This work was partially supported by the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie Grant Agreement No. 676060 (LONGPOP). Disclaimer: This publication reflects only the author’s view, and the Research Executive Agency is not responsible for any use that may be made of the information it contains.

1. Introduction

1The social and spatial trajectories of individuals’ life courses are thought to have life-long consequences for health and wellbeing [Morris et al., 2018; Pearce et al., 2018]. Both demographic and place characteristics are considered the main arguments to explain health outcomes in rural and urban settings all over the world. Yet, the possible pathways by which compositional and contextual factors influence health are challenging to determine since there is diversity in individuals’ lives and spatial mobilities through the life course [Sparks, 2012: 18]. Studies about mortality and contextual (place-based) characteristics have been often conducted by using multilevel models at area level in different time points [Naess et al., 2008; Naess and Leyland, 2010; Ohlsson and Merlo, 2011]. In this study, we propose an alternative approach based on the identification of long-term trajectory patterns by clustering individual sequences of residential trajectories over multiple birth cohorts, and their subsequent association with later life mortality risks.

2This paper examines the influence of social and long-term trajectories of municipalities of residence on old age mortality, a research field which has received little attention in historical research, particularly concerning the impacts of residential trajectories on later life mortality [Vallin and Meslé, 2004: 38]. There is considerable evidence that what the pioneer in gerontology Robert Barlet coined as the ‘longevity revolution’ – a rise in average life expectancy of more than 30 years in industrialized countries - was a long-term process that accelerated in the late 19^th century and is still ongoing today [Caselli, Meslé and Vallin, 1999]. A growing number of studies adopt a life course approach to analyze the impact of early life and adult conditions on older adults’ health and mortality [for example, Smith et al., 2014; Pudrovska and Anikputa, 2012; van der Linden et al., 2018]. In comparison, there is little work about the role of long-term residential trajectories in the construction of health in later life.

3Indeed, specific research about the long-term effects of residential patterns on later-mortality is scant. A number of studies have reported how mortality is associated with duration of migrant residence. During residence in different municipalities migrants are exposed to risky or, conversely, protective contexts in specific stages of life [e.g., Kesztenbaum and Rosenthal, 2011; Ginsburg et al., 2016]. However, in many of these studies, the exposure length or the duration of staying in a place is often collapsed in few categories and little time points. We need to consider trajectory indicators of places across the life course and not only those at one or two points in time that may underestimate the health-place relationship [Cummins et al., 2007]. Also, the study of specific mechanisms as mediating and moderating demographic factors may help to understand these life-place trajectory effects, build theory and, ultimately, enhance knowledge of the complexity of the ageing process and mortality [Spini et al., 2016: 681].

4Our study relies on a “life course of place approach” that is longitudinally applied to the study of later health and considers the contextual places inhabited through life as dynamic temporal and spatial processes that shape people’s social and health inequalities [Pearce et al., 2018]. Adapted from Bailey et al. [2016] and Gauthier and Viry [2019], we define “life-place trajectories” here as changing municipality of residence for each time of life from birth until early old age. Specifically, life-place trajectories across municipalities are conceived as residential pathways based on two aspects. On the one hand, the geographic proximity to the municipality of origin as a proxy of social support that an individual received. On the other hand, the exposure to residential contexts according to the population size of the individual’s place of residence. The spatial development and individualization processes which imply a dispersion of family members could imply that proximal support from families becomes more difficult, with diverse health implications for the elderly [Mulder and van der Per, 2009]. The residential history of different elderly generations according to geographic proximity to their families in origin, may also reflect inequality in the acquisition of resources within the narrowing of the rural-urban gap across the 20^th century [van Hooijdonk et al 2008]. By applying a process approach, we are able to provide long-term patterns of past residential mobility that might be associated with mortality later in life [Bailey et al., 2016: 200].

5Adopting a life course perspective - to study how patterns of past residential history based on proximity to origin and residential context affect later-life mortality across generations- implies the application of advanced analytical techniques. A valuable strategy for examining long-term longitudinal exposures in changing contexts is to apply state sequence techniques complemented with survival analysis [Gabadinho et al., 2011:3]. Therefore, the main research aims of this study are as follows: a) To identify the most typical patterns of life-place trajectories derived from two aspects that, based on the literature, we assume are important to health: proximity to support in the place of origin and residential context or environment; b) To study how old-age mortality was associated with these life-place trajectory patterns and possible early-life, adult and elderly pathways; c) To examine how these associations varied by birth cohorts and marital status.

6This study is based on individual-level data from birth cohorts born in the Netherlands between 1850 and 1922. Thus, we cover a broad period in the history of the Netherlands from 1850 to 2015, in which important health and societal changes were taken place. The Netherlands rapidly moved from a life expectancy below the European average in 1870 (37.7 years) to a life expectancy higher than expected by its national income between the 1920s and 1960s, due to a fast infant mortality decline [Mackenbach, 2020: 292]. Among the combined factors, a rapid economic progress, advances in medicine, particularly antibiotics, and high levels of literacy facilitated the introduction of modern hygienic measures from last quarter of 19^th century until at least mid-20th century, after which the expansion of the welfare state contributed to health improvements [Mackenbach, 2020: 113, 300]. However, in the second half of the 20^th century increase in life expectancy was much slower [Oeppen and Vaupel, 2002].

7Overall, this study makes a number of contributions to the growing body of research examining the influence of past trajectories on health and social outcomes from a life course perspective [Burton-Jeangros et al., 2015; Ritschard and Studer, 2018]. First, it is based on detailed population register data of multiple birth cohorts from the Netherlands, which allow to distinguish past longitudinal residential places across the 20th century, as it is expected that the influence of places on old-age mortality had a specific long time lag of residential exposures [Macintyre et al., 2002:135]. Second, the study adopts key principles of the life course paradigm by linking health and place domains, and by describing previous life-place trajectories to various historical contexts in multiple time points, for a more sophisticated understanding of the underlying processes affecting health and inequalities in later life [Pearce, 2015: 28]. To achieve this, we use Sequence Analysis, a data mining technique which allows to detect main patterns of individual residential trajectories from birth until age fifty. This is based on the proximity to the community of origin, cumulated distance covered over the life course and residential contexts. Third, we model old-age mortality according to clusters of life-place trajectories, testing the mediating and moderating effects of life course factors.

2. Theoretical framework and hypotheses

8Two integrated bodies of research bear on the general research question of examining the influence of life-place trajectories on health and mortality: a “life course of place” approach and a life course epidemiology perspective. First, the literature about life course of places and health conceives a dynamic approach to places where individuals resided during their life course, as diverse environments associated with later life health consequences. Our study follows the life course of place approach of Pearce et al. [2018]. In their work, these authors study the effect of area level social deprivation and green space accessibility over the life course on mental health in older age. In our study, we look at how residential pathways constructed from sequences of municipalities across the life course impact old-age mortality. Places not only offer opportunities and constraints in specific historical contexts, but are intertwined with individual resources and restrictions related to life course mechanisms, such as family strategies (e.g. low or higher support) and work (e.g. low or high socio-economic status). That is why living in specific places can have cumulative and non-reversible effects on later life outcomes, through social or epidemiological mechanisms [Elder et al., 2003; Cummins et al., 2007].

9Second, the life course epidemiology literature proposes three interrelated theoretical models to clarify the mechanisms through which trajectories can be explained and, at the same time, explain outcomes [Burton-Jeangros et al., 2015]. First, the “critical’ or “sensitive” period model posits that experiences during critical life course stages, such as infancy or adolescence, have lasting effects on people’s well-being [Lynch and Smith, 2005]. Second, the cumulative dis/advantage model underlines that life course trajectories reflect cumulative processes over time such as the exposure to risk or, conversely, protective factors in specific places and historical time with possible long-lasting consequences [Dannefer, 2003]. Several studies with at least three points of measurements of residential place [Culatti et al., 2014] have demonstrated the relative contribution of the cumulated risk area of residence over the life course to mortality among older age groups [Naess et al., 2008; Naess and Leyland, 2010]. Place of residence over the life course has been found to have a small influence on mortality among the elderly, but this influence may change over time [Ohlsson and Merlo, 2011].

10Third, the pathway model generally explains these processes as sequences of social exposures that, in a probabilistic way, interact on the risk of health decline and mortality in older age, taking into account possible sequential relationships at any given point in time [Braveman et al., 2009; Spini et al., 2016]. In practice, the cumulative model has been interpreted in different ways, and tested through an accumulation model of risk factors over time in additive ways, or through pathways models where the disadvantages are mediated or moderated by risk factors [Cullati et al., 2014: 19]. These considerations are relevant for our research objectives, which are to describe patterns of individual’s residential trajectories and to understand the socio-demographic mechanisms that mediate and moderate a possible association between these residential patterns from birth until age fifty and old-age mortality. Residential trajectories can be viewed as divergent careers, with positive and negative exposures over the life course mediated through diverse mechanisms, that may have a lasting effect on survival at old ages. We examine longitudinal exposure trajectories from different residential contexts over time and their mediation by diverse early life, adult and elderly factors from birth until death. In the following, and according to our research objectives, we formulate a series of hypotheses derived from the literature on health/mortality, demography and life course place studies.

2.1. Aspects of life-place trajectories (residential pathways) and mechanisms of ageing and mortality processes

11Two main aspects of life-place trajectories associated with health and mortality can be distinguished as follows: (a) proximity to the community of origin, and (b) residential context or environment over the life course [Gauthier and Viry, 2019: 192]. The first aspect refers to the longitudinal geographic distance/proximity from birth place, as a proxy of proximity to social support that an individual can potentially receive every time that she or he changes municipality of residence over the life course from family and friends of the community of origin [Gauthier and Viry, 2019]. Proximity is considered an important predictor of support between family members, as the larger the geographical distance between them, the lower the social support received [Knijn and Liefbroer, 2006; Mulder and van der Meer, 2009; Heady et al., 2010]. Contemporary studies in the Netherlands have shown, indeed, a decline in support exchange by distance. Growing distance was associated with a decrease in family support, which may be indirectly influenced by the individualization processes which imply a dispersion of family members for reasons of education, job or partnership [Knijn and Liefbroer, 2006; Mulder and van der Meer, 2009].

13At the same time, individuals may have a better stock of knowledge concerning spaces and the opportunities they offer if they cumulated past migrant experiences. This enhances their capacity to move, conceptualized as “motility”. Motility improves with an accumulation of cultural, relational and spatial resources [Viry et al., 2010]. These cumulative processes of advantage in individuals with more active migration histories may contribute to inequalities in terms of wealth, health or social capital in later life between mobile and non-mobile persons [Viry et al., 2010:18]. It may be also expected that individuals with distinct cumulative life-place-trajectories – in terms of diverse capital of resources– become unequal in later mortality. We can hypothesize that long-term trajectories with greater cumulative distances (km) from origin are related to a decreased risk of mortality in old age (Hypothesis H1b).

14Further, it is possible that the impact of life-place trajectories on old-age mortality varies by marital status. Research has found that marital status has a significant impact on mortality at old age, with negative health outcomes consistently identified in single, as well as in divorced or separated persons [Manzoli et al., 2007; Robards et al., 2012]. In this respect, a key current debate focuses on the importance of considering a life course or long-term marital history at old-age beyond short transitions, to provide a more thorough explanation of mortality outcomes [Robards et al., 2012]. However, as far as we know, few researches have examined the moderating role of marital status on the association between different definitions of migrant status and mortality, suggesting the buffer or protective effect of marriage on this association. Several studies have demonstrated changes in the risk of mortality among migrants by marital status, being those unmarried at higher risk of mortality than those who were married [Maxwell and Harding, 1998]. Long-distance return migrants have been found to have higher risks of mortality in Sweden, suggesting poor living conditions among low educated and unmarried returned men [Andersson and Drefahl, 2017]. These authors suggest that future studies on long-term migrant trajectories, old age mortality and proximity support would give more insights in this topic [Andersson and Drefahl, 2017: 10]. Therefore, we hypothesize as follows: Marital status moderated the detrimental association with increased risk of old-age mortality in specific life-place trajectories and old-age mortality. Particularly, this effect was more detrimental for widowed, divorced and unmarried persons with long-term distal trajectories from the community of origin over the life course, and weaker for married persons with these trajectories (Hypothesis 2).

15Longitudinal exposure to residential context constitutes a second core aspect of life-place trajectories associated with health. Part of the literature shows the negative health consequences of rapid urbanization and industrialization processes in which many European and US cities were involved in the last quarter of the 19^th century and, in particular, an excess mortality for persons with a long exposure in large cities [Oris and Alter, 2001; Cain and Hong, 2009; Kesztenbaum and Rosenthal, 2011]. This so-called urban penalty, which increased with urban expansion after 1850 in the Netherlands, has been identified as a result of a combination of factors including, among others: compositional effects, overcrowding, concentration of infectious diseases, lack of labor regulations and weak public health measures (water sewage systems) [van der Woud, 2010; Revuelta and Ramiro, 2016; van Rossem et al., 2017; Helgertz and Önnerfors, 2019]. This is particularly important for the long historical window in this study, from 1850 until the end of 20^th century. With the turn of the century, European countries followed different health processes at different times and speed in improvements of living standards, medicine and public health interventions, which lead to a mortality decline and higher life expectancy [Mackenbach, 2020: 5, 12].

16Aside from these determinants mentioned in the historical debate, the development of the welfare state in Europe after WWII has also been suggested to contribute to the mortality decline, and concretely in the Netherlands with the rapid expansion of individual healthcare expenditure between the 1960s and 1970s, followed by a certain stabilization and a peak in expenditures reached at the beginning of the 2000s [Mackenbach et al., 2011: 904, 911]. In this respect, the study of van Poppel and Joung [2001: 301] about long-term mortality differentials according to marital status in the Netherlands (1850-1970) suggested further research to test if older cohorts of migrants far away from home, particularly singles, would have had worst health in a context of pre-welfare state with a potential lack of family support, than younger generations who possibly benefited from health and social benefits. In a context of more welfare developments after 1945, we would expect that long-distance patterns from origin over the life course had less negative impact on mortality in older age among those younger than in older cohorts. We would expect, as well, long-distance moves to have become less dependent on good health over time, so the effect of poor health/kin support to become less important. Can we expect some mitigation or change over time in the selection? Therefore, we hypothesize that birth cohort moderated the detrimental impact that certain life-place trajectories based on proximity to origin had on old-age mortality. Specifically, the impact was weaker among younger cohorts with trajectories characterized by long-distance migration patterns from the community of origin than among older cohorts (Hypothesis 3).

17From a rural-urban perspective, it is still possible to debate about the timings of transitions from an urban penalty to an urban premium [Reis, 2009; Oris and Ramiro, 2016: 6; Kok et al., 2017] and whether certain residential trajectories were more advantageous concerning mortality at old-age in cohorts born between the end of the 19^th century, and first twenty years of the 20^th century, as compared to earlier cohorts. However, there are few studies about rural-urban mortality differentials at old-age in historical populations. Edvinsson and Broström [2012: 650] found no indication of higher mortality risks among the elderly living in unhealthy urban areas during the industrialization period in northern Sweden. Contemporary studies in the Netherlands offer mixed results, as some show a slight urban mortality advantage among the oldest old (+80) [van Hooijdonk et al., 2008], while others show higher urban risks among persons aged 65 years [Reijneveld et al., 1999]. These diverse findings may be produced partially by the diverse urbanity measures and the time points considered [van Hooijdonk et al., 2008: 504; Edvinsson and Brostom, 2012: 652]. Hence, whereas an old-age urban advantage among the young cohorts with long exposures in large cities is possible, more historical and recent evidence is required. To our knowledge, few studies have applied a processual perspective, consisting of longitudinal residential exposures within long life courses across historical cohorts [Pearce et al., 2018]. Based on these considerations, we hypothesize that, after adjusting for covariates, past life-place trajectories with long exposures in large cities elevated the risk of mortality in old-age mortality in cohorts born between 1850 and 1922, but this impact was weaker among younger cohorts with a long life-course exposure residing in large cities than among older cohorts (Hypothesis 4).

3. Methods

3.1. Sample

19Three main characteristics of HSN data make it a valuable tool for our research objectives: longitudinal residence information, sample stratification by birth cohorts, and life course features. Longitudinal municipal histories were derived from local population registers, which started in the Netherlands in 1850 with obligatory report of residential change for residents in the country. A stratified random sample of birth cohorts born between 1850 and 1922, with life course characteristics (childhood socio-economic status, adult occupational status, religious affiliation, marital history, residential history) was achieved from population registers, birth and marriage certificates.

3.2. Measures of life-place trajectories and additional covariates

22Finally, the third geographical indicator code the residential context from birth until fifty years old depending of the places where individuals spent all or a part of their lives. Statuses are: 1) Large cities with >100,000 inhabitants, 2) Large-medium cities, 25,000-100,000 inhabitants; 3) Medium cities, 5,000-25,000 inhabitants, 4) rural municipalities, less or equal to 5,000 inhabitants. The election of the number of statuses or categories was governed by a parsimony principle, keeping those necessary for our purposes with the maintenance of the essential information needed. As the number of sequences increases with the number of statuses, we needed to simplify the typology to the more meaningful modalities to our research.

3.3. Methodological research strategy

24In this study, we perform a series of analyses. First, Sequence Analysis (SA) is applied to describe longitudinal patterns of life-place trajectories. Hamming distance is used to measure the (dis) similarity between sequences in order to capture changes in time [Studer and Ritschard, 2016: 28]. According to comparison of quality measures, seven and four cluster groups, respectively, for life-place trajectories based on spatial proximity to origin and cumulated distance are retained as the best cluster solution (Figures 1 and 2). Another seven cluster groups are retained for trajectories based on residential context (Figure 3).

25Second, univariate analyses are performed to explore the possible inclusion of variables in continuous cox proportional hazard models. All variables were potential candidates according to results from the non- parametric long-rank test of equality applied to categorical variables and from the univariate cox proportional hazard regression used for the continuous variable rate of moves. Predictors with a p-value higher than 0,25 would be excluded from the final model. Finally, continuous Cox proportional regression models are applied to examine the association between cluster trajectories and old-age mortality after age 50 while adjusted by early-life, adult and later-life factors [Blossfeld and Rohwer, 2002].

26Four models for every residential trajectory type based on the proximity of origin, cumulated distance and residential contexts are analyzed (Tables 1, 2 and 3). Model 1.a. adjusts for sex, age and birth cohorts. In model 1.b, early-life factors are added. Model 1.c incorporates adult factors and model 1.d adds migrations after age 50. All three additive models control for sex, age and birth cohorts. Additionally, two two-way interaction analyses (one at the time) are conducted to examine how the association between life-place trajectories was moderated by marital status, and birth cohorts (Tables 4, 5, S4, S5). The focus of our research is on the interactive effects of life-place trajectories and marriage, and life-place trajectories and birth cohorts, but for readers who might be interested, we include results for the main effects. The specification of the models is determined, on the one hand, according to our hypotheses, and on the other hand, adjusted by the Likelihood estimation test and the Bayesian Information Criterion (BIC) statistics. These latter measures allow us to evaluate the fit of the models [Mills, 2011]. All statistical calculations are performed using the software Rstudio [R Core Team, 2019], as well as the TraMinerR package [Gabadinho et al., 2011] for sequence analysis and the WeightedCluster package [Studer, 2013] for cluster analysis. Time to mortality at old age is assessed for persons from age 50 to the death date, occurring from April 1902 until September 2015 (end of the study period).

27The validation of results is assessed using several sensitivity analyses. First, the models will be checked for multi-collinearity between independent variables and for collinearity in the interaction between variables of interest and trajectory-types, using variance inflation factor (VIF). There is no one standard criteria to interpret if a VIF shows multi-collinearity but it is commonly accepted that values greater than four and less than 0.2 require action [Smith and Smith, 2015]. In this study, it is not possible to include in same additive models the three categorical variables composing the trajectory clusters due to multi-collinearity (variance inflation factor (VIF) was > 200). We opted to run separated models for the three key typologies of trajectory clusters. Second, whereas the quadratic age as control does not alter estimates in models, the logarithm of age (ln) alters substantially coefficients in cluster trajectories and marital status. Age (time-variant covariate) is not collinear in any model and is our final inclusion. The inclusion of periods was disregarded as they suppressed most of demographic covariates effects, suggesting that ‘overcontrolling” could mask the effect of main mediators [Braveman and Barclay, 2009: 170]. Lastly, the Cox proportional assumption was tested with the scaled Schoenfeld residuals with the cox.zph function in R studio, for each covariate and a global test on all models. When the proportional assumption was violated, an interaction between age and the trajectory cluster was applied to deal with the non-proportionality.

4. Results

4.1. Patterns of individual’s life-place trajectories: findings from sequence analysis

28Figure 1 visualizes the seven trajectory clusters about residential trajectories based on proximity to origin by plotting their state distributions, with the ages on the horizontal axis from birth to 50 [Gabadinho et al., 2011]. The percentage of individuals in each state by cluster membership is relative to the size of the study subsample. Whereas cluster “High proximity-early migration” (13% of the subset) contains typically trajectories with a rapid short-distance migration from early childhood until adolescence, cluster “High-proximity-late migration” (16%) is characterized by a transition from birthplace to short-distance locations in adulthood. The trajectory type “High-medium proximity” (11%) follows a long-medium distance migration pattern from 11 to 40 km, while cluster “Medium proximity” (9%) represents individuals experiencing internal migrations from their birthplace to municipalities located between 41 and 100 km. Cluster “distal or low proximity” (8%) from the community of origin distinguishes long-distance migrations, and also starts during childhood but is more often experienced around age 39. Finally, cluster “Long stayers” (32%) comprises persons living in the municipality of origin from birth at least until late adulthood, and cluster “Stayers-NA” (11%) contains incomplete trajectories, particularly in later adulthood.

29Figure 2 shows four cluster groups of residential trajectories based on cumulated distance from origin. The first cluster is called “small cumulated distance”, comprising 59% of individuals in our subsample. Most of them correspond to the category of stayers but also to those persons with few residential moves, mostly concentrated in municipalities proximal to the birth municipality. There is almost no age variation. The second cluster “small-to-medium cumulative distance” (17%) represents individuals who cumulated a distance migrated from the municipality of origin between 50 and 250 km. The fourth cluster “small-to-larger cumulated distance” (13%) characterizes individuals who started cumulating great distance relatively early after age 20. Lastly, the third cluster (Small-NA cumulative distance, 11%) agglutinates individuals with a transition from small to medium cumulated distance who were lost of follow-up relatively early, with a peak around 38 years old. The quite similar shape of the state 0-50 km in both clusters at the bottom (Figure 1) may suggest long-distance migrations for the gap missing data but they could have also been frequent short distance moves.

30The distribution plots of the seven life-place trajectory patterns based on residential context are depicted in Figure 3. First, stable large city trajectories (cluster 4) represent 23% of individuals with residential continuity in this settlement type over fifty years. As the most common trajectory, it will be considered the reference category in regression models. Second, stable medium city trajectories (cluster 2) include about a similar percentage of cases and characterize individuals with a long spell residing in a medium-city context. Next, the stable large-medium city trajectories (cluster 7) correspond to individuals with a uniform pattern of residence in municipalities between 25,000 and 100,000 inhabitants, with few transitions to large cities. Fourth, stable rural trajectories (cluster 1) comprise 14% of individuals with residential continuity in rural contexts. The remaining three trajectories (clusters 3, 5 and 6) combine long spells with transitions to another residential context. Individuals who experienced a Rural-medium city-NA trajectory (cluster 6) typically started residing in rural and medium cities, followed by a missing transition to a possible different context with a peak around 32 years old. Another trajectory type, Medium-large city step migration (cluster 3), represents 10 % of persons and essentially concerns individuals moving from medium to large cities. Rural-urban step migration trajectories (cluster 5) include a first stage of individuals residing in rural contexts up to the age of 36-37. This is followed by a transition to a second stage in which people start living in urban context, becoming a stable state for the last fifteen years. Overall, the trajectories reveal substantial continuity over time for a majority of the study sample, particularly on the “stable large city” and the “stable medium city” trajectories, and discontinuity in the urban hierarchy for the “rural-urban” and “medium-large city” step migration, typically during young adulthood.

4.2. Multivariate results: effects of life-place trajectories on mortality in later life

31Table 1 shows findings from multivariable Cox regression analyses for the association of trajectories based on proximity to origin with the risk of mortality in later life. Tables 2 and 3 show these findings for trajectories based on cumulative distance from origin and trajectories based on residential context, respectively (see Supplementary material for full models, Tables S1, S2 and S3). Table 4 presents two-way interaction terms between trajectories based on proximity to origin and marital status, and Table 5 does the same for the interaction of these trajectories with birth cohorts (see Supplementary material for interactions between trajectories based on residential context with marital status and birth cohorts, respectively, Tables S4 and S5).

32Hypothesis 1a proposes that the mortality risk in old age relatively increased among individuals with trajectories of moderated-long distance from birthplace. As Table 1 indicates, stayers with loss of follow-up trajectories (Stayers-NA) were associated with lower mortality risks, after adjusting for sex, age and cohorts (Model 1a). This association remained significant, albeit modest, when adjusted for early-life, adult and later-life factors (Table 1, Models 1.b. ,1.c and 1.d). However, the risk of mortality was higher in persons with medium proximity and high-medium proximity trajectories when controlling for adult factors. Adjustment for potentially mediating migrations in later life increased the HRs (Hazard Ratios) in most of proximity trajectories, with the strongest effects in medium proximity trajectories and high proximity-late migrations (HR. 1.148 and HR. 1.182, respectively, Table 1). Thus, Hypothesis 1a is mostly confirmed, as the results show a relative significant association of four out of six trajectories with increased risk of mortality in old-age, mediated by later-life migrations. An interaction between cluster trajectories and age groups showed that the effects in medium proximity trajectories and high proximity-late migrations persisted among age ≥ 70 but were less significant (Supplementary material, Table S6).

Table 1. Association of past life-place trajectories based on residential proximity to origin and old-age mortality, estimated from Cox proportional hazard models (n = 20,359)

	M1a		M1b		M1c		M1d
	HR (95% CI)		HR (95% CI)		HR (95% CI)		HR (95% CI)
	1,029		1,045		1,05		1.182 ***
High proximity-late migration (ref: Long stayers)	(0.981-1.08)		(0.995-1.097)		(1.000-1.104)		(1.118-1.251)
High-medium proximity	1,041		1.050*		1.054*		1.105 ***
	(0.998-1.086)		(1.006-1.095)		(1.009-1.101)		(1.055-1.156)
Medium proximity	1,051		1,053		1.062 *		1.148 ***
	(0.996-1.108)		(0.998-1.112)		(1.005-1.121)		(1.08-1.219)
Stayers-NA	0.854***		0.861***		0.853 ***		0.893***
	(0.813-0.897)		(0.819-0.904)		(0.812-0.897)		(0.844-0.944)
High proximity-early migration	1,007		1,013		1,018		1.077**
	(0.962-1.054)		(0.967-1.06)		(0.972-1.067)		(1.027-1.13)
Distal proximity	0,986		0,997		0,995		1,018
	(0.935-1.041)		(0.945-1.053)		(0.942-1.051)		(0.959-1.082)

*p < 0.05, **p < 0.01, ***p < 0.001. Data are Hazard Ratios (HRs).
Model 1a adjusted for sex, age and birth cohorts, Model 1b M1a + adjusted for early life factors, Model 1c M1b + adjusted for adult factors, Model 1d M1c + adjusted for elderly migrations

Source: Own elaboration based on The Historical Sample of the Netherlands, Data Set Life Courses Release 2010.01 and Data Set Personal Cards version 3:51.

33From a cumulative spatial life-course perspective, Hypothesis 1b posits that long-term trajectories with larger number of cumulative distance (km) from origin are related to decreased mortality risk in later life. The results for the Small-to-larger cumulative distance trajectory support this hypothesis in the full model with later-life migrations as covariate, but the magnitude of the effect is not very large (Table 2, M2d). Moreover, Small-to-medium cumulative distance trajectories were associated with high mortality risk after adjusting for age, sex and cohorts (Table 2, Model 2a). The results remained similar after adjustment for early, adult and later-life factors, suggesting that these factors did not mediate the association between this trajectory and old-age mortality (Table 2, Models 2b, 2c and 2d). Then, Hypothesis 1b was supported, albeit the effect is modest.

Table 2. Association of past life-place trajectories based on cumulative distance from origin and old-age mortality, estimated from Cox proportional hazard models (n = 20,359)

	M2a		M2b		M2c		M2d
	HR (95% CI)		HR (95% CI)		HR (95% CI)		HR (95% CI)
	1.076***		1.077***		1.079***		1.079***
Small-to-medium cum. distance (ref: Small)	(1.036-1.118)		(1.037-1.12)		(1.038-1.122)		(1.036-1.123)
	0.873***		0.876***		0.863***		0.839***
Small-NA cum. distance	(0.834-0.915)		(0.836-0.917)		(0.824-0.904)		(0.799-0.881)
	0,981		0,984		0,979		0.934**
Small-to-larger cum. distance	(0.94-1.023)		(0.942-1.027)		(0.937-1.022)		(0.892-0.978)

*p < 0.05, **p < 0.01, ***p < 0.001. Data are Hazard Ratios (HRs).
Model 2a adjusted for sex, age and birth cohorts, Model 2b M2a + adjusted for early life factors, Model 2c M2b + adjusted for adult factors, Model 2d M2c + adjusted for elderly migrations

Source: Own elaboration based on The Historical Sample of the Netherlands, Data Set Life Courses Release 2010.01 and Data Set Personal Cards version 3:51.

34Regarding life-place trajectories based on residential context, Hypothesis 4 states that a prolonged exposure pattern in large cities was a risk factor of mortality in elderly people but that the negative effect of residing in such large cities was higher in the older than in the younger cohorts. As shown in Table 3, most of trajectories were associated with lower mortality risks compared to trajectories marked by long residences in large cities, when adjusted for sex, age and cohorts (Model 3a). These effects remained when early-life factors were incorporated into the model. However, with the exception of rural-medium city trajectories, which were associated with a lower risk of dying earlier in life (HR. 0.814), the rest were no longer associated with mortality after adjustment for adult factors, suggesting that they mediated the association of the trajectories based on residential context with the risk of old-age mortality. These results persisted significant after including later-life migrations (Table 3, Models 3b, 3c, and 3d). Moreover, contrary to expectations, those in younger cohorts following stable medium-large city trajectories showed higher risk of mortality, compared to those from old cohorts (HR. 1.180). No significant associations were found for the other trajectories based on residential context and changes over time (Table S5, Supplementary material). Therefore, according to these findings, we have to reject Hypothesis 4.

Table 3. Associations of past life-place trajectories based on residential context and old-age mortality, estimated from Cox proportional hazard models (n = 20,359)

	M3a		M3b		M3c		M3d
	HR (95% CI)		HR (95% CI)		HR (95% CI)		HR (95% CI)
Stable rural (ref: Stable large-city)	0.933**		0.932**		0,957		0,951
	(0.89-0.979)		(0.886-0.980)		(0.909-1.007)		0.904-1.001
Stable medium-city	0.932**		0.936**		0,955		0,971
	(0.894-0.971)		(0.893-0.982)		(0.911-1.002)		(0.925-1.019)
Medium-large city step migration	0,98		0,98		0,997		0,987
	(0.93-1.033)		(0.928-1.036)		(0.943-1.054)		(0.932-1.045)
Rural-urban step migration	0.935*		0.933*		0,956		0,972
	(0.879-0.995)		(0.876-0.995)		(0.897-1.020)		(0.91-1.038)
Rural-medium city-NA	0.814***		0.812***		0.816***		0.814***
	(0.773-0.857)		(0.77-0.856)		(0.773-0.861)		(0.77-0.860)
Stable medium-large city	0,965		0.951*		0,964		0,963
	(0.92-1.011)		(0.905-1.000)		(0.917-1.013)		(0.916-1.012)

*p < 0.05, **p < 0.01, ***p < 0.001. Data are Hazard Ratios (HRs).
Model 3a adjusted for sex, age and birth cohorts, Model 3b M3a + adjusted for early life factors, Model 3c M3b + adjusted for adult factors, Model 3d M3c + adjusted for elderly migrations.

Source: Own elaboration based on The Historical Sample of the Netherlands, Data Set Life Courses Release 2010.01 and Data Set Personal Cards version 3:51.

35The association between life-place trajectories and old-age mortality differed with marital status and birth cohort multiplicative interactions. We also present results for the main effects. The adjusted results suggested that unmarried persons after the age of 50 had a lower mortality risk than those who were married (main effects of marital status in table 4). However, this effect may vary according to life-place trajectory patterns. Hypothesis H2 postulates that marital status moderated the association between life-place trajectories geographically distant from home and old age mortality. As indicated in Table 4, following distal trajectories increased the mortality above 50 among widows/widowers and unmarried compared to married persons. Hence, hypothesis H2 is supported. Similar results were found for rural-to-medium city trajectories (Supplementary material, Table S4). Our results supported earlier studies indicating that old age mortality declined over time in 20^th century Netherlands. Compared to those born between 1850 and 1879, those born between 1880 and 1889 had a 16 percent lower risk of dying in old age, while those born between 1890 and 1922 had a 20 percent lower risk (main effects of birth cohorts in table 5). Hypothesis H3 postulates some beneficial effect of distal proximity trajectories from origin on old-age mortality over time. As indicated in Table 5, a negative association was larger in persons from younger cohorts (1890-1922) than those from old ones (1850-1879) who experienced distal proximity trajectories (HR.1.222), as well as those who followed medium proximity, high-medium proximity and high-proximity-later migrations (HR. 1.381, HR.1.308 and HR. 1.359, respectively). Thus, Hypothesis H3 is not supported.

Table 4. Old-age mortality, Hazard Ratios (HRs) for interaction between life-place trajectories based on residential proximity to origin and marital status

Model: main effects and interaction adjusted for demographic variables¹		HR (95% CI)
Marital status, main effect
Widowed (ref: married)		0,968	(0.908-1.033)
Divorced		0,911	(0.666-1.247)
Unmarried		0.884***	(0.833-0.938)
Life-place trajectories based on residential proximityₐ
High proximity-late migration (ref: Long stayers)		1,046	(0.965-1.132)
High-medium proximity		0,993	(0.929-1.061)
Medium proximity		1,002	(0.918-1.094)
Stayers-NA		0.656***	(0.596-0.722)
High proximity-early migration		0,973	(0.904-1.047)
Distal proximity		0.837***	(0.761-0.921)
Interaction: Life-place trajectories based on residential proximity x Marital Statusₐ
High proximity-late migration (ref: Long stayers)	x Widowed	1.198**	(1.066-1.347)
High-medium proximity	x Widowed	1,105	(0.999-1.222)
Medium proximity	x Widowed	1.206**	(1.058-1.375)
Stayers-NA	x Widowed	1.554***	(1.375-1.756)
High proximity-early migration	x Widowed	1.225***	(1.096-1.370)
Distal proximity	x Widowed	1.352***	(1.185-1.543)
High proximity-late migration (ref: Long stayers)	x Divorced	1,493	(0.868-2.569)
High-medium proximity	x Divorced	1,211	(0.777-1.889)
Medium proximity	x Divorced	1,691	(0.997-2.868)
Stayers-NA	x Divorced	1,538	(0.874-2.706)
High proximity-early migration	x Divorced	1,236	(0.745-2.049)
Distal proximity	x Divorced	1,393	(0.808-2.403)
High proximity-late migration (ref: Long stayers)	x Unmarried	1.258***	(1.118-1.416)
High-medium proximity	x Unmarried	1.361***	(1.219-1.521)
Medium proximity	x Unmarried	1.296***	(1.140-1.475)
Stayers-NA	x Unmarried	1.588***	(1.397-1.805)
High proximity-early migration	x Unmarried	1.131*	(1.012-1.263)
Distal proximity	x Unmarried	1.365***	(1.200-1.553)

*p < 0.05, **p < 0.01, ***p < 0.001. Data are Hazard Ratios (HRs).
¹Model adjusted for sex, age, birth cohort, birth region, father’s occupational status at birth, own occupational status at age 50, religious affiliation, ratio of moves after age 50.
^ₐ Main effect of life-place trajectories according to residential proximity
^ₐ Marital Status, reference category Married

Source: Own elaboration based on HSN.

Table 5. Old-age mortality, Hazard Ratios (HRs) for interaction between life-place trajectories based on residential proximity to origin and and birth cohort membership

Model: main effects and interaction adjusted for demographic variables¹		HR (95% CI)
Birth cohort, main effect
1880-1889 (ref: 1850-1879)		0.834***	(0.786-0.886)
1890-1922		0.680***	(0.638-0.726)
Life-place trajectories based on residential proximityₐ
High proximity-late migration (ref: Long stayers)		1,058	(0.958-1.168)
High-medium proximity		1,025	(0.947-1.111)
Medium proximity		0,998	(0.898-1.109)
Stayers-NA		0.834**	(0.740-0.940)
High proximity-early migration		1,041	(0.952-1.138)
Distal proximity		0,925	(0.807-1.061)
Interaction: Life-place trajectories based on residential proximity x Birth cohortₐ
High proximity-late migration (ref: Long stayers)	x Birth cohort 1880-1889	1,049	(0.928-1.186)
High-medium proximity	x Birth cohort 1880-1889	0,989	(0.891-1.099)
Medium proximity	x Birth cohort 1880-1889	1,12	(0.980-1.280)
Stayers-NA	x Birth cohort 1880-1889	1,059	(0.915-1.226)
High proximity-early migration	x Birth cohort 1880-1889	0,99	(0.883-1.110)
Distal proximity	x Birth cohort 1880-1889	1,115	(0.940-1.323)
High proximity-late migration (ref: Long stayers)	x Birth cohort 1890-1922	1.359***	(1.198-1.541)
High-medium proximity	x Birth cohort 1890-1922	1.308***	(1.173-1.460)
Medium proximity	x Birth cohort 1890-1922	1.381***	(1.204-1.585)
Stayers-NA	x Birth cohort 1890-1922	1.187*	(1.034-1.362)
High proximity-early migration	x Birth cohort 1890-1922	1.144*	(1.016-1.287)
Distal proximity	x Birth cohort 1890-1922	1.222*	(1.049-1.423)

*p < 0.05, **p < 0.01, ***p < 0.001. Data are Hazard Ratios (HRs).
^ₐ Main effect of life-place trajectories according to residential proximity
Birth cohort, reference category 1850-1979
Model adjusted for sex, age, birth region, father’s occupational status at birth, own marital status, own occupational status at age 50, religious affiliation, ratio of moves after age 50.

Source: Own elaboration based on HSN.

5. Discussion and conclusion

36During the last decades, there has been considerable research aimed at enhancing knowledge about demographic risk factors for mortality in adulthood. When the residential context was considered, a static approach was often adopted, probably resulting in overlooking heterogeneity in mortality inequalities at old-age. Therefore, this paper sought to analyze the dynamic interplay between residential trajectories throughout the life course and inequalities in old-age mortality from a long-term perspective.

37The analyses in this study were mainly based on explanatory models and demographic mechanisms of old-age mortality risks. By using descriptive cluster analysis to uncover long-term migration patterns and combining it with semi-parametric Cox proportional hazard models to estimate mortality risks in old age, this paper added to the body of knowledge on the association between longitudinal residential places and health outcomes as well as to the discussion about rural-urban mortality differentials among elderly people in historical populations [Naess and Leyland, 2010; Edvinsson and Broström, 2012; Pearce et al., 2018]. These models are appropriate when socio-demographic mechanisms of old-age mortality risk, including long-term migratory trajectories, are poorly understood.

38Our first aim was to identify longitudinal patterns extracted from individuals’ life course trajectories of municipality of residence and creating two proxy indicators: (1) access to support in the place of origin, and (2) exposure to a specific social and health environment- during their first fifty years of life in the Netherlands during the 19^th and 20^th centuries. First, we found seven distinctive trajectories from birth until young-old age based on the proximity or distance to the kin network of origin. About one- third of people followed non-mobile trajectories (long stayers), and a high-proximity and late migration trajectory emerged as the second most typical one. In addition, we observed that four distinct patterns described the cumulated distance from home through the life course until fifty years of age. We found that the most common pattern was a small cumulative distance trajectory, followed by a small-to-medium trajectory. With regard to life-place trajectories based on residential context, we detected seven patterns and observed that over three quarter of persons followed trajectories of continuity in a similar residential context. The most common patterns were stable large city trajectories and stable medium city trajectories. Even if this study did not allow going into detail about the kin spatial proximity and frequency of contacts, to our knowledge, it is the first research to examine the protective and risk effects of these long-term residential patterns of migration on old-age mortality for a large number of cohorts across the 20th century in the Netherlands.

39Following the life course principle of linked residential and health domains, our second aim was to determine to what extent these patterns were associated with mortality inequalities at old-age, and if this association was mediated by early-life, adult and elderly factors. A major finding of this study was that individuals with trajectories of moderate-long distance from birthplace had a relatively higher risk of mortality in old age compared to long-term residents. In particular, those who followed medium proximity trajectories and high proximity-late migrations from origin had a 11–18 percent higher risks of mortality after age 50. In this regard, a second major finding in this study was that marital status and later-life migrations mediated the relationship between moderately long-distance patterns from birth to late adulthood and an elevated risk of mortality in old age. These mechanisms suggest that living far from one’s family of origin may lead to a reduction in in-person emotional support. This may be related to the cumulative effect of weak social networks and unhealthy living situations, which may result in a decline in the amount of community support in old age [Shor and Roelfs, 2015].The study’s overall findings provided certain support for the pathway model, particularly when it comes to family support in the community of origin, which was determined by measuring longitudinal residential closeness to birthplace across the first 50 years of life. Hence, this study may serve as an argument for the development of more social capital and support measures that may mitigate group differences in the patterns of old-age mortality among migrants.

40At this stage, however, our results of main effects are in line with previous studies reporting a weak effect of different measures of social support on mortality, after adjusting by additional factors [Shor and Roelfs, 2015; Nyqvist et al., 2014; Choi et al., 2014]. This is suggestive that life course proximity to origin is not a good enough condition to benefit health in later life. Social support can indeed not only come from the group of origin, but also from the family created by the individuals we follow. In that perspective, our findings suggest a protective effect of marriage on mortality, which call for further research into gender-specific differences and should account for the effect of living close to children [van der Berg and Gupta, 2015; Tatangelo et al., 2017]. However, in comparison with previous research, our study is expanded to examine the influence of past life-place trajectories proximal to the community of origin, and both the mediating role of marital history and the more recent elderly migration history, following the life course dynamic concept of “the shadow of the past” [Bernardi et al., 2020]. Indeed, as we show past life trajectories are not often a linear processes across time. Regarding the negative mediating role of elderly migrations between past life-place trajectories and mortality, a possible explanation might be that migrations in later life are found to be more likely in persons with deteriorated health [Walter, 2002; van de Pers et al., 2015], suggesting that the more recent past matters more for mortality in old age.

41Our third aim was to study how marital status and birth cohorts moderated the association between life-place trajectories and mortality at old age. Two final major findings were observed. On the one hand, past life-place trajectories were associated both with increased or decreased survival in heterogeneous groups of elderly persons. As compared with married persons, widows, widowers and unmarried persons with past distal trajectories from origin were found to be particularly vulnerable to mortality. On the other hand, the study results indicated a substantive negative association between certain trajectories and old-age mortality over time. Specifically, individuals of the younger cohorts with long-distance patterns from origin as well as those who had medium proximity, high-medium proximity and high-proximity-later migration patterns were associated with a 30-38 percent increased risk, after adjusting for demographic factors. These findings were somewhat unexpected considering van Poppel and Joung’s [2001] suggestion that individuals from the older cohorts who migrated far from their origin places had worse health in a pre-welfare state context as opposed to those from the younger cohorts of migrants with better health who benefited from welfare advancements after WWII. The observation that long-term marital status moderated the association between life-place trajectories and mortality in old age helps to interpret this result. Further analyses suggest that this was particularly the case of unmarried persons from the younger cohorts with residential history patterns far distant from the community of origin (results available on request). According to literature on migration, marriage, and mortality in the elderly, late adult migration may be a disruptive process associated with the effects of divorce and widowhood, meaning that persons without social support after loss of partner could have higher health risks and higher tendency to seek support elsewhere [Manzoli et al., 2007; Robards et al., 2012].However, our research goes beyond this by considering the conjoined effect of long-term trajectories distal from origin over the life course and widowhood, as a combination of two proxy measures of weak social relationships with negative consequences in later life.

42Concerning old-age mortality risks associated with life-place trajectories based on residential context, the results first suggested that the positive influence of most trajectories were inhibited by adult factors, excepting for rural to medium-sized cities trajectories. Indeed, whereas these results meet the initial expectation of a large urban mortality disadvantage at old age, a suppressive positive effect due to potential benefits of rural residence patterns – or other combined patterns with medium cities- seemed mediated through marital status. This result again supports previous studies that emphasized the protective effect of marriage on longevity [Goldman et al., 1995]. Furthermore, long rural to medium residential patterns appeared to be beneficial for survival at old-age. This association was largely independent of early, adult and later life factors, suggesting an impact of this residential trajectory itself, in contrast to mortality differentials explained by socio-demographic characteristics. Interpretations are limited due to many missing follow-up places concentrated in this pattern.

43Next, we did not find, as expected, a clear advantage associated with trajectories living in large cities over time in 20^th century Netherlands. However, our results over time are overall in line with current studies reporting a lack of urban-rural mortality differences [House et al., 2000; Congdon et al., 1997] and, to some extent, corroborate previous findings in contemporary Netherlands of small urban-rural differences in mortality, after adjustment of population composition and socio-economic neighborhood [van Hooijdonk et al., 2008] or no differences for perceived health among persons moving into urban or rural areas, controlling for socio-demographic factors [Verheij et al., 1998]. We further found that widowed and unmarried persons with rural to medium city trajectories had increased mortality risks in old age compared with those with large city trajectories. These results in general are in line with evidence from van Hooijdonk et al.’s study [2008] which also has shown lower mortality risks among widowed persons and never married in the most urban category.

44The research’s findings are about the Netherlands, where profound societal changes occurred during the study period. The latter half of the 19th century and at least the middle of the 20th century saw fast economic growth, high literacy rates, medical advancements, a decrease in the mortality gap between rural and urban areas, the development of the welfare state, and subsequent improvements in health. Thus, the findings might be relevant to countries with comparable societal developments over a similar time period. However, the Netherlands is a relatively small country with a high population density and an important urban network already before the industrialization. These particularities limit the potential to generalize our results on migratory trajectories. But the discoveries on the associations between these trajectories and mortality in old age call for cross-country comparisons across time and space that take into consideration the contextual and geographic differences between rural and urban areas over the life course. This would shed light on the extent to which the study’s findings are valid to other societies and periods or only to the Dutch context.

45The HSN is a longitudinal representative sample of multiple cohorts in the Netherlands, but it is worthy to mention that it does not contain measurements for other recognized predictors of mortality. Some of the most relevant considerations in the study of old-age mortality particularly include psychological predictors such as perceived social support, subjective well-being and intellectual functioning [Maier and Smith, 1999], health behaviors and genetic factors [van den Berg et al., 2019]. Also, the comparability of findings deserves caution as the studies use different indicators of urbanicity which may lead to different results [Gheasi et al., 2019]. These limitations should be considered for interpreting results.

46In the future, a promising research perspective would be to analyze gender differences in migratory trajectories and old age mortality, which deserves specific investigation. We also aim to explore the impact of migration histories on cause-specific mortality requiring linkage of the Historical Sample of the Netherlands to individual cause-of-death data [Yeung et al., 2014]. We also expect to apply other innovative methods to integrate diverse life domains and critical life course transitions [Eerola, 2018], which are useful to demographers and migration scholars interested in explaining life course trajectories. Our analytical strategy can be highly useful to approach part of the complexity of the ageing process by studying the influence of longitudinal socio-residential patterns on mortality from communicable and non-communicable diseases, other health and wellbeing outcomes across adult and elder populations, different historical time and geographic scales.