Who is aging out of place? The role of migrant selectivity in international retirement migration

ABSTRACT

International retirement migration gained popularity with the rise of globalisation and individualisation, but little is known about whom the retirement migrants are compared to retirees who do not migrate. To gain insight into who migrates compared to who stays, we examine a broad set of individual determinants. We collected data for the survey of Dutch Retirement Migrants Abroad, a new dataset based on a probability sample of Dutch nationals with an oversample of retirement migrants (ages 66–90). The survey includes 5225 migrants who migrated from the Netherlands and permanently reside in one of forty different destination countries and 1339 Dutch retirees who reside in the Netherlands. Using discrete-time event-history models, we test the effect of socioeconomic status, social ties, personality traits, and cultural values on the likelihood of migration. Having a partner and a higher occupational status raised the likelihood of migration. Additionally, retirement migrants were more likely to be adventurous, postmaterialist, and identify with counterculture of the sixties, such as being involved in the hippie culture, than non-migrants. Having more social ties in the Netherlands decreased the likelihood of migration. This study highlights the complex interplay of determinants influencing who migrates at older ages and who stays.

KEYWORDS:

• International migration
• retirement
• migrant selectivity
• event-history

Moving abroad around retirement has become a viable option for many older adults looking for a new lifestyle in the next chapter of their lives with the rise of globalisation, individualisation, and advances in cheaper travel. Examples of international retirement migration flows are from higher-income countries, such as the UK, US or Japan, to lower-income countries, such as Spain, Mexico, or Thailand. This type of migration challenges our understanding of aging in place, which assumes that people prefer to stay in their neighbourhoods as they get older to be with their family and existing social network (Wiles et al. 2012). It suggests that a substantial number of older adults are not satisfied with the post-retirement lifestyle options in their country of origin and instead choose to retire abroad. While the motives of international retirement migration have received much attention (Casado-Díaz 2006; Hayes 2014; Rodriguez, Fernandez-Mayoralas, and Rojo 1998), the determinants of retirement migration have been studied less frequently (Savaş et al. 2023). In this article, we quantitatively examine a comprehensive set of characteristics that foster migrating to a new country around retirement.

Research investigating the characteristics of retirement migrants often focused on their demographic and socioeconomic profiles (Bahar et al. 2009; Casado-Díaz 2006; Huber and O’Reilly 2004; Williams and Patterson 1998). According to these studies, most retirement migrants were married and living in two-person households, had middle to high levels of education, and had pre-retirement jobs requiring high-level skills. However, recent studies have shown that the characteristics of retirement migrants are diversifying over the years to include those who move to healthcare facilities (Bender, Hollstein, and Schweppe 2017) and those of lower socioeconomic status (Repetti, Phillipson, and Calasanti 2018; Truly 2002). Beyond socio-demographic characteristics, qualitative studies offered in-depth insights into sociological and psychological characteristics of specific migrant groups. For example, recent research focuses on single retirees (Bell 2015; Bender and Schweppe 2022; Gambold 2013; Thang and Sone 2011; Thang, Sone, and Toyota 2012). Bender and Schweppe (2022) reported on single men from Germany or Switzerland who were looking for new partners in the destination country. Thang and colleagues (2012) showed that single retirement migrants from Japan residing in Australia or Thailand had reasons involving ‘staying away from the confines of Japanese structures’ (250). These studies show the importance of examining sociological and psychological characteristics in addition to socio-demographic characteristics as they shed light on the decision-making process that influences people’s choices to migrate after retirement.

Researchers have taken different approaches to studying characteristics of retirement migrants, but generally focused on one or a few destination countries. Comparative studies such as Rodriguez and colleagues (1998) studied retirement migrants of different origin countries (UK, Germany, Nordic countries, and Benelux) in Spain, and King and colleagues (1998) investigated British retirement migrants in four different regions of the Mediterranean. Despite valuable insights existing research provided into who retirement migrants are in different destination countries around the world (King, Warnes, and Williams 1998; Lizarraga, Mantecón, and Huete 2015; Rodriguez, Fernandez-Mayoralas, and Rojo 1998; Rojas, LeBlanc, and Sunil 2014; Unguren, Tekin, and Bayırlı 2021), they are less suitable for getting a comprehensive picture of how international retirement migrants differ from those who stay behind.

Migrant selectivity, a concept from classic migration research, suggests that migrants are not a random selection of individuals but differ systematically from stayers (Ichou 2014; Polavieja, Fernández-Reino, and Ramos 2018). Migrant selectivity is argued to have implications for the social and economic integration of the migrants. For example, positive selection in terms of educational attainment has been associated with higher earnings of first- and second-generation migrants (Borjas 1993). However, it is difficult to draw conclusions on migrant selectivity regarding retirement migration due to the lack of a ‘control group’ in the origin country. It is also difficult to disentangle which characteristics existed before migration and which were acquired or changed after, as the characteristics inquired about in surveys regarding retirement migration were often not retrospective. For example, income is likely to be different from the moment of migration to the moment of inquiry, which affects the evaluation of how socioeconomic status is associated with the likelihood of migration. Additionally, qualitative research, while providing valuable demographic, sociological and psychological insights (Banks 2004; Gustafson 2001; Oliver 2011), face challenges in generalising findings due to small sample sizes and the non-representative data frames. Thus, a more comprehensive analysis with a representative sample is essential to understanding the determinants influencing selectivity of retirement migration.

In this study, we make theoretical and methodological advances by investigating the determinants of migration to a new country around retirement age. Our contribution is threefold. First, we collected representative data from retirees born in one origin country (the Netherlands), migrated around retirement age, and were living permanently in one of forty destination countries at the time of the survey (2021). These destinations included frequently studied destination countries, such as Spain, and destinations that are investigated less often, such as the Scandinavian countries (Appendix). The data collection allowed us to get a comprehensive view of retirement migration from one origin country (Dutch Retirement Migrants Abroad; Henkens et al. 2021); however, we did not focus on the variations created based on the destination country. Second, we employed a novel stratified retrospective design that combined our survey with data on a representative sample of stayers. Using similar measurements in both samples, we developed an event-history model to test hypotheses about the determinants of international retirement migration. The event-history analysis aims to explain why certain individuals are at a higher risk of experiencing an event than others (Vermunt 2009), which in this case we used to test a discrete event, migrating versus not migrating. The discrete-time event-history analysis can include time-constant variables, such as educational attainment, and variables that change over time, such as children’s ages. Third, we developed and tested hypotheses on the impact of socioeconomic status, social ties, personality traits, and cultural values on international retirement migration. Although these themes have received attention in previous literature, we broaden the understanding of how the combination of these four sets of characteristics affects who migrates compared to who stays in a quantitative manner.

The context of this study is the Netherlands, a densely populated country with a high GDP per capita in Europe. In the past few decades, the retirement age in the country was 65, but this age started increasing with the growing aging population. At the moment of this research (2021), the Netherlands had an official retirement age of 66.3 years. Dutch residents acquire 2% of state pension for every year they lived in the Netherlands in the fifty years prior to their official retirement age, in addition to the pension schemes provided by their work. The Netherlands has a long history of emigration that fluctuated in the nineteenth and twentieth centuries. A study on recent emigration from the Netherlands showed that age, education, income, and social networks played key roles in emigration decision (Van Dalen and Henkens 2007). Furthermore, the number of retirement migrants has risen since the 2000s (Van Dalen and Henkens 2008, 77), with approximately 24,000 registered retirement migrants receiving their pensions abroad in 2021 (Henkens et al. 2021).

Framework and hypotheses

Socioeconomic status

Education and occupation are amongst the most studied determinants of international migration. The focus on these variables is motivated by the human capital theory, which emphasises that migration is an investment with costs and returns. An individual decides to move when the future benefits outweigh the expected costs (Borjas 1989; Sjaastad 1962). Most often, those with higher human capital migrate as they can afford to do so (De Haas et al. 2019).

Higher-educated people have better job opportunities and higher salaries (Pregi and Novotný 2019; Spörlein et al. 2020), and especially important for the present case, a higher language efficiency (Chiswick and Miller 2014). In the case of retirement migration, language efficiency is deemed a crucial part of integration in the destination country as it is needed for social interactions and for instances where one needs to obtain information or care. However, many retirement migrants struggle to be fluent in the destination language (Savaş et al., 2023). We hypothesise that higher-educated people would be more likely to migrate than less-educated people.

Occupational status is relevant for international migration as it strongly relates to income and wealth, especially for older cohorts (Ganzeboom, De Graaf, and Treiman 1992). Several surveys on retirement migration showed that retirement migrants belonged to a higher occupational class (Bahar et al. 2009; King, Warnes, and Williams 1998; Rodriguez, Fernandez-Mayoralas, and Rojo 1998), while other researchers showed that over the years, many migrated to manage their finances as they were not affluent in their country of origin (O’Reilly 2007; Repetti, Phillipson, and Calasanti 2018). However, these findings only reflect the characteristics of the retirement migrants in the destination and do not take into account the selectivity of the migrants. Migration might not be attractive for individuals of the highest occupational status, as they could own houses in both countries and have transnational practices rather than living permanently in a new country (e.g. seasonal retirement migrants). Additionally, for those with the lowest occupational status, the costs of migration might be too high, making them less likely to have opportunities to migrate to a new country. For those in the middle occupational status, moving to a country with a lower cost of living could help them live a more luxurious life, maximising their benefits while handling the costs of migration. We hypothesise that there will be a curvilinear effect of occupational status: those in middle occupational status positions would be more likely to migrate than those with lower or higher occupational status.

Social ties

Social ties are crucial not only for migrants’ integration and well-being in the destination country (Casado-Diaz 2009) but also for healthy aging as they provide support and other resources in times of need (Cornwell, Laumann, and Schumm 2008). For retirement migrants, two types of social ties exist: ties in the destination country and in the origin country. While ties in the destination country may increase the likelihood of moving by enhancing the network and support before and after the migration process (Williams et al. 2000), ties in the origin country may reduce the likelihood of moving as existing social ties become more important with age (Fingerman et al. 2020; Lubben and Gironda 2003).

Different actors come into play while considering social ties. Family is an important source of support, especially for older adults (Litwin and Landau 2000). Around the age of retirement, family ties may include parents, children, and grandchildren simultaneously. Each of these family members can play a role as an anchoring tie to the origin country. Additionally, considering that people create community and civic ties in the country of origin throughout their lives through activities such as volunteering and community involvement, these ties could also play a role as anchoring ties to the origin country.

We first hypothesise that people with a partner will be more likely to migrate than singles; although migrating might be a way to search for a new partner for singles, having a partner would be an essential support system during the migration process. Additionally, we expect a positive effect of having a partner with a migration background (partner or the partner’s parents born outside of the Netherlands) compared to having a Dutch partner, as a partner with a migration background might have ties in the destination country and fewer ties in the Netherlands.

Second, we hypothesise that people whose parents are alive would be less likely to migrate compared to people whose parents have passed away, as they would have fewer family obligations to the origin country. Additionally, we hypothesise that people with children or grandchildren would be less likely to migrate than those without children or grandchildren. The anchoring effect might be absent for younger children as they are likely to migrate with their parents. We also do not expect a relationship between having older grandchildren and the likelihood of migration, as grandparental support would be required less often once the grandchild is an adult. We further hypothesise that people with a stronger civic or community engagement in the country of origin (e.g. volunteering) between the ages 50–65 are less likely to migrate than people with a weaker civic engagement in the origin as these ties would act as anchoring ties to the origin.

Personality disposition

Several studies suggest that personality characteristics might aid the migration process (Canache et al. 2013; Silventoinen et al. 2008). Certain personality traits aid people to perform better in novel situations; migration to a new country is a case in point. In research studying interstate migration of older adults in the US, higher levels of extraversion and openness to experience predicted a higher likelihood of migration (Crown, Gheasi, and Faggian 2020; Jokela 2009). These personality traits can have the same effect in the case of international retirement migration, as they affect how people approach new situations and cope with the challenges of migration, such as making new social ties. While qualitative studies have explored various psychological aspects of retirement migrants, such as perceiving migration as a mean to chase new adventures (Hayes 2018), as well as showing that retirement migrants’ identities evolve following migration, for some helping them create ‘new’ identities (Oliver 2011), personality traits of retirement migrants in comparison to stayers is yet to be investigated.

We study personality traits in retirement migration by investigating adventurousness, a facet of openness to experience (Goldberg et al. 2006), and extraversion. We chose the adventurousness facet instead of openness to experience as the questions regarding adventurousness were more appropriate for the case of retirement migration (see measures section for the questions). We hypothesise that adventurousness will be positively associated with the likelihood of migration. However, we expect a more nuanced effect of extraversion. Highly introverted people might be less affected by the decreasing face-to-face contact with their social ties in the origin country; hence, the barrier to migrate would be lower. Additionally, extroverted people would also be more likely to migrate as it would be more pleasurable to make new ties in the destination country with their excitement-seeking and sociable qualities (Goldberg 1993). Thus, we hypothesise a curvilinear effect of extraversion on the likelihood of migration, with people higher and lower in extraversion being more likely to migrate than those positioned in the middle.

Cultural values

Cultural sociology has emphasised the importance of the era’s values in shaping people’s identities. Our sample mostly consists of baby boomers (born in 1940–1950s), who have been the centre of much research due to the value and belief systems that have changed in their lifetime. One of the most important cultural changes is the shift from materialist to postmaterialist views following an increase in prosperity (Inglehart 1990). A cultural shift to postmaterialism is associated with an emphasis on non-material needs, such as a sense of community, self-actualisation, and quality of life, over material needs, such as physical security and wealth. Although most research has studied the effects of postmaterialism on economic and political outcomes, some studies have made links to migration. One study showed that, in Germany, those who held postmaterialist values were more likely to express migration intentions (Samarsky 2020). Although the research’s sample was young, we expect similar effects among older adults. A study of international retirement migration emphasised migrants’ desire for ‘self-fulfillment’ (Hayes 2018), which is often associated with postmaterialist values. Following these arguments, we expect those with postmaterialist values to be more likely to migrate than those with materialist values.

A related indicator of a cultural shift baby boomers experienced lies in the emergence of countercultural movements in the 1960s and 1970s. The 1960s and 1970s are associated with an experimental lifestyle, political activism, and a direction against consumer society, giving rise to different countercultures, such as the hippie culture (Braunstein and Doyle 2002). Previous research showed that a strong counterculture identification in youth was linked to more active retirement views, such as seeing retirement as a new beginning rather than a phase where one slows down and diminishes activity (Tunney, Henkens, and van Solinge 2022). One of the effects of identifying with the counterculture could be a stronger propensity to migrate after retirement due to the non-conformist nature of the counterculture identity. Thus, we hypothesise that those who used to identify with the counterculture of the 1960s and 1970s will be more likely to migrate than those who did not identify as such.

Methods

Data and sample

To test our hypotheses, we collected data from Dutch retirees living abroad (migrants) and Dutch retirees residing in the Netherlands (non-migrants). The sample was drawn by the Social Insurance Bank (SVB), which executes the public pension system in the Netherlands. The data of the SVB covered the entire Dutch population. The population was defined as people who were born in the Netherlands, who were between the ages of 66–90 in 2021, who lived at least 35 years in the Netherlands after reaching age 15, and who were receiving their pension in a country outside of the Netherlands (Henkens et al. 2021). This assured that our respondents’ residence was the destination country, excluding those who migrate seasonally. We limited the population to the forty most common destination countries, thereby covering 98% of the population (Appendix). People who lived in Belgium or Germany were excluded beforehand as these countries are in very close proximity to the Netherlands, often involving border migration. Return migrants, people who had initially migrated to the Netherlands when they were younger due to reasons such as employment but later returned to their country of origin upon reaching retirement age, were excluded as this would require a separate conceptual and empirical treatment. A probability sample was drawn from the population. The sample was contacted via SVB and our fieldwork agency for a web-based or paper-and-pencil questionnaire. The response rate was 45% resulting in an effective sample size of 6110. Further information on the sample and fieldwork can be found in the codebook (Henkens et al. 2021).

In order to match the migrant group to a group of non-migrants residing in the Netherlands, a survey with similar questions was carried out via the LISS panel (Longitudinal Internet studies for the Social Sciences) administered by CentERdata (Tilburg University, The Netherlands). LISS is a representative study of the Dutch population. The non-migrant group included 1364 Dutch citizens between the ages 66–90 in 2021, born and still residing in the Netherlands.

For this study, from the migrant group, we included only those who migrated after age 50 in our sample. This decision was made as mobility related to traditional retirement transition starts at age 50, (Sander and Bell 2014). We excluded those who did not fill out their gender, work status, retirement age, and migrants who reported living in the Netherlands. From the non-migrant group, we excluded those over 90 to match our age range with the migrant sample. Thus, our sample consisted of 6564 participants (65.9% male, 79.7% with a partner, M_age= 73.95, SD_age = 5.18), of which 1339 stayed in the Netherlands and 5225 migrated.

Event-history analysis

A discrete-time event-history analysis was used to examine predictors of retirement migration (Allison 1984). This analysis was chosen for four main reasons: (1) it provides estimates of which individuals are more likely to experience an event than others, (2) it considers not only whether someone migrated but also the timing of migration, (3) it accounts for censoring in the non-migrant group (i.e. people who could still leave after the moment of observation), and (4) it allows time varying independent variables to be included in the model.

A person-year file was created to accommodate the estimation of event-history models. The dependent variable was the likelihood of migrating after age 50, given that a person was still at risk (not migrated). Our observation period started at age 50 and ended in the year of migration. The dependent variable was coded as 0 for all the years before the year of migration, while the year of migration was coded as 1. The migrants were truncated after the year of migration. For those who did not migrate, the observation period started at the age of 50 and ended at the time of the survey. The dependent variable for the non-migrants was 0 for all person-year records. After these criteria were applied, there were 102,546 person-year records. Because of our oversample of migrants, the baseline hazard of migration is too high, but differentials in the hazard are valid. Similar designs have been used in event-history analyses of other uncommon events (e.g. Kalmijn and Poortman 2006).

Measures

Our independent variables capture four groups of determinants of retirement migration: socioeconomic status, social ties, personality traits, and cultural values and identity. Table 1 presents the mean, coding and psychometric properties, time frame, survey questions, and answer categories of all independent variables in the person-year file. Some questions were retrospective. The retrospective questions differed in how they were asked to migrants and non-migrants. For example, the question regarding occupational status was phrased as ‘What was your employment position just before you emigrated?’ to migrants and ‘What was your employment position just before you became 65 years old?’ to non-migrants. Additionally, variables were either constructed as time-constant or time-varying variables. The decision for which time frame was chosen was made by considering the nature of each variable. For example, while the last obtained education was constructed as time-constant, the age of their children was constructed as a time-varying variable.

Table 1. Sample size, means, proportions, coding and psychometric properties, time frame, questions, and answers of all variables in the person-year file.

Variables	Migrated		Stayed		Coding and psychometric properties	Time frame	Questions/answers
	N	Mean/%	N	Mean/%
Age categories	69,072		33,474		Categorical variable	Time varying	In which year were you born? (This variable was created by subtracting the year of birth from the year of observation)
50–55		35.8%		20.0%
55–60		29.6%		20.0%
60–65		21.1%		20.0%
65–70		9.7%		18.7%
70–75		2.8%		12.4%
75–80		0.7%		0.6%
80–85		0.1%		2.3%
85–90		0.0%		0.5%
Partnership status × Gender*	68,785		33,474		Categorical variable	Time constant	What is your gender? & Did you have a partner when you emigrated? / Did you have a partner when you turned 65 years old?
Partnered fe/male		79.0%		77.6%
Single female		7.3%		8.3%
Single male		13.7%		14.1%
Partner’s migration background	48,160	38.6%	22,161	11.1%	Dummy variable coded 0–1, 1 = Partner or partners’s parents born outside NL	Time constant	Where were your partner and their parents born? (Referring to the partner mentioned above)
Retirement status (Retired)	69,072	30.8%	33,474	41.9%	Dummy variable coded 0–1, 1 = Retired	Time varying	What was your employment position: just before you emigrated? / just before you became 65 years old? & At what age did you stop working in the Netherlands? / At what age did you stop working?
Chronic health conditions*	66,545				Categorical variable, questions taken from Vanajan, Bültmann, and Henkens (2021)	Time constant	Did you have the following conditions before you emigrated? / Did you have the following conditions when you were 65 years old?
None		48.6%		46.4%
One		28.3%		30.7%
Comorbidity		23.1%		22.9%
Education	68,574	60.8	32,361	55.8	Continuous variable, coding is based on the linear ISLED scale (Schröder and Ganzeboom 2014)	Time constant	What is the highest level of education you have completed with a diploma?
Occupation*	67,156		32,718		Categorical variable	Time constant	What was your (last) occupation in the Netherlands? / What was your (last) occupation?
Higher professional		16.6%		7.0%
Higher managerial		17.4%		7.1%
Secondary professional		19.1%		24.7%
Secondary managerial		17.8%		15.4%
Non-manual		12.7%		23.2%
Skilled manual		9.0%		11.8%
Unskilled manual		7.4%		10.7%
Children	67,799		33,454		Categorical variable	Time varying	Do you have (your own) children?^a & How old is this child?
None		26.3%		17.0%
Below age 18		6.9%		4.9%
Above age 18		66.8%		78.2%
Grandchildren	68,626		33,474		Categorical variable	Time varying	Do you have grandchildren? & How old is your oldest grandchild?
None		71.3%		50.3%
Below age 18		26.4%		39.2%
Above age 18		2.3%		10.5%
Parents	60,515		29,499		Categorical variable	Time varying	Is your mother still alive? / If no, year of death … /Is your father still alive? / If no, year of death …
Both alive		10.2%		5.2%
One parent alive		32.3%		21.6%
Both deceased		57.4%		73.2%
Civic engagement in the origin country (alpha = .60)	63,869	3.7	33,474	5	Five-item scale, coded 0 (None for all items) to 15 (Often for all items)	Time constant	When you were still living in the Netherlands, between the ages of 50 and 65, did you do the following … /Between the ages of 50 and 65, did you do the following? (a) practising a sport (b) participate in an association or club (c) being active for a volunteer organisation (d) attend religious service (e) being active for the neighbourhood
Extraversion (alpha = .83)	65,763	3.4	33,311	3.3	Seven-item scale, range 1 (Completely wrong) to 5 (Absolutely correct). From IPIP personality inventory (Goldberg et al. 2006)	Time constant	Can you describe how accurately each expression describes you as a person? Describe yourself as you generally are now and not as you want to be in the future. (a) don’t talk much (b) feel good in the company of people (c) stay in the background (d) start conversations (e) have little to say (f) talk to many different people at parties (g) am quiet in the company of strangers
Adventurousness (alpha = .80)	65,768	3.2	33,311	3	Seven-item scale, range 1 (Completely wrong) to 5 (Absolutely correct). From IPIP personality inventory (Goldberg et al. 2006)	Time constant	Can you describe how accurately each expression describes you as a person? Describe yourself as you generally are now and not as you want to be in the future. (a) prefer variety to routine (b) I like to keep everything the way it is (c) love to start something new (d) I am attached to fixed patterns and habits (e) I hate change (f) I am a creature of habit (g) I am interested in many things
Counterculture Identification (alpha = .81)	65,656	1.5	33,335	1.4	Four-item scale, range 1 (No) to 4 (Very)	Time constant	Did you identify with traits from the 1960s and 1970s when you were young? (a) hippie culture (b) protest generation/ anti-establishment (c) feminism (d) alternative lifestyles
Postmaterialism	60,414		33,335		Inglehart’s four-item value indicator was used (Inglehart and Abramson 1999), but the item ‘fighting inflation’ was changed to ‘stimulating economic growth’ Coding was done following Halman’s (2009) steps	Time constant	If you had to choose, which of the following do you think is the most important? Which one comes in the second place? (a) maintaining the rule of law (b) giving people a bigger say in important government decisions (c) stimulating economic growth (d) protecting free speech
Materialist		13.9%		18.8%
Mixed		57.8%		69.0%
Postmaterialist		28.3%		12.2%

Source: DRM 2021.

Note: The descriptive statistics are based on the person-year file prior to imputation.

*Questions differed between stayers and migrants: For questions that included the phrase ‘before you emigrated?’ for the migrants, the phrasing was changed for the stayers for an appropriate comparison.

^a The question asked about their child's age was specifically about the child who had their birthday closest to the date of response to ensure that there would be a quasi-random selection of a child.

We constructed five time-varying variables using the information on the years in which events happened. A categorical time-varying age variable was created with the current age (year of observation minus the year of birth) variable divided into five-year intervals. Similarly, a time-varying retirement variable was created by combining the retirement status and the retirement age variables. At last, we combined questions to create the time-varying parent, children, and grandchildren variables. For the variable concerning parents, we combined questions on whether their mother and father were alive and, if not, their year of death. This variable was then made into three categories ‘(a) both parents alive, (b) one parent alive, (c) both parents deceased’. For the child variable, we combined two questions. One question asked whether the respondent had a child, and the other inquired about the child’s age. The answers from these two questions were combined to make a time-varying variable with three categories ‘(a) no children, (b) child below age 18, (c) child above age 18’. The question asked about their child’s age was specifically about the child who had their birthday closest to the date of response to ensure the anonymity of the respondents. For the grandchild variable, we combined questions on whether they had grandchildren; if yes, the age of their oldest grandchild. The answers to these questions were combined to make three categories ‘(a) no grandchildren, (b) grandchild below age 18, (c) grandchild above age 18’.

Overall, the number of missing values was low. The variables with the most missing values were the variable that indicated the parents’ age (12% missing) and the variables in the postmaterialism scale (8% missing). The item nonresponse was lower than 4% for the rest of the variables. The missings were dealt with using multiple imputation procedures using mi impute in Stata 17. We imputed the variables with missing values 20 times and used information from dependent, independent, and an extra variable for the continent of residence of the respondents. Logit models with the mi estimate command were used to test our hypotheses.

We estimated five discrete-time event-history models to estimate the effects of socioeconomic status, social ties, personality traits, and cultural values on the likelihood of migration. Each model estimated age, retirement status, and health as baseline variables. Continuous independent variables were standardised for easier interpretation, and extraversion, which we hypothesised could have curvilinear effects, was squared and added to the models. We conducted a separate analysis to test the effect on the partner’s migration background. This analysis focused on people with a partner and included the baseline variables and the partner’s migration background variable in a logit model.

Results

To gain insights into the effects of age, retirement status, and health on the likelihood of migration, Table 2 presents the results of these baseline predictor variables. The results showed that migration was most likely to occur between the ages of 65–70, which is in line with previous research on internal retirement migration (Sander and Bell 2014). Being retired was associated with a higher likelihood of migration, confirming that what we observe was indeed retirement migration. Poor health, indicated as having one or multiple chronic health conditions (comorbidity), was negatively associated with the likelihood of migration compared to having no chronic health conditions. The negative association between health and migration is noteworthy as it shows a different profile than what some of the previous research shows, which is a type of retirement migrant who migrates to a warmer climate to manage health issues such as rheumatism (Rodriguez, Fernandez-Mayoralas, and Rojo 1998).

Table 2. Results of baseline event-history model of retirement migration. Regression coefficients and standard errors in parentheses (N = 102,546).

Baseline variables	Full model
Age categories (Ref: 50–55)
55–60	.512*
	(.051)
60–65	1.000*
	(.052)
65–70	1.626*
	(.057)
70–75	1.163*
	(.073)
75–80	.942*
	(.105)
80–85	.462*
	(.202)
85–90	−.018
	(.497)
Retirement status (Ref: Not retired)	.601*
	(.035)
Chronic health conditions (Ref: None)
One chronic disease	−.183*
	(.036)
Comorbidity	−.241*
	(.037)

Source: DRM 2021.

*p < .05.

The first column of Table 3 presents the results of Model 1, in which occupation and education were included as predictor variables. The second column presents the results of Model 2, in which partnership, parents, children, and grandchildren were added as predictor variables. Model 3 includes extraversion and adventurousness, while Model 4 in the fourth column includes postmaterialism and counterculture identification as predictors. The last column combines all variables in one model.

Table 3. Event-history models of retirement migration: Regression coefficients and standard errors in parentheses (N = 102,546).

Independent variables	Model 1	Model 2	Model 3	Model 4	Full model
Occupational status (Ref: Higher professional)
Higher managerial	−.003				.023
	(.050)				(.051)
Secondary professional	−.378*				−.343*
	(.051)				(.052)
Secondary managerial	−.163*				−.139*
	(.056)				(.057)
Nonmanual	−.567*				−.478*
	(.063)				(.064)
Skilled manual or service	−.399*				−.382*
	(.073)				(.073)
Unskilled manual	−.436*				−.376*
	(.081)				(.080)
Education (ISLED)	.045*				.009
	(.020)				(.020)
Partnership (Ref: Partnered)
Single female		−.329*			−.319*
		(.059)			(.061)
Single male		−.203*			−.198*
		(.043)			(.042)
Parents (Ref: Both parents alive)
One parent alive		−.016			.011
		(.070)			(.071)
Both parents deceased		−.032			.019
		(.070)			(.071)
Children (Ref: No child)
Children below 18		−.210*			−.268*
		(.076)			(.077)
Children above 18		−.213*			−.219*
		(.039)			(.039)
Grandchildren (Ref: No grandchild)
Grandchild below 18		−.240*			−.186*
		(.038)			(.038)
Grandchild above 18		−.251*			−.177*
		(.073)			(.073)
Civic engagement in the origin		−.164*			−.174*
		(.016)			(.017)
Extraversion			−.015		.000
			(.016)		(.017)
Extraversion squared			.009		−.000
			(.011)		(.011)
Adventurousness			.180*		.116*
			(.016)		(.017)
Materialism (Ref: Materialist)
Mixed				.026	−.006
				(.049)	(.048)
Postmaterialist				.351*	.283*
				(.052)	(.052)
Counterculture identification				.117*	.069*
				(.015)	(.016)
Constant	−3.519*	−3.525*	−3.780*	−3.869*	−3.448*
	(.054)	(.073)	(.043)	(.058)	(.094)

Source: DRM 2021.

Note: All models are controlled for baseline variables. The partner’s migration background is not included in the models and in the table, as it was analyzed in a separate model.

*p < .05.

The results of Model 1 did not support our hypothesis that higher education was associated with a higher likelihood of migration. However, our findings showed that higher professionals (e.g. doctors and teachers) were more likely to migrate than secondary professionals or managers (e.g. department managers), as well as those in non-manual and manual labour. We did not find support for our hypotheses on a curvilinear effect of occupation on the likelihood of migration. Instead, this positive and nearly linear association with occupational status is in line with human capital theory, suggesting a higher propensity to migrate among higher socioeconomic status positions.

Results of Model 2 supported our hypothesis that people with a partner were more likely to migrate compared to single women and men, suggesting that the spousal relationship is an important support system during the migration process. Additionally, people with a partner with a migration background were more likely to migrate than people with a partner with a Dutch background (b = .478, p < .001), indicating that the partner’s characteristics play an important role in the migration decision.

The results of the second model generally provided support for the hypothesis that ties in the country of origin are associated with a lower likelihood of migration. As hypothesised, people with children and grandchildren had a lower likelihood of migration than people without children and grandchildren. Similarly, civic engagement in the origin between ages 50–65, such as volunteering, lowered the likelihood of migration. The main refutation came from the presence of parents. Unlike hypothesised, having living parents compared to having deceased parents did not affect the likelihood of migration.

We also formulated hypotheses about the role of the age of children and grandchildren. Contrary to our argument that having children below 18 would not affect the likelihood of migration, results showed that the mere presence of children lowered the likelihood of migration, regardless of their age. We also expected that having grandchildren above 18 would not affect the likelihood of migration. However, Model 2 showed that having grandchildren, regardless of their age, lowered the likelihood of migration.

The results of Model 3 provided partial support for our hypotheses about the importance of personality traits. People higher in adventurousness were more likely to migrate than those lower in adventurousness, supporting our hypothesis on the positive effect of adventurousness on the likelihood of migration. There was neither the hypothesised curvilinear effect of extraversion on the likelihood of migration nor a linear effect. The lack of an effect of extraversion could mean that retirement migrants are less focused on socialising in the destination country than other migrants.

The results of Model 4 showed that having postmaterialist views increased the likelihood of migration compared to having materialist views, providing support for our hypothesis on postmaterialism. Results also showed that identifying with the counterculture of the 1960s and 1970s, which is argued to lead to more active views on retirement, was associated with a higher likelihood of migration, providing support for our hypothesis on the long-lasting effects of counterculture identification in youth.

Exploratory analyses – destination regions

We conducted a sensitivity analysis to investigate whether the determinants of retirement migration differed between destinations. Table 4 presents the results of the multinomial event-history analysis conducted to investigate which determinants played a role in the likelihood of migrating to Europe, Asia, or other countries. We combined the regions outside of Europe and Asia as the sample sizes in separate regions were too small.

Table 4. Multinomial event-history models of retirement migration: Regression coefficients and standard errors in parentheses (N = 102,546).

Independent variables	Europe		Asia		Other countries
Age categories (Ref: 50–55)
55–60	.555*	(.062)	.955*	(.174)	.232*	(.101)
60–65	1.059*	(.064)	1.645*	(.179)	.532*	(.106)
65–70	1.713*	(.070)	2.279*	(.194)	1.039*	(.121)
70–75	1.303*	(.086)	1.627*	(.237)	.482*	(.165)
75–80	1.124*	(.124)	1.416*	(.316)	.034	(.281)
80–90^a	.386	(.235)	1.124*	(.498)	.087	(.645)
Retirement status (Ref: Not retired)	.586*	(.041)	.619*	(.098)	.645*	(.074)
Chronic health conditions (Ref: None)
One chronic disease	−.125*	(.043)	−.142	(.100)	−.417*	(.080)
Comorbidity	−.174*	(.045)	−.066	(.106)	−.652*	(.093)
Occupational status (Ref: Higher professional)
Higher managerial	.055	(.061)	.179	(.163)	−.176	(.114)
Secondary professional	−.271*	(.061)	−.473*	(.162)	−.523*	(.113)
Secondary managerial	−.132	(.068)	.022	(.170)	−.264*	(.122)
Nonmanual	−.437*	(.077)	−.643*	(.206)	−.512*	(.137)
Skilled manual or service	−.400*	(.087)	−.092	(.204)	−.520*	(.162)
Unskilled manual	−.400*	(.096)	−.230	(.221)	−.406*	(.177)
Education (ISLED)	.023	(.023)	−.095	(.055)	.025	(.044)
Partnership (Ref: Partnered)
Single female	−.418*	(.070)	−.817*	(.242)	.164	(.115)
Single male	−.579*	(.057)	.958*	(.092)	−.070	(.096)
Parents (Ref: Both parents alive)
One parent alive	.064	(.089)	−.304	(.212)	−.003	(.144)
Both parents deceased	.073	(.087)	−.212	(.210)	−.031	(.143)
Children (Ref: No child)
Children below 18	−.401*	(.097)	−.054	(.207)	−.010	(.150)
Children above 18	−.171*	(.047)	−.488*	(.112)	−.227*	(.084)
Grandchildren (Ref: No grandchild)
Grandchild below 18	−.169*	(.045)	−.306*	(.118)	−.160	(.086)
Grandchild above 18	−.156	(.086)	−.156	(.196)	−.292	(.191)
Civic engagement in the origin	−.213*	(.019)	−.110*	(.047)	−.072*	(.036)
Extraversion	.032	(.020)	−.099*	(.048)	−.041	(.038)
Extraversion squared	−.014	(.014)	.033	(.030)	.002	(.025)
Adventurousness	.154*	(.020)	−.013	(.052)	.058	(.041)
Materialism (Ref: Materialist)
Mixed	.009	(.057)	.255	(.142)	−.198	(.104)
Postmaterialist	.271*	(.061)	.646*	(.153)	.109	(.116)
Counterculture identification	.092*	(.019)	.046	(.046)	−.008	(.036)
Constant	−3.96*	(.115)	−6.35*	(.296)	−4.37*	(.190)
N events	3636		600		970

Source: DRM 2021.

Note: Countries included in Europe and Asia are selected geographically. Israel and Turkey are included in Asia.

^a This age group consists of a 10-year interval as separation resulted in a small sample size.

*p < .05.

As seen in Table 4, there were more similarities than differences in the determinants of retirement migration to different regions, showing that the overall results were mainly consistent across different regions. Perhaps the most striking difference was in partnership and gender, which showed that single males were more likely to migrate to Asia (mainly Thailand and the Philippines in our study) than people with a partner. The phenomenon of single males migrating to these Asian countries has been gaining the interest of researchers, especially due to the status and age differences between the migrant and the host society (Bell 2015; 2017; Statham 2019).

Discussion

International retirement migrants are a special group of people who challenge our thinking about aging and migration. Retirement migrants face the challenges associated with a new language, bureaucracy, and healthcare system at a later stage of their lives. Although gerontological literature shows that older adults prefer to age in place (Boldy et al. 2011; Stones and Gullifer 2016), research on retirement migration shows that a small group of older adults seek amenities that they cannot readily access in their origin country.

In this paper, we investigated for whom aging in a new country might be more appealing. We found that people with a certain demographic, socioeconomic, psychological, and cultural profile are more likely to partake in international retirement migration. The sharp decline in the likelihood of migration after age 75 indicated a critical period around retirement in which people migrate. Other characteristics in the profile provided the means to migrate, such as having higher occupational status, which is in line with the majority of quantitative research on retirement migrants (Bahar et al. 2009; King, Warnes, and Williams 1998; Lizarraga, Mantecón, and Huete 2015; Rodriguez, Fernandez-Mayoralas, and Rojo 1998). The profile also included personal characteristics that would make the idea of moving abroad after retirement more attractive, such as being more adventurous, seeking self-actualisation, and being affiliated with the counterculture.

People generally build strong family, community, and professional ties throughout their lives. These ties can be jeopardised by international borders between the individual and their network. As people retire, professional ties weaken, while family and community ties become more important. It is often assumed that retirement migrants leave their network behind. We showed that people with fewer family and community ties are more likely to migrate after retirement than people with more social ties in the origin. This might suggest that retirement migrants may encounter fewer adverse social network-related effects of migration, such as the potential loss of contact with family and friends. Since retirement migrants are less likely to have children, grandchildren, and community ties, they may be able to migrate more ‘freely’ once their professional ties are severed.

Our study has a number of innovative elements. This is the first study that provides a comprehensive view of retirement migration from a single country. By using event-history models, we showed how migrants differed from non-migrants not only in their socio-demographic profiles but also in their social integration in the country of origin and their psychological profiles. Overall, this study gave a comprehensive look at whom retirement migrants are compared to non-migrants by employing a novel methodology asking comparable retrospective questions to representative samples of migrants and non-migrants.

There are three main limitations to this research. First, some subjective retrospective questions might be prone to response bias. For example, we asked about people’s current postmaterialist values, but we used them to reflect on their past actions (migration). However, we assume these values are relatively constant in late life (Grünwald, Damman, and Henkens 2022) and thus are not dependent on the migration experience. Second, our sample consisted of people receiving their pensions abroad and therefore settled in their country of destination. We did not capture seasonal migrants (e.g. snowbirds) or people who own second houses abroad but are still officially living in the origin country. We also did not have information on property ownership in the destination before permanent migration, which could be considered in future studies. Third, while we collected data from forty destination countries, we investigated one country of origin. The Netherlands has an extensive welfare state and a generous pension system. Thus, further research should investigate to what extent our findings can be generalised to other countries with different welfare arrangements.

This study is the first to compare who migrates to who stays after retirement by using a representative retrospective survey on international retirement migrants from the Netherlands. We found several determinants that foster international retirement migration. Understanding these determinants of retirement migration may provide valuable insights into the motivations, experiences, and impact of this type of migration as well as their structural incorporation into the country of destination. It is clear that international retirement migrants break the stereotype of older people being set in their ways and resistant to change, suggesting that they are more flexible and adaptable than previously thought.

Author Contributions

E. B. Savaş wrote the main part of the article and performed the statistical analyses. K. Henkens and M. Kalmijn substantially contributed to the manuscript. The authors jointly developed the idea and the design of the study.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Dutch National Science Foundation [grant number 406.18.SW.022].

Appendix

Population of study by country and sample numbers: persons 66–90 with Dutch nationality abroad with 70% pension accumulation.

Table

	Country	Population	Sample	Fraction
0	Belgium	10,617	0	0.00
0	Germany	7834	0	0.00
1	France	5910	2364	0.40
2	Spain	5605	2242	0.40
3	Portugal	1265	759	0.60
4	Thailand	1028	617	0.60
5	USA	987	592	0.60
6	Hungary	667	400	0.60
7	Great-Britain	530	318	0.60
8	Switzerland	528	317	0.60
9	Italy	512	307	0.60
10	Austria	472	472	1.00
11	Sweden	463	463	1.00
12	Curaçao	439	439	1.00
13	Canada	360	360	1.00
14	Philippines	343	343	1.00
15	Indonesia	326	326	1.00
16	Greece	295	295	1.00
17	South Africa	291	291	1.00
18	Turkey	285	285	1.00
19	Australia	273	273	1.00
20	Poland	247	247	1.00
21	Bonaire	210	210	1.00
22	Ireland	204	204	1.00
23	Suriname	197	197	1.00
24	Brazil	184	184	1.00
25	Canary Islands	175	175	1.00
26	Aruba	160	160	1.00
27	New Zealand	158	158	1.00
28	Israel	157	157	1.00
29	Norway	134	134	1.00
30	Luxembourg	126	126	1.00
31	Czech Republic	124	124	1.00
32	Romania	105	105	1.00
33	Denmark	95	95	1.00
34	Malta	93	93	1.00
35	Bulgaria	87	87	1.00
36	Malaysia	61	61	1.00
37	Colombia	55	55	1.00
38	Cyprus	46	46	1.00
39	Croatia	44	44	1.00
40	Dominican Republic	43	43	1.00
41+	Other countries	764	0	0.00
	Total top 40*	23,284	14,168
	Total all countries*	24,048
	Coverage	0.97

*Excluding Belgium and Germany.

Source: DRM 2021.