001 for women, and ∆ ‐2LL = 512.22; ∆ df = 164; p  < .001 for men). The results of Model 5c are shown in Figure 1 (women) and Figure 2 (men). In both figures, districts were ranked by the mortality hazards in the native population, resulting in the familiar pattern that has been documented extensively in the literature: Flemish districts show the lowest mortality, with mortality being lowest in districts located in the provinces of East and West Flanders, followed by districts located in the provinces of Antwerp, Flemish‐Brabant and Limburg. With the exception of Mouscron, the Walloon districts show higher mortality than Flemish districts, with mortality levels being somewhat more favourable in districts located in the provinces Walloon Brabant and Luxembourg and overall less favourable for districts in the province of Namur. Districts in the provinces of Hainaut and Liège show somewhat more heterogeneity compared to the districts in the other Walloon provinces. Finally, the capital region of Brussel takes an intermediate position, with mortality being higher than in all of the Flemish districts, as well as a limited group of Walloon districts.3TableHazard ratios (HR) comparing all‐cause mortality in migrant men and women to natives by migrant generation and duration of residence, ages 25–75, Belgium, 2011–2015Model 5aModel 5bModel 5cHRSig.HRSig.HRSig.A. Women (native women reference)Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1.5 and 2nd generation0.809***0.753***0.840***Agemig>18, 00–09 yrs residence0.374***0.375***0.447***Agemig>18, 10–29 yrs residence0.559***0.556***0.679***Agemig>18, 30 + yrs residence0.720***0.682***0.805***B. Men (native men reference)Belgian born, Belgian origin1,000Ref.1,000Ref.1,000Ref.1.5 and 2nd generation0.830***0.775***0.852**Agemig>18, 00–09 yrs residence0.387***0.408***0.444***Agemig>18, 10–29 yrs residence0.525***0.546***0.658***Agemig>18, 30 + yrs residence0.731***0.700***0.745***Model legend: Model 5a reports the migrant mortality advantage controlling for age, living arrangements, activity status and income. Model 5b incrementally includes district, whereas Model 5c additionally includes the interaction between migrant typology and district. The estimates of the migrant mortality advantage in Model 5c reflect the average migrant mortality advantage across districts.Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), Calculations by authors.*Significance levels: ‐ not significant. p < .050,**p < .010,***p < .001.1FigureDistrict‐level variation in all‐cause mortality hazards of native versus migrant women by migrant generation and duration of residence controlling for household position, socio‐economic and housing characteristics (Model 5c), Belgium, 2011–20152FigureDistrict‐level variation in all‐cause mortality hazards of native versus migrant men by migrant generation and duration of residence controlling for household position, socio‐economic, and housing characteristics (Model 5c), Belgium, 2011–2015Turning to spatial variation in migrant mortality, the familiar spatial variation of all‐cause mortality in native women does not emerge among migrant women who settled in Belgium as adults and had a duration of residence between 0 and 9 years at the time of the 2011 Census (Figure 1a). Allowing for differences in socio‐demographic profile and socio‐economic position, the mortality advantage relative to natives is subject to substantial variation between districts, with differentials generally being smallest in the Flemish districts and considerably larger in Brussels and the majority of Walloon districts, where migrant groups are predominantly located (Table 1). With increasing duration of residence, the mortality advantage becomes smaller (Figure 1b) and the spatial gradient in all‐cause mortality among migrants with a duration of residence of 30 years or longer converges to the pattern of native women, albeit that the mortality advantage remains somewhat larger in the Walloon districts (Figure 1c). Among women of migrant origin who were born in Belgium or immigrated under the age of 18 (Figure 1d), the mortality advantage relative to natives is virtually absent in most of the Flemish districts and much smaller in Walloon districts compared with the other migrant groups. This suggests that health selection has only carried over to the intermediate and second generations to a limited extent. Alternatively, these groups may continue to enjoy a limited mortality advantage compared to natives as a result of group‐specific behaviours, but this advantage is nullified by the unfavourable socio‐economic position of the groups considered. Similar patterns emerge with respect to the mortality advantage of migrant men compared to natives (Figures 2a–d). For both men and women, the results largely confirm hypothesis 2 stating that health selection and acculturation mechanisms imply a gradual convergence of mortality risks between migrants and natives with increasing duration of residence, and also a gradual convergence of spatial variation in all‐cause mortality, leading to significant between‐district variation in the migrant mortality advantage, particularly among migrant groups with short durations of residence.Finally, considering the between‐district variation of the migrant mortality advantage, Model 5c calculates the average mortality advantage of each migrant group as the geometric mean of the mortality advantages within districts (Table 3). Allowing for living arrangements, socio‐economic position, and housing characteristics, mortality hazards in men and women of the intermediate and second generation are on average approximately 15% lower than is the case for natives. Among men and women with a duration of residence between 0 and 9 years, the mortality advantage amounts to 55% on average across districts, compared with 35% among men and women with durations of residence between 10 and 29 years, and 20% to 25% among women and men with durations of residence exceeding 30 years, respectively (Model 5c, Table 3). The comparison of Model 5c to the results of Model 5a provides clues as to how settlement patterns and spatial variation in the mortality advantage can partially explain the overall migrant mortality advantage in Belgium. Allowing for living arrangements, socio‐economic position and housing characteristics, the average mortality advantage within districts (Model 5c) is smaller for all migrant groups than the country‐level mortality advantage (Model 5a). In line with the acculturation and migration as rapid health transition arguments, mortality in several migrant groups is less subject to spatial variation than mortality in natives (Figures 1 and 2). In regions with low native mortality such as the majority of Flemish districts, this gives rise to a smaller mortality advantage for migrants compared to districts with higher mortality in the native population such as Brussels and the majority of Walloon districts. As the migrant population is strongly concentrated in districts with high native mortality and large migrant‐native differentials in all‐cause mortality (Table 1), the country‐level mortality advantage largely reflects the sizeable migrant‐native mortality differentials in these districts.DISCUSSION AND CONCLUSIONThis paper builds on two separate bodies of literature that consider spatial variation in mortality and migrant‐native differentials in mortality, respectively. The first body of literature has documented for various contexts that all‐cause and cause‐specific mortality are typically subject to spatial variation net of compositional effects in terms of socio‐economic position and selective internal migration between regions by health status. The second body of literature has documented for various contexts that recent migrants typically have lower mortality than natives as a result of health selection and that the mortality advantage gradually diminishes with increasing duration of residence and among the second generation in line with acculturation theories. The latter arguments suggest, however, that research on migrant‐native differentials in all‐cause and cause‐specific mortality may benefit from explicitly introducing a spatial dimension. First, arguments referring to health selection and acculturation suggest that spatial variation in migrant mortality may largely differ from the pattern found in the native population, implying that the migrant‐native differentials may be subject to considerable regional variation allowing for migrant generation and duration of residence. Second, migrant groups often have specific settlement patterns, implying that consideration of these spatial aspects may provide additional insight into the nature of the country‐level mortality advantage.This paper contributes to the literature by introducing a spatial dimension in the analysis of migrant‐native differentials controlling for socio‐demographic characteristics, socio‐economic position, and housing characteristics. Considering variation in the mortality advantage migrant generation and duration of residence, our findings are consistent with earlier findings on migrant‐native differentials in all‐cause mortality in Belgium. Similar to the study by Vandenheede et al. (2015) for the period 2001–2010, we find that the mortality advantage compared to natives is largest among first generation migrants with short durations of residence between 0 and 9 years and is progressively smaller among first generation migrants with durations of residence between 10 and 19 years, 20–29 years and 30 years and over, in line with health selection and acculturation arguments (Hypothesis 1). Migrants who immigrated as children (intermediate generation) and the second generation, where health selection can be largely presumed absent, have mortality risks similar to natives. For all groups, migrant‐native differentials in all‐cause mortality increase when controlling for living arrangements, socio‐economic position and housing characteristics, indicating that the socio‐economic profile of migrant groups is less favourable than the profile of natives and partially conceals the mortality advantage.Health selection and acculturation theories suggest that the gradual convergence of mortality patterns may also apply to spatial variation of all‐cause mortality, with spatial variation in migrant mortality being dissimilar from natives among first‐generation migrants with short durations of residence and being more similar among first generation migrants with long durations of residence, as well as the intermediate and second generation (Hypothesis 2). Controlling for socio‐demographic profile, socio‐economic position, and housing characteristics, we find that spatial variation of all‐cause mortality differs significantly between migrant groups and natives, with spatial variation in mortality becoming increasingly similar to natives with increasing duration of residence. Although migrant‐native differentials in all‐cause mortality are virtually absent in the Flemish districts among first‐generation migrants with an extended duration of residence, as well as the intermediate and second generation, the mortality advantage persists in Brussels and the majority of Walloon districts and predominantly reflects the high mortality level of natives in these districts.The finding that spatial variation in all‐cause mortality differs between natives and a sizeable part of the migrant population implies that the mortality advantage is subject to variation between districts. The country‐level mortality advantage is obtained by comparing the nation‐wide mortality of migrants to the nation‐wide mortality rate of natives and presents a weighted average of the mortality advantage within districts which partially reflects the spatial distribution of both migrant and native populations. As a result, the migrant mortality advantage at the country level largely reflects the mortality advantage in districts where migrants are primarily located, which may not necessarily reflect the mortality advantage in other districts. Given these mechanisms, the last part of the paper explored whether and to what extent the country‐level mortality advantage in Belgium can be partially accounted for by the settlement pattern of the migrant population (Hypothesis 3). In the Belgian case, a substantial part of the migrant population is located in Brussels and Wallonia, characterised by higher mortality among natives, which gives rise to a large mortality advantage in these districts. In contrast, the mortality advantage is substantially smaller in the majority of Flemish districts. Whereas a high level of convergence in mortality risks is reached between migrants and natives in Flemish districts, the convergence is lower in Brussels and the majority of Walloon districts. This is also true in groups with extended durations of residence exceeding 30 years as well as the intermediate and second generations, indicating that only partial exposure or acculturation with local risk factors takes place in migrant populations in these districts (Grimmeau et al., 2015). The partial convergence of spatial variation in all‐cause mortality of migrants and natives is likely to be driven by selective exposure to or adoption of risk behaviours by migrants, resulting in a partial adoption of the spatial patterns of cause‐specific mortality found in natives. The country‐level mortality advantage partially reflects the concentration of the migrant population in districts characterised by high mortality among natives and is therefore somewhat larger than the average mortality advantage within districts.Despite their strengths, the analyses reported in this paper also suffer a number of limitations. First, although the gender differences in mortality advantage reported in this paper are consistent with the healthy migrant hypothesis, the analyses could not explicitly control for migration motive, although this has been shown to affect the health selection at immigration (Syse et al., 2018). As the Belgian Census and population register only cover the resident population, the data at hand do not allow to determine directly whether immigrants are selective in terms of health status compared to the resident population in the country of origin. Second, given that the observation window covers the period from 2011 to 2015, the different groups in terms of duration of residence pertain to different migration cohorts. Hence, the finding that migrant‐native mortality differentials are smaller in migrant groups with longer duration of residence may suggest convergence of mortality risks or may indicate that health selection was less articulate in older cohorts than is the case in more recent migrant cohorts. Particularly for men, we consider it unlikely that health selection would have increased over time as labour migration was the prime motive in older migrant cohorts (suggesting strong health selection and even explicit health screening for some groups), whereas migration motives have become more diverse in recent cohorts (suggesting weaker health selection). Third, the analysis has not used spatial lag models, which are less straightforward to implement in the context of hazard models. We feel that future work could further explore the clustering of mortality risks in neighbouring districts which may violate the assumption of independence.Despite the limitations, we feel that our analysis illustrates the relevance of introducing a spatial component into the analysis of the migrant mortality advantage, and several avenues for future research can be identified. First, the heterogeneity in the migrant population could be further explored; migration motives and socio‐economic position are known to vary between EU and non‐EU migrants, but also by the specific country of origin, particularly in the case of non‐EU migrants (UNIA & Fod WASO, 2017). An obvious extension of the analyses presented in this paper would therefore be to explore how selective exposure to local conditions affects the mortality advantage found in different migrant groups given their specific settlement patterns, which may enhance our understanding of the factors driving between‐group variation in mortality outcomes. Although we consider such differentiation to be a potentially fruitful avenue for future research, analyses for specific origin groups will have to deal with substantially smaller sample sizes and specific patterns of residence that preclude a nation‐wide comparison of spatial variation in mortality patterns, as was the case in this paper. As a result, analyses for specific subgroups are likely to require more specific research designs. Second, the analyses have focused on individuals who have been consistently exposed to local conditions for a period of 15 years in line with previous research that suggests that change in context does not affect mortality risks immediately (Kravdal, 2009; Riva et al., 2011). The results in this paper refer to a (large) subset of the population, but more work is needed to quantify how exposure to different spatial contexts over the life course affects mortality risks, in both migrant and native populations. Third, although the possibility of unregistered emigration in the population register was deemed unlikely, it is clear that a fair amount of registered emigration took place throughout the observation period, with emigration being considerably more prevalent than mortality among migrants with shorter duration of residence. To the extent that emigration among recent migrants is selective in terms of health status (e.g., as a result of limited access to social security and/or health care), the impact on the mortality advantage found among recent immigrants may be substantial. We consider more detailed work on selective emigration among recent immigrants and its impact on the mortality advantage found in this group to be a relevant avenue for future research. The decline of the mortality advantage with increasing duration of residence and the convergence of spatial variation in all‐cause mortality between natives and migrants with an extended duration of residence suggest that acculturation of mortality risks is taking place that is not offset by selective (r)emigration. Fourth and finally, a logical extension would be to consider regional patterns in cause‐specific mortality given that regional variation in all‐cause mortality is the outcome of regional variation in patterns of cause‐specific mortality.ACKNOWLEDGEMENTSWe thank Patrick Lusyne, Cloë Ost, and Lien Co at Statistics Belgium for making the coded longitudinal microdata from the 2011 Belgian Census, the Population Register (2000–2015), and the tax return register (2000–2015) available.DATA AVAILABILITY STATEMENTThe data that support the findings of this study are available from Statistics Belgium. Restrictions apply to the availability of these data, which were used under licence for this study.REFERENCESAldridge, R. W., Nellums, L. B., Bartlett, S., Barr, A. L., Patel, P., Burns, R., Hargreaves, S., Miranda, J. J., Tollman, S., Friedland, J. S., & Abubakar, I. (2018). Global patterns of mortality in international migrants: A systematic review and meta‐analysis. Lancet, 392(10164), 2553–2566. https://doi.org/10.1016/s0140-6736(18)32781-8Andersson, G., & Drefahl, S. (2017). Long‐distance migration and mortality in Sweden: Testing the salmon bias and healthy migrant hypotheses. Population Space and Place, 23(4), 11. https://doi.org/10.1002/psp.2032Arenas, E., Goldman, N., Pebley, A. R., & Teruel, G. (2015). Return migration to Mexico: Does health matter? Demography, 52(6), 1853–1868. https://doi.org/10.1007/s13524-015-0429-7Bauwelinck, M., Deboosere, P., Willaert, D., & Vandenheede, H. (2017). Suicide mortality in Belgium at the beginning of the 21st century: Differences according to migrant background. European Journal of Public Health, 27(1), 111–116. https://doi.org/10.1093/eurpub/ckw159Bos, V., Kunst, A. E., Garssen, J., & Mackenbach, J. P. (2007). Duration of residence was not consistently related to immigrant mortality. Journal of Clinical Epidemiology, 60(6), 585–592. https://doi.org/10.1016/j.jclinepi.2006.08.010Bostean, G. (2013). Does selective migration explain the Hispanic paradox? A comparative analysis of Mexicans in the US and Mexico. Journal of Immigrant and Minority Health, 15(3), 624–635. https://doi.org/10.1007/s10903-012-9646-yBrazil, N. (2017). Spatial variation in the Hispanic paradox: Mortality rates in new and established Hispanic US destinations. Population Space and Place, 23(1), 17. https://doi.org/10.1002/psp.1968Castaneda, H., Holmes, S. M., Madrigal, D. S., Young, M. E. D., Beyeler, N., & Quesada, J. (2015). Immigration as a Social Determinant of Health. In J. E. Fielding (Ed.), Annual review of public health, Vol 36 (Vol. 36) (pp. 375–392). Palo Alto: Annual Reviews.Chiswick, B. R., Lee, Y. L., & Miller, P. W. (2008). Immigrant selection systems and immigrant health. Contemporary Economic Policy, 26(4), 555–578. https://doi.org/10.1111/j.1465-7287.2008.00099.xClark, S. J., Collinson, M. A., Kahn, K., Drullinger, K., & Tollman, S. M. (2007). Returning home to die: Circular labour migration and mortality in South Africa. Scandinavian Journal of Public Health, 35, 35–44. https://doi.org/10.1080/14034950701355619Corijn, M., & Lodewijckx, E. (2009). De start van de gezinsvorming bij de Turkse en Marokkaanse tweede generatie in het Vlaams gewest. Een analyse op basis van Rijksregistergegevens. Studiedienst van de Vlaamse Regering.Deboosere, P., & Fiszman, P. (2009). De la persistance des inégalités socio‐spatiales de santé. Le cas Belge. Espace, Populations, Sociétés, 2009(1), 149–158.Deboosere, P., & Gadeyne, S. (2002). Can regional patterns of mortality in Belgium be explained by individual socio‐economic characteristics? Reflets et perspectives de la vie économique, XLI(4), 87–103.Deboosere, P., & Gadeyne, S. (2005). La sous‐mortalité des immigrés adultes en Belgique: une réalité attestée par les recensements et les registres. Population, 60(5), 765–811.Duchene, J., & Thiltgès, E. (1993). La mortalité des plus de 15 ans en Belgique: les disparités régionales en 1985–1987. Espace, Populations, Sociétés, 1, 61–74.Eggerickx, T., Sanderson, J. P., & Vandeschrick, C. (2018). Les inégalités sociales et spatiales de mortalité en Belgique: 1991–2016. Espace Populations Sociétés, 2018(1–2), 1–37.Fuller‐Thomson, E., Brennenstuhl, S., Cooper, R., & Kuh, D. (2015). An investigation of the healthy migrant hypothesis: Pre‐emigration characteristics of those in the British 1946 birth cohort study. Canadian Journal of Public Health‐Revue Canadienne De Sante Publique, 106(8), E502–E508. https://doi.org/10.17269/cjph.106.5218Gadeyne, S. (2006). The ultimate inequality: Socio‐economic differences in all‐cause and cause‐specific mortality in Belgium in the first part of the 1990s (Vol. 39). Brussels & The Hague: NIDI‐CBGS Publications.Gimeno‐Feliu, L. A., Calderon‐Larranaga, A., Diaz, E., Laguna‐Berna, C., Poblador‐Plou, B., Coscollar‐Santaliestra, C., & Prados‐Torres, A. (2019). The definition of immigrant status matters: Impact of nationality, country of origin, and length of stay in host country on mortality estimates. BMC Public Health, 19, 8. https://doi.org/10.1186/s12889-019-6555-1Grimmeau, J.‐P., Deboosere, P., Eggerickx, T., Gadeyne, S., Hermia, J.‐P., Marissal, P., … Willaert, D. (2015). Atlas van België: Bevolking. Gent: Academia Press.Guillot, M., Khlat, M., Elo, I., Solignac, M., & Wallace, M. (2018). Understanding age variations in the migrant mortality advantage: An international comparative perspective. PLoS ONE, 13(6), 17. https://doi.org/10.1371/journal.pone.0199669Hamilton, T. G. (2015). The healthy immigrant (migrant) effect: In search of a better native‐born comparison group. Social Science Research, 54, 353–365. https://doi.org/10.1016/j.ssresearch.2015.08.008Helgesson, M., Johansson, B., Nordquist, T., Vingard, E., & Svartengren, M. (2019). Healthy migrant effect in the Swedish context: A register‐based, longitudinal cohort study. BMJ Open, 9(3), 10. https://doi.org/10.1136/bmjopen-2018-026972Hofmann, E. T. (2012). The burden of culture? Health outcomes among immigrants from the former Soviet Union in the United States. Journal of Immigrant and Minority Health, 14(2), 315–322. https://doi.org/10.1007/s10903-010-9436-3Ichou, M., & Wallace, M. (2019). The healthy immigrant effect: The role of educational selectivity in the good health of migrants. Demographic Research, 40, 61–93. https://doi.org/10.4054/DemRes.2019.40.4Jervelund, S. S., Malik, S., Ahlmark, N., Villadsen, S. F., Nielsen, A., & Vitus, K. (2017). Morbidity, self‐perceived health and mortality among non‐Western immigrants and their descendants in Denmark in a life phase perspective. Journal of Immigrant and Minority Health, 19(2), 448–476. https://doi.org/10.1007/s10903-016-0347-9Keenan, K., Kulu, H., & Cox, F. (2021). Editorial introduction: Social and spatial inequalities in health and mortality: The analysis of longitudinal register data from selected European countries. Population, Space and Place.Kennedy, S., Kidd, M. P., McDonald, J. T., & Biddle, N. (2015). The healthy immigrant effect: Patterns and evidence from four countries. Journal of International Migration and Integration, 16(2), 317–332. https://doi.org/10.1007/s12134-014-0340-xKesteloot, C. (1985). De geografische spreiding van buitenlandse minderheden in België. Beschrijvingen en verklaringen. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen & Amsterdam: De Nederlandsche Boekhandel.Kibele, E., Scholz, R., & Shkolnikov, V. M. (2008). Low migrant mortality in Germany for men aged 65 and older: Fact or artifact? European Journal of Epidemiology, 23(6), 389–393. https://doi.org/10.1007/s10654-008-9247-1Kravdal, O. (2009). Mortality effects of average education in current and earlier municipality of residence among internal migrants, net of their own education. Social Science & Medicine, 69(10), 1484–1492. https://doi.org/10.1016/j.socscimed.2009.08.035Lievens, J. (2000). The third wave of immigration from Turkey and Morocco: Determinants and characteristics. In R. Lesthaeghe (Ed.), Communities and generations. Turkish and Moroccan populations in Belgium (Vol. 36). Brussels & The Hague: NIDI/CBGS Publications.Martens, A. (1985). Het na‐oorlogs immigratiebeleid. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen & Amsterdam: De Nederlandsche Boekhandel.Martinez, J. N., Aguayo‐Tellez, E., & Rangel‐Gonzalez, E. (2015). Explaining the Mexican‐American health paradox using selectivity effects. International Migration Review, 49(4), 878–906. https://doi.org/10.1111/imre.12112Mehta, N. K., & Elo, I. T. (2012). Migrant selection and the health of US immigrants from the former Soviet Union. Demography, 49(2), 425–447. https://doi.org/10.1007/s13524-012-0099-7Myria. (2019). Migratie in cijfers en in rechten. Brussel: Federaal Migratiecentrum.Norredam, M., Agyemang, C., Hansen, O. K. H., Petersen, J. H., Byberg, S., Krasnik, A., & Kunst, A. E. (2014). Duration of residence and disease occurrence among refugees and family reunited immigrants: Test of the 'healthy migrant effect' hypothesis. Tropical Medicine & International Health, 19(8), 958–967. https://doi.org/10.1111/tmi.12340Norredam, M., Hansen, O. H., Petersen, J. H., Kunst, A. E., Kristiansen, M., Krasnik, A., & Agyemang, C. (2015). Remigration of migrants with severe disease: Myth or reality?—A register‐based cohort study. European Journal of Public Health, 25(1), 84–89. https://doi.org/10.1093/eurpub/cku138Rasulo, D., Spadea, T., Onorati, R., & Costa, G. (2012). The impact of migration in all‐cause mortality: The Turin longitudinal study, 1971‐2005. Social Science & Medicine, 74(6), 897–906. https://doi.org/10.1016/j.socscimed.2011.10.045Renard, F., Tafforeau, J., & Deboosere, P. (2015). Mapping the cause‐specific premature mortality reveals large between‐districts disparity in Belgium, 2003‐2009. Archives of Public Health, 73, 18. https://doi.org/10.1186/s13690-015-0060-5Reniers, G. (1999). On the history and selectivity of Turkish and Moroccan migration to Belgium. International Migration, 37(4), 679–713. https://doi.org/10.1111/1468-2435.00090Reus‐Pons, M., Vandenheede, H., Janssen, F., & Kibele, E. U. B. (2016). Differences in mortality between groups of older migrants and older non‐migrants in Belgium, 2001‐09. European Journal of Public Health, 26(6), 992–1000. https://doi.org/10.1093/eurpub/ckw076Riva, M., Curtis, S., & Norman, P. (2011). Residential mobility within England and urban‐rural inequalities in mortality. Social Science & Medicine, 73(12), 1698–1706. https://doi.org/10.1016/j.socscimed.2011.09.030Singer, J., & Willett, J. (2003). Applied longitudinal data analysis. Modeling change and event occurrence. Oxford: Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195152968.001.0001Stols, E. (1985). Emigratie en immigratie in historich perspectief. In A. Martens & F. Moulaert (Eds.), Buitenlandse minderheden in Vlaanderen‐België. Antwerpen/Amsterdam: Uitgeverij De Nederlandsche Boekhandel.Syse, A., Dzamarija, M. T., Kumar, B. N., & Diaz, E. (2018). An observational study of immigrant mortality differences in Norway by reason for migration, length of stay and characteristics of sending countries. BMC Public Health, 18, 12. https://doi.org/10.1186/s12889-018-5435-4UNIA, & Fod WASO. (2017). Socio‐economische monitoring 2017: Arbeidsmarkt en origine. Brussel: Interfederaal Gelijkekansencentrum en Federale Overheidsdienst Werkgelegenheid en Sociaal Overleg.Urquia, M. L., O'Campo, P. J., & Heaman, M. I. (2012). Revisiting the immigrant paradox in reproductive health: The roles of duration of residence and ethnicity. Social Science & Medicine, 74(10), 1610–1621. https://doi.org/10.1016/j.socscimed.2012.02.013Vallin, J., Meslé, F., & Valkonen, T. (2001). Trends in mortality and differential mortality. Strassbourg: Council of Europe Publishing.Van Hemelrijck, W. M. J., de Valk, H. A. G., & Vandenheede, H. (2016). Cancer mortality by migrant background in the 2000s in Belgium: Patterns and determinants. European Journal of Public Health, 26, 383–383.Van Imhoff, E., & Keilman, N. (1991). Lipro 2.0: An application of a dynamic demographic projection model to household structure in the Netherlands (Vol. 23). The Hague & Brussels: NIDI‐CBGS.Vandenheede, H., Willaert, D., De Grande, H., Simoens, S., & Vanroelen, C. (2015). Mortality in adult immigrants in the 2000s in Belgium: A test of the ‘healthy‐migrant’ and the ‘migration‐as‐rapid‐health‐transition’ hypotheses. Tropical Medicine & International Health, 20(12), 1832–1845. https://doi.org/10.1111/tmi.12610Vanthomme, K., & Vandenheede, H. (2019). Trends in Belgian cause‐specific mortality by migrant origin between the 1990s and the 2000s. BMC Public Health, 19, 16. https://doi.org/10.1186/s12889-019-6724-2Wallace, M., & Darlington‐Pollock, F. (2020). Poor health, low mortality? Paradox found among immigrants in England and Wales. Population, Space and Place, e2360. https://doi.org/10.1002/psp.2360Wallace, M., Khlat, M., & Guillot, M. (2019). Mortality advantage among migrants according to duration of stay in France, 2004‐2014. BMC Public Health, 19, 9. https://doi.org/10.1186/s12889-019-6652-1Wallace, M., & Kulu, H. (2014). Low immigrant mortality in England and Wales: A data artefact? Social Science & Medicine, 120, 100–109. https://doi.org/10.1016/j.socscimed.2014.08.032Wallace, M., & Kulu, H. (2018). Can the salmon bias effect explain the migrant mortality advantage in England and Wales? Population Space and Place, 24(8), 18. https://doi.org/10.1002/psp.2146Weitoft, G. R., Gullberg, A., Hjern, A., & Rosén, M. (1999). Mortality statistics in immigrant research: Method for adjusting underestimation of mortality. International Journal of Epidemiology, 28, 756–763. https://doi.org/10.1093/ije/28.4.756Wilson, B., Drefahl, S., Sasson, I., Henery, P. M., & Uggla, C. (2020). Regional trajectories in life expectancy and lifespan variation: Persistent inequality in two Nordic welfare states. Population, Space and Place, 26(e2378). https://doi.org/10.1002/psp.2378AAPPENDIXA1TableSocio‐demographic and socio‐economic profile of native and migrant women aged 25–75 by migrant generation and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):DeathsBelgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7][N]Age (time‐varying)‐ Mean age51.4940.5432.2540.3347.3854.5457.9645,565Household position [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ SING14.3511,079.2013.9510.1113.4116.4612,734‐ MAR028.4410.193.9712.9512.5917.6628.3417,341‐ MAR+30.7435.6634.8340.9950.7947.0735.046,507‐ CMAR2,839.2715.440.660.230.120.53353‐ UNM05,134.925.155.822.101.941.981,882‐ UNM+6,829.389.007.195.042.772.30767‐ CUNM0,180.420.930.110.020.020.0219‐ H1PA8.1712.8611.109.6816.0014.9612.163,528‐ C1PA1.593.956.690.430.200.170.72579‐ NFR0.580.941.835.951.820.801.02699‐ OTHER1.111.331.832.241.081.041.40886‐ COLL0.060.020.030.030.020.030.05270Activity status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Student1.484.3510.472.060.230.070.0442‐ Employed58.3761.3247.4140.2746.0440.2630.549,680‐ Homekeepers20.1320.4629.3852.6742.7546.2971.7416,412‐ Pensioner16.082.540.220.983.296.6019.9618,117‐ Never worked0.151.181.730.040.030.030.2939‐ Unemployed9.7910.1610.783.987.676.767.431,275Income decile [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ 1st decile (low)0.831.924.269.945.375.461.42983‐ 2nd decile0.821.934.569.985.405.361.35964‐ 3rd decile1.683.328.3710.796.325.832.301,185‐ 4th decile16.1317.7425.7027.1627.2328.6533.5211,073‐ 5th decile16.9619.4916.8815.0720.2220.5022.289,822‐ 6th decile12.3611.9010.468.119.969.6013.658,280‐ 7th decile12.0511.7110.367.478.917.989.655,109‐ 8th decile13.0513.069.934.637.016.736.623,203‐ 9th decile14.0511.276.223.414.454.614.802,850‐ 10th decile (high)12.067.653.263.445.145.294.412,096N deaths40,4221,134667936284572,06545,565N individuals1,990,659143,10816,858155,47766,43130,75386,3732,489,659N person‐years9,702,778765,33999,637667,120314,227145,723401,68912,096,513Ownership status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Home owner81.2669.9754.7836.7358.5068.8372.6331,480‐ Renter18.7430.0345.2263.2741.5031.1727.3714,085Dwelling with bath or shower [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available0.801.170.971.240.790.710.90665‐ Bath/shower99.2098.8399.0398.7699.2199.2999.1044,900Dwelling with central heating [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available12.8414.4912.3310.3310.2310.0714.248,040‐ Central heating87.1687.5187.6789.6789.7789.9385.7637,525Type of building [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ 1 dwelling84.9775.2758.2645.6860.3768.5472.5234,325‐ 2 dwellings2.724.325.875.924.634.675.071,607‐ 3+ dwellings12.0520.1535.5748.0934.6726.5122.169,516‐ Nonresidential0.260.260.300.320.340.290.25117Number of rooms per inhabitant [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ less than 0.500.070.371.031.520.830.460.3346‐ 0.50–0.991.215.1313.1111.8111.036.304.70537‐ 1.00–1.245.2811.5718.7116.5816.5112.488.901,439‐ 1.25–1.498.1612.8814.4812.7413.2210.798.331,722‐ 1.50–1.9918.7422.5819.4416.2918.6419.5415.724,300‐ 2.00–2.4919.0319.1015.5216.0015.0516.9517.257,430‐ 2.50–2.9913.278.915.646.996.888.6012.077,018‐ 3.00 or more34.2519.4612.0618.0617.8424.8832.7123,073Construction period of dwelling [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Before 191917.1828.8525.7826.2321.7524.4531.759,452‐ 1919–194512.3714.7616.5515.1415.3715.7015.415,890‐ 1946–196010.4012.0114.1013.0514.0012.7311.374,897‐ 1961–197010.808.8611.8912.8512.1910.668.846,570‐ 1971–198017.0810.5111.8812.2513.3112.6514.109,435‐ 1981–199011.507.075.424.555.969.267.203,925‐ 1991–200012.569.307.026.7710.4710.047.203,459‐ 2001–20054.074.023.314.413.942.632.271,070‐ 2006 or later4.054.634.064.763.011.881.86867N deaths40,4221,13466793628457206545,565N individuals1,990,659143,10816,858155,47766,43130,75386,3732,489,659N person‐years9,702,778765,33999,637667,120314,227145,723401,68912,096,513Note: Legend Household Positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child of a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors.A2TableSocio‐demographic and socio‐economic profile of native and migrant men aged 25–75 by migrant generation and duration of residence, Belgium, 2011–2015Belgian born:Foreign born by age at immigration in years (AI) and duration of residence in years (R):DeathsBelgian originMigrant originAI < 18AI > 18, 00–09 RAI > 18, 10–19 RAI > 18, 20–29 RAI > 18, 30+ R[1][2][3][4][5][6][7][N]Age (time‐varying)‐ Mean age50.7440.1631.3341.4548.4355.5958.99Household position [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ SING14.5916.7517.5323.0515.6514.3712.5819,371‐ MAR027.309.553.4112.2812.0117.9632.0331,837‐ MAR+31.7530.5923.3538.9258.1656.3943.2612,510‐ CMAR5.6515.5625.641.140.360.191.081,049‐ UNM05.675.474.955.772.632.512.493,174‐ UNM+7.149.387.087.625.703.382.941,503‐ CUNM0.320.701.320.160.040.010.0373‐ H1PA2.182.000.801.672.313.122.481,765‐ C1PA3.196.7110.550.670.330.231.201,813‐ NFR0.671.051.973.950.980.690.83866‐ OTHER1.472.193.334.731.811.111.031,470‐ COLL0.060.050.080.060.030.040.04312Activity status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ student1.564.099.681.730.100.060.0189‐ employed69.2070.5159.2357.2465.2760.3844.0021,998‐ Homekeepers6.4711.6119.1733.0519.9821.4815.5812,545‐ pensioner19.573.710.071.513.697.8332.8738,295‐ never worked0.110.610.720.030.010.000.0641‐ unemployed3.089.4711.136.4410.9610.247.472,775Income decile [time‐varying, distribution of person‐years exposure (%) and deaths (N)]‐ 1st decile (low)0.882.325.469.294.444.300.931,186‐ 2nd decile0.882.345.359.274.354.440.921,186‐ 3rd decile1.343.728.559.424.393.981.091,204‐ 4th decile3.388.2013.1813.1012.7112.616.663,148‐ 5th decile6.829.7610.9310.7613.6015.4116.7511,708‐ 6th decile10.609.169.349.4010.5910.9921.1618,270‐ 7th decile13.9012.8213.4712.0213.5611.5919.9315,577‐ 8th decile16.8118.3316.5511.4314.6412.7212.269,139‐ 9th decile19.9817.9611.257.0910.4711.0810.237,199‐ 10th decile (high)25.4115.385.918.2211.2412.8810.087,126N deaths66,5712,206971,2699216644,01575,743N individuals1,939,833149,09615,684146,66459,07925,50189,0952,424,952N person‐years9,441,797794,87494,502597,194275,607119,067407,61711,730,658Ownership status [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Home owner82.8372.9655.0232.4557.4470.5576.7253,316‐ Renter17.1727.0444.9867.5542.5629.4523.2822,427Dwelling with bath or shower [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available1.051.281.111.341.020.861.021,627‐ Bath/shower98.9598.7298.8998.6698.9899.1498.9874,116Dwelling with central heating [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Not available13.6715.1311.8610.5310.7010.4314.3015,055‐ Central heating86.3384.8788.1489.4789.3089.5785.7060,688Type of building [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ 1 dwelling85.8676.1958.4343.9057.6166.9675.0058,716‐ 2 dwellings2.764.626.196.305.235.205.152,629‐ 3 + dwellings11.0618.9235.0549.4136.8627.5119.6014,133‐ Nonresidential0.320.270.330.400.290.320.26265Number of rooms per inhabitant [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ less than 0.500.070.401.211.851.040.650.39119‐ 0.50–0.991.244.8012.1211.7613.099.275.66989‐ 1.00–1.245.6111.3817.6816.1818.7614.6910.432,868‐ 1.25–1.498.4512.2013.3011.8212.4810.899.072,978‐ 1.50–1.9919.3822.1518.0215.1516.4018.2316.657,486‐ 2.00–2.4919.3818.6315.9015.2212.6215.2617.0812,623‐ 2.50–2.9912.638.135.616.215.627.4812.0712,013‐ 3.00 or more33.2422.3116.1721.8119.9823.5228.6536,667Construction period of dwelling [time‐constant, distribution of person‐years exposure (%) and deaths (N)]‐ Before 191918.0130.9927.0128.6524.3925.8631.7517,669‐ 1919–194512.6914.8016.4815.0415.9716.4515.6010,456‐ 1946–196010.6411.4513.9512.4313.9412.4711.018,036‐ 1961–197010.078.4111.2812.3011.8810.448.239,843‐ 1971–198016.3710.1611.9211.5312.4412.1314.2115,372‐ 1981–199011.746.975.604.335.108.177.406,283‐ 1991–200012.549.156.996.539.569.757.575,110‐ 2001–20053.973.882.994.373.712.672.301,591‐ 2006 or later3.984.203.774.823.012.071.921,383N deaths66,5712,206971,2699216644,01575,743N individuals1,939,833149,09615,684146,66459,07925,50189,0952,424,952N person‐years9,441,797794,87494,502597,194275,607119,067407,61711,730,658Note: Legend Household Positions: (i) single (SING), (ii) married without children (MAR0), (iii) married with children (MAR+), (iv) child of a married couple (CMAR), (v) cohabiting without children (UNM0), (vi) cohabiting with children (UNM+), (vii) child of an unmarried couple (CUNM), (viii) head of a single parent household (H1PA), (ix) child of a single parent household (C1PA), (x) not family related household member (NFR), (xi) other household position (OTHER), and (xii) member of a collective household (COLL).Data sources: 2011 Census, Population Register (2001–2015) and annual tax returns (2011–2015), calculations by authors.

Journal

"Population, Space and Place" – Wiley

Published: Apr 1, 2022

Keywords: all‐cause mortality; Belgium; migrant mortality advantage; socio‐economic position; spatial variation

There are no references for this article.