Journal of Population Research

, Volume 31, Issue 1, pp 1–27 | Cite as

Migration and retirement in the life course: an event history approach

Article

Abstract

Migration at older ages is commonly explained by reference to the search for greater amenity, and subsequently by the onset of greater dependency, but the links between mobility and specific life course transitions have rarely been articulated. We aim to establish the timing of migration in relation to retirement from the labour force, and to determine how its intensity varies around the retirement event. We also seek to identify how household and individual characteristics shape the propensity and timing of migration, differentiating moves according to distance and with particular attention to the characteristics of the spouse. Data are drawn from the first six waves of the Household, Income and Labour Dynamics in Australia survey, a nationally representative panel study covering the period 2001–2006. Migration events are identified relative to retirement and event history methods are employed to establish the characteristics predisposing households to relocate around retirement. Results demonstrate that retirement acts as a trigger to migration but the propensity to move falls as retirement age rises and the hazard is increasingly concentrated in the year retirement occurs. Within this framework the presence, health, education and retirement status of a spouse exert a significant influence on the likelihood of migration, though with different effects for long and short distance moves. Results highlight the importance of variations in underlying life-course trajectories in shaping retirement migration and demonstrate that only a minority of moves at ages 55–69 are directly associated with retirement, underlining the need for caution when identifying retirement migration using age as a proxy measure.

Keywords

Internal migration Retirement Life course Event history analysis Australia 

Introduction

As in most developed countries, increasing longevity and declining fertility have caused an unprecedented ageing of the Australian population (Lutz et al. 2008). The associated increase in the number of retirees, which is set to rise further as the large baby-boom cohorts reach retirement, has been paralleled by changes in the life-course patterns of Australians. This means that the period spent in retirement is expanding (Himes 2001; Warnes et al. 2004). At the same time, changes in the socio-economic status of retirees characterized by rising affluence, improved health and more flexible pension schemes, coupled with loosening of intergenerational relationships, have markedly increased the mobility of older people (Law and Warnes 1982; Warnes and Law 1984). The nature of the transition into retirement has also undergone fundamental change over recent decades. Part-time employment, telecommuting and dual-career households have all contributed to a longer and less clear-cut transition to retirement, which is no longer constrained to the ‘normal’ retirement age of 65 years. These changes raise the crucial question of how the transition into retirement influences the propensity for retirement migration and its timing.

This paper explores the links between the two discrete events of retirement and migration. In the literature, retirement migration is often referred to as those moves that occur at or in retirement, implicitly assuming that withdrawal from the labour force occurs instantaneously at age 65 (Law and Warnes 1975; Wroe 1973). Mainly because of data and methodological constraints, this assumption has not yet been tested and the behavioural models (Law and Warnes 1982; Litwak and Longino 1987; Wiseman 1980) that form the theoretical basis of most cross-sectional studies are yet to be clearly calibrated. Contemporary thinking on retirement migration suggests that withdrawal from the labour force acts as a trigger for mobility (see Bradley et al. 2008; Bures 1997; Frey 1986; Longino et al. 2008). However, variations in the sequencing and timing of life-course events mean that retirement cannot be directly tied to a specific age. Adding further complexity to this issue, migration around the time of retirement may in turn trigger other life events or may be undertaken in anticipation of another event occurring, such as the onset of disability. Further, family or housing careers can constrain or facilitate the triggering effect of retirement (Mulder and Hooimeijer 1999).

In recognition of these complexities, the focus of research has shifted from a cross-sectional to a life-course approach. The conceptual framework of the life course that underpins several recent studies is helpful in understanding the patterns and processes pertaining to mobility of retirees (Bradley et al. 2008; Courgeau and Lelièvre 1992; Longino et al. 2008). The analysis presented here aims to tease out the links between migration and retirement by determining the timing of migration relative to retirement. Setting this in a life-course framework enables a number of other life-course influences to be taken into account, including the complexity of household structure, spouses’ employment career, and type of move.

This study departs from previous work by defining retirement migration by reference to moves that take place close to retirement, irrespective of the age at which retirement itself has occurred. Thus our attention is focused on the influence of the retirement event, rather than on older-age retirement in general. The study uses data from the Housing, Income and Labour Dynamics in Australia survey (HILDA), a nationally representative panel survey which began in 2001 with a sample of 13,969 individuals in 7,682 households. Discrete-time event history models are estimated to predict the probability of moving between 2 years before and 4 years after the retirement event, and to determine how household composition, demographic characteristics, financial resources, educational attainment and health status influence the hazard of moving. The objective of focusing on the links between these two life-course events is to determine if the triggering effect of the retirement event on migration is as strongly related to age as commonly conceived. Since we expect the strength and nature of this connection to differ by distance moved, a distinction is made in the analysis between short and longer-distance moves.

The paper begins with a brief description of the HILDA dataset, followed by a description of how the transition into retirement was defined, how spouses were linked to analyse migration in a household context, and how the discrete-time person-period file was created. Next, it outlines how the duration from retirement to migration was measured, which personal and household characteristics were included in the model, and how different types of move were distinguished. The introduction of the discrete-time event history model is followed by a discussion of the age profile of retirement among respondents in the HILDA survey, the timing of migration relative to retirement, and the link between these events using hazard profiles. We then report the results from the event history models to establish the determinants of retirement migration at differing spatial scales.

Data

The HILDA sample

The data are drawn from waves 1–6 of the HILDA survey, 2001–2006, a full description of which can be found in Watson and Wooden (2002). To summarize its key features, HILDA is a nationally representative sample of Australian households and contains information at both the individual and household level. Data on economic and subjective well-being, labour market dynamics, and family dynamics were collected in annual waves, six of which were conducted between 2001 and 2006. A total of 13,969 individuals in 7,682 households were interviewed in the first wave in 2001. Of the initial sample, 3,884 persons were lost to attrition, so that 10,085 of the initial sample (72 %) were re-interviewed in the sixth wave. As we will show, sample attrition had no significant effects on the results of our event history models. Personal interviews with all household members aged 15 years and over were supplemented with a household questionnaire and a self-completion questionnaire. The HILDA survey has been used extensively in Australia to study a wide range of family, health, housing and labour market dynamics, circumstances and transitions, but little attention has previously been given to its potential for the analysis of internal migration.1 This neglect is surprising given the high level of mobility in Australia and the dearth of longitudinal datasets that track residential histories (Bell and Hugo 2000).

Defining the retirement transition

The transition to retirement was measured using each individual’s self-reported retirement status. This transition is not always clear-cut or immediate as people may go through several intervening years of part-time work or unemployment before full retirement, or they may return to the workforce following a period of retirement. To account for the range of possible pathways, retirement was defined as a change in employment status from (a) ‘full-time employment’, (b) ‘part-time employment’ or (c) ‘unemployed but looking for work’ to ‘retired’. Many women do not participate in the workforce or may show a pattern of sporadic employment over the life course. Others define themselves as housewives until they are unable to care for themselves and their spouses owing to declining health. Women’s self-reported retirement status can therefore be vague, complicating analysis. For the analysis presented here, women with interrupted or missing employment history were classified as housewives, and it was assumed that they did not undergo a formal transition to retirement. Since female partners play a key role in the decision to migrate, however, it was important to retain them in the study. Hence, a household-based approach was adopted, whereby, in the case of a two-person, single-earner household, the focus was placed on the retirement of the male income-earner.

Linking spouses in a household-based approach

The timing of retirement needs to be seen in a household context (Boyle et al. 2008; van Solinge and Henkens 2005). Around retirement age, single or couple households are most common, as the majority of resident children have already left the parental home and few people live in multigenerational households. In couple households, spouses influence the timing of each other’s retirement (Smith and Moen 1998); retirement is therefore not purely an individual decision but affects both the retiree and the spouse (van Solinge and Henkens 2005). The HILDA data allow household members to be linked, and event occurrence and timing to be analysed at a household level. In the analysis presented here, each household is represented by the retiring household member. The covariates for each retiree include their own characteristics, the characteristics of their spouses and details of household composition. In dual-earner households, a female and a male retiree represent one household. Since in dual-earner households most female spouses retired in the same year as the husbands, the latter were chosen to represent the household. Therefore, the sample contains only wives; households that experienced union formation or dissolution during the survey period were excluded. The literature shows that at least one member of these households tends to migrate around the time of formation or dissolution, thus potentially confusing the expected link between migration and retirement (Boyle et al. 2008; Flowerdew and Al-Hamad 2004; Grundy 1985). Same-sex couples and singles living in shared accommodation were also excluded.

The discrete-time person-period file

To estimate the discrete-time event history model, a person-period file had to be created, to which the model was then applied (Allison 1984). In the HILDA survey, each respondent is asked once a year whether he or she changed address, which means that the event of migration is measured in discrete intervals. The dataset holds a sequence of observations for each individual. This means one observation for each time interval until event occurrence (moved) or censoring (did not move until end of observation period), unless the person was lost through attrition. Since less than ten per cent of all respondents moved more than once during the observation period, an individual was deemed to drop out of the risk set once he or she moved. To construct an unbalanced panel (since there are missing values), the sequences of observations for all respondents were merged across waves, which resulted in a person-period file containing 77,810 records. This file contains all respondents who contributed at least two waves to the HILDA data set, whether they retired during the survey period, were still in the labour force or had already retired before the survey. The analytic sample of retirees at risk of migrating was then obtained by applying selection criteria to the person-period file. The criteria were that the person was aged between 50 and 75 years in at least one survey wave; had retired during the survey period; and had no gaps in survey participation (missing waves). After these criteria were applied, the sample contained 1,223 person-period records representing 315 retiree households.

Measuring the duration from retirement to migration

The variable ‘Retirement Time’ was created to measure the duration from retirement to migration (see Fig. 1). In the HILDA survey, individuals retired in Waves 2–6, since persons already retired at the beginning of the survey in 2001 were excluded (Fig. 1a). The data were then pooled across waves and five cohorts were defined based on the wave in which individuals retired (Fig. 1b). Not all five cohorts were observed across the six waves. Observations in Wave 1 were not included for cohorts 1 and 2, since no transitions to retirement or migration events were observed in Wave 1 of the HILDA survey, although observations in the first wave are within 2 years before retirement for these cohorts. Observations in Waves 1–3 were excluded for cohorts 3–5 if observations were more than 2 years before retirement.
Fig. 1

Creating ‘Retirement Time’ periods

Retirement Time was then defined by comparing migration to a reference date whereby Retirement Time period three is defined as the year in which a respondent retired (see Fig. 1c). Periods one and two refer to the 2 years before retirement and periods four to six refer to the years after retirement. This means that retirees can be observed for a maximum of 4 years after retirement and allows moves undertaken on the eve of retirement but while still in the labour force to be accounted for. Because of data limitations, not all respondents were observed over all six Retirement Time periods. These gaps occur because data were available but the observations refer to survey data collected more than 2 years before a respondent’s retirement (marked by ‘a’ in Fig. 1b); data were available but no transitions to retirement or migration events were collected in Wave 1 (marked by ‘b’ in Fig. 1b–c); and no data were available for the Retirement Time periods (marked by ‘c’ in Fig. 1c).

Adding personal and household characteristics

The probability of migration is influenced by personal and household characteristics; previous studies have shown that a number of factors are likely to affect migratory behaviour. The age selectivity of migration is well established in the literature (Rogers and Castro 1981; Rogers et al. 1978; Thomas 1938) and other research indicates that high income, an advanced level of educational attainment, good health and substantial housing assets increase the risk of migration around retirement age (Biggar 1980; Bradley 2011; Clark and Davies 1990; Sommers and Rowell 1992), although there is contravening evidence that people with limited financial resources (Hugo and Bell 1998) and those in need of assistance and care are also more likely to move (Litwak and Longino 1987). In family households, the personal characteristics of the female spouse may also be influential because decision-making is collaborative (Boyle et al. 2001; Kitching 1990), and the income or retirement pension of a (formerly) full-time employed spouse provides additional resources that may be required to realize a lifestyle-related move. According to previous work, lifestyle-related migration is primarily undertaken by couples, whereas lone-person households tend to move over shorter distances for health-related reasons (Biggar 1980; Haas and Serow 1993; Sommers and Rowell 1992).

Not all individual and household characteristics that are important in explaining the migration decision among retirees could be readily captured in the HILDA dataset. In the analysis presented here, we include both time-varying and time-constant explanatory variables representing personal and contextual factors that have been identified in the literature as determinants of retirement migration. The covariates used are listed in Table 1. Preliminary analysis showed that net wealth (an indicator of household wealth that incorporates income, assets and pension payments) and house value had no significant effects on the hazard, so these variables were excluded.
Table 1

Definition of covariates and sample characteristics

Variable name

Variable description

% of households in sample

All households

Contributed 6 waves (n = 217)

‘attrited’ from panel (n = 98)

Retirement time

Time since 2 years before retirement in years

   

Household composition

 Female spouse nonea

Dummy = 1 if single household

25.7

25.0

27.6

 Female spouse also retireda

Dummy = 1 if both spouses of dual-earner household retire

15.2

13.9

18.4

 Female spouse in labour forcea

Dummy = 1 if only the male spouse of a dual-earner household retires

18.1

18.5

17.3

 Female spouse not in labour forcea

Dummy = 1 if husband in a single-earner household retires

41.0

42.6

36.7

Spouse characteristics

 Retiree bad healthb

Dummy = 1 if the respondent stated fair or poor health

27.0

23.6

34.7

 Female spouse bad healthb

Dummy = 1 if the female spouse stated fair or poor health

19.0

17.1

23.5

 Retiree education tertiarya

Dummy = 1 if the respondent has postgraduate degree, bachelor degree or a graduate diploma

19.7

19.9

18.4

 Female spouse education tertiarya

Dummy = 1 if the female spouse has postgraduate degree, bachelor degree or a graduate diploma

11.7

11.6

12.2

 Retiree education post-secondarya

Dummy = 1 if the respondent has non-university certificate or diploma

39.7

40.3

38.8

 Female spouse education post-secondarya

Dummy = 1 if the female spouse has a non-university certificate or diploma

20.3

21.8

16.3

 Retiree education high schoola

Dummy = 1 if the respondent finished high school

40.6

39.8

42.9

 Female spouse education high schoola

Dummy = 1 if the female spouse finished high school

67.9

66.7

71.4

Household characteristics

 Home ownerb,c

Dummy = 1 if the household lives in self-owned property or rent-free

88.9

88.9

88.8

 Low household incomeb

Dummy = 1 if yearly household income is below A$57,000 (mean)

55.6

52.8

62.2

Retirement and occupation characteristics

 Retirement age below 60 yearsa

Dummy = 1 if the respondent’s retirement age is below 60 years

36.2

34.7

39.8

 Low-impact occupationa,d

Dummy = 1 if respondent was a manager, professional or service worker in the year before retirement

52.1

55.6

43.9

aTime-constant

bTime-varying

cReference: renting

dReference: farmer, tradesperson, elementary clerk or labourer

Table 1 also shows the characteristics of households in the sample in the year of retirement (n = 315). To account for possible attrition bias, we provide summary statistics for households which contributed all six waves (n = 217), and for households which left the panel (n = 98). ‘Non-attritors’ and ‘attritors’ varied on many characteristics, but these differences had no significant effect on retirement migration behaviour.

In the sample as a whole, three-quarters of all retiree households were couple households; the largest single group consisted of those in which the spouses of retirees had either retired before the survey period or classified themselves as housewives. Only 15 % of all households were dual-earner couples in which both spouses retired during the survey period, although the share was slightly higher (18.4 %) among households who left the panel. Differences between households who contributed all six waves to the panel and those who left the survey were most pronounced for health, with the former being less likely to suffer from poor health than the latter. Overall, the female spouses in the sample had slightly better health (approximately 80 % stated good health) than the male retirees (approximately 73 % stated good health), which may relate to female spouses’ younger ages. Retirees had a higher level of educational attainment (60 % had a post-secondary or tertiary degree) than their female spouses (only 32 %). Almost 90 % of households in the sample lived in owner-occupied property or did not pay rent, and just over half lived in a household with a yearly income below A$57,000. About 35 % of retirees withdrew from the labour force aged under 60 (about 40 % among retirees in ‘attriting’ households).

Differentiating between types of moves

It is well established that the determinants of local moves differ from those shaping moves over longer distances (e.g., Biggar 1980; Wiseman and Roseman 1979). In the HILDA survey, internal migration was measured using a question in the personal questionnaire, asked in Waves 2–6, as to whether the respondent had changed address since the last interview. Thus, the survey effectively measured moves as transitions over a one-year period, and captured all such residential relocations, irrespective of distance (Bell et al. 2002). Respondents did not have to cross an administrative boundary to be classified as internal migrants. However, the In-Confidence Release of the HILDA survey codes the location of each individual at each wave to one of the 66 statistical divisions (SDs) into which Australia is divided. For the analysis presented here, it was therefore convenient to distinguish two types of movements: short-distance moves which involved a relocation within the same SD, and longer-distance moves involving relocation to a different SD. Because of the physical size and settlement geography of Australia, some non-metropolitan SDs are large in area so that moves within these SDs could be over relatively long distances, but these represent a minority of all moves and can be considered ‘local’ in that they remain within the same functional region, since this is how SDs are defined (Bell and Hugo 2000).

During the sample period, 42 households (13 %) undertook a short-distance move and 38 households (12 %) moved over a longer distance. Only the first move of a household was considered to ensure the sample was not biased by chronic movers. Moves undertaken in the year before the first wave had to be excluded since the place of origin was not captured in HILDA and, therefore, the type of move could not be determined; this resulted in 19 moves being excluded from the analytic sample. The propensity to move within the same SD among retirees in the HILDA survey was similar to that observed using census data for 2001–2006: 14 % among those aged 55–69 (ABS 2006 Census, unpublished data). The propensity to move to a different SD was slightly higher in the HILDA sample, 12 % compared to 10.5 % in the census for those aged 55–69. To identify the determinants of retirement migration by type of move, separate event history models were run for short-distance and for long-distance moves.

Method

Discrete-time event history models were used to analyse the timing of migration around the time of retirement. As noted above, migration was defined as a change of usual residence over a 1-year time interval. Thus, although migration occurs in continuous time, the data derived from the HILDA survey are interval censored. Households which remained in the same residence throughout all waves of the survey are right-censored and contribute to the hazard calculation in the event history models. Standard econometric tools such as logistic regression cannot handle censored data adequately. In event history analysis, however, both migrants and non-migrants, that is both censored and non-censored individuals, contribute to the hazard calculation (Allison 1984). In discrete-time, the hazard (Hj) is the conditional probability P that a person migrates in a particular year (time interval tj), given that no migration occurred before tj. The discrete-time hazard is defined as:
$$ H_{j} = P(T = t|T \ge t_{j} ) $$
where T is a discrete random variable that indicates the time of the event (Jenkins 1995:131).
A widely used model in discrete-time event history analysis is the complementary log–log model, which is the discrete-time counterpart of an underlying continuous-time proportional hazard model (Prentice and Gloeckler 1978). The complementary log–log link function is suitable if a proportional hazards model holds in continuous time and the survival times are interval censored. In proportional hazard models, the hazard rate \( \theta (t,X) \) satisfies an important separability assumption:
$$ \theta (t,X) = \theta_{0} (t)\exp (\beta 'X) $$
Thus, it is the product of a non-parametric baseline hazard \( \theta_{0} (t) \), which may differ in each interval, and \( \exp (\beta 'X) \) where \( \beta ' \) is a vector of parameters to be estimated and \( X \) is a vector of covariates that captures the observed differences between individuals. The hazard function \( h(t_{j} ,X_{ij} ) \) shows the yearly hazard of migration for the time interval tj (i.e. the time between two annual HILDA interviews for person i). The discrete-time hazard in the jth interval thus has the following form:
$$ h(t_{j} ,x_{ij} ) = 1 - \exp \left\{ { - \exp \left( {\beta 'X_{ij} + \gamma_{j} } \right)} \right\} $$
where \( \gamma_{j} \) refers to the baseline hazard. A set of dummy variables was used to represent the effect of time since 2 years before retirement using the Retirement Time variable. The dependent variable was a binary indicator, dit = 1 if a person i moved in year t, and dit = 0 otherwise. The discrete-time hazard hij (the probability that the ith retiree would migrate in year j, given no previous migration), was estimated in STATA Version 10 using maximum likelihood estimates of the \( \beta ' \) parameters. Period-specific dummies for each of the six Retirement Time periods were included to estimate the non-parametric baseline hazard (Jenkins 1995; Singer and Willett 1993).

Migration and retirement

The age profile of retirement

While it has been shown that the retirement peak in the age profile of migration occurs at ages 62–65 in most Western countries (Rogers 1988; Rogers et al. 1978), age at retirement has changed markedly over recent decades, both in Australia and elsewhere (OECD 2011). Our findings show that that the transition to retirement is spread across a wide age range, extending over 30 years from 40 to 70 (see Fig. 2). Wave three of the HILDA survey includes a Retirement Module in which all respondents who had already retired before the third wave were asked in which year they retired. Age at retirement could be identified for 1,542 persons.
Fig. 2

Age profile of retirement among retired respondents in the HILDA survey

The results indicate that most Australians retire well before the traditional retirement age of 65. The pattern of early retirement was slightly more pronounced for women than for men: 88 % of all retiring men and 92 % of all retiring women did so before age 65. The most common age for HILDA respondents to retire was 59 or 60, with a secondary peak at ages 64 and 65. However, these four ages accounted for only 32 % of all retirements. The profile for men shows peaks at 59–60 and 64–65 years, whereas the profile for women has a single peak at 59–60 years. Despite this, the spread in the age profile of retirement is even more pronounced for women than for men: 50 % of retirements for men occurred within a 7-year range around the median retirement age of 60, whereas retirements for women were spread across a 9-year age range around a median of 57.

The spread of retirement suggests an equally broad spread in the age profile of retirement migration, and indicates that chronological age is not a good indicator of retirement-related moves. Moreover, it is potentially misleading to conceptualize all residential relocations around age 65 as retirement moves. According to the HILDA data, the majority of moves are undertaken by members of the labour force and long-term retirees, while the proportion of moves undertaken by the recently retired is very small. Figure 3 shows the employment status of all respondents in the person-period file who moved during the survey period by sex and age in the year of migration. If males and females are considered jointly, only 25 % of all moves undertaken at ages 55–69 within the survey period were ‘true’ retirement moves in the sense that the migrants had retired from the labour force within 3 years of moving. In contrast 33 % were in the labour force and 26 % had retired at least 4 years before moving. At ages 55–59, more than half of all migrants were in the labour force at the time of the move, while at ages 65–69 and 70–74 more than 50 % of all movers had retired at least 4 years before moving. As expected, the share of migrants who were still in the labour force at the time of the move was substantially higher for males than for females.
Fig. 3

Migration by employment status at the time of move by age for males (n = 596) and females (n = 590), 2001–2006

These results present a dilemma for analysts studying retirement migration using sources of data such as the census, where retirees can only be identified using age as a proxy measure, since no direct measure of retirement or reasons for migration are commonly given. When census data are used to explore the intensity or spatial patterns of retirement migration and the age-based definition of retirement has to be used, it is important to recognize that the results at best represent moves undertaken around the age of retirement, rather than moves that are necessarily triggered by the event of retirement. From the evidence presented here, only about a third of all moves in the traditional retirement age bracket are closely associated in time with the retirement event.

The timing of migration relative to retirement

The spread of retirement ages observed in the HILDA data as a whole is also apparent for households in the analytic sample. Figure 4 shows the distribution of retirement ages for all retirees in the 315 sample households who retired during the survey period. The majority left the labour force well before age 65. As for all retirees in the HILDA survey, retirement is spread widely across the age profile, from 50 to 75. Again, spikes are apparent around ages 60 and 65, with a third at age 55 indicating a cluster of early retirees.
Fig. 4

Retirement age profile for households in the sample, n = 315 (proportions of all sample households)

The HILDA data provide clear evidence of event dependence between retirement and migration, but the timing is not completely coincident. Figure 5 shows the relative frequencies of migration around the time of retirement for the 111 retiree households in the analytic sample who moved during the survey period, including repeat moves and moves in the year before the first survey wave. Approximately 40 % of moves were undertaken in the year of retirement, and a further 30 % occurred within 4 years following retirement. A small secondary peak in the profile is found 2 years before retirement, indicating that moves were also undertaken on the eve of retirement but while still in the labour force. It is important to note that the 6-year survey period (2001–2006) limits the observation window to 2 years before and 4 years after retirement, so it may be that a broader window would reveal a wider spread of retirement-related moves. However, the distribution in Fig. 5 clearly shows a distinct tailing off in the propensity to migrate with increasing time away from the year of retirement.
Fig. 5

Migration around the time of retirement (n = 111)

The data also demonstrate that the propensity to move in the year of withdrawal from the labour force varies systematically by age at retirement. Figure 6 charts the probability of moving by retirement age among the 315 households in the analytical sample and shows that the probability of moving around retirement declines steadily as age at retirement increases. Thus, the age-dependent decline in migration commonly observed in national migration age profiles is also found if attention is confined to retirement-related moves.
Fig. 6

Percentage of households in the sample which moved during the survey period, by age at retirement (111 moves)

The hazard of migration around retirement

To better understand the timing of migration relative to retirement, the baseline hazard of all moves among the 315 retiree households in the analytic sample was calculated, providing a sensitive lens to detect when migration is most likely to occur relative to the timing of retirement. Following Fig. 1, the hazard profiles were calculated for Retirement Time periods, where period three is the year in which a respondent retired, periods one and two refer to the 2 years before retirement, and periods four to six refer to the years after retirement.

Figure 7 shows the hazards for the sample households plotted as a step function. The hazard of migration is highest in the year of retirement (i.e. Retirement Time period three), increasing sharply from a low level in the year before retirement. The high hazard value in period six (3 years after retirement) is due to small numbers as the count of households decreases sharply in periods five and six.
Fig. 7

Discrete-time hazard function for all households in the analytic sample

In Fig. 8 the baseline hazard function is calculated separately by type of move. For both short and long distance moves, the hazard is highest in the year of retirement, which is consistent with the observation of Longino et al. (2008), who reported that recent retirement was a strong predictor of non-local moves in the US. Differences in the hazard by distance moved are most pronounced in the year after retirement where the hazard of moving over short distances remains high whereas the risk of moving over longer distances registers a decline.
Fig. 8

Discrete-time hazard function by distance moved

When the sample of retirees is disaggregated by age at retirement, considerable differences are apparent between the sample hazards for respondents who retired aged 59 or younger and respondents who retired at later ages. Figure 9 confirms the conclusion from Fig. 6 that the hazard of moving for early retirees is higher than for persons who retire at an older age. This difference is maintained across the hazard function, indicating that in every year around retirement, early retirees are more likely to move. However, Fig. 9 also shows that, as retirement age increases, the propensity to move becomes increasingly concentrated in the year of retirement. The peak in the hazard in the year of retirement is more pronounced for those aged 60 and over than for their younger counterparts. Thus, the hazard for young retirees remains relatively high after retirement, while for older retirees it falls sharply. Perhaps younger retirees have more place ties based upon the labour force status of the spouse, property ownership and relatives, so that moves are delayed until such ties are loosened, for example as the spouse retires and children leave the parental home. Conversely, it may be that older retirees have had more time to plan a move, and have greater incentive to implement the change in order to maximize the benefits of a move over their shorter remaining life expectancy.
Fig. 9

Discrete-time hazard functions for early and late retirees

Differences in household composition also play a role in migration behaviour around retirement (Fig. 10). Differences in the propensity to move are most apparent in the year of retirement and in the 2 years after retirement. In the year of retirement the risk of migration is highest for dual-earner couples. This suggests that members of this household type adjust their retirement timing to each other and may be more likely than other household types to have sufficient financial resources to realize a move. These findings are broadly consistent with the results reported by Speare and Meyer (1988) who found that the older household type with the highest mobility was the married couple, while single-person households showed intermediate mobility levels. In the HILDA data, likewise, single-person households showed significantly lower mobility in the year of retirement, but with hazards that are still well above those of couples in which the spouse was still working, or was not in the labour force (Fig. 10). The retirement status of the spouse clearly influences the likelihood that a couple will migrate, and its timing. A married couple was less likely to move in the year of the income-earner’s retirement if the spouse was still in the labour force than if the spouse was also retired. The decision to remain in the labour force might be motivated by financial necessity with the job ties of the partner limiting the scope the couple has to move. Two-person households in which there was a single earner displayed the lowest risk of moving in the year of retirement, probably due to limited resources and strong person and place ties in the home community, particularly among housewives.
Fig. 10

Discrete-time hazard functions by household composition

Determinants of retirement migration

Event history models were fitted to determine how household composition, combined with other individual and household characteristics, influences local and long-distance migration around retirement. Using a flexible non-parametric baseline hazard specification, the proportionate change in the baseline hazard caused by changes in the independent variables is indicated by the model coefficients. The time-varying covariates cause the hazard of migration in year j to be dependent on the value of the time-varying covariates in that year. Six event history models were estimated. Model 1 contains the time indicator variables and household composition predictors. Model 2 is an extension of Model 1 in that spouse characteristics were included. Model 3 adds household characteristics and Model 4 contains additional covariates that capture retirement and occupation in the year before retirement. Models 5 and 6 replicate model 4 separately for short and long distance moves. The parameter estimates, significance levels and goodness-of-fit statistics for the six models are shown in Table 2. The goodness-of-fit statistics improve (i.e. Chi square decreases) from Model 1 to Model 4, for the loss of ten degrees of freedom, which underlines the benefit of including additional covariates. Bootstrap standard errors confirmed the stability of the parameter estimates and the robustness of Model 4.
Table 2

Determinants of retirement migration, 2001–2006

Variable

Model 1

Model 2

Model 3

Model 4

Marginal effectc

Bootstrap

Std. Err.

Model 5

Model 6

Exp(β)

Exp(β)

Exp(β)

Exp(β)

Short-distance

Exp(β)

Longer distance

Exp(β)

Baseline

 Period 1

0.054**

0.040**

0.101**

0.086**

−0.102**

0.037

  

 Period 2

0.040**

0.028**

0.072**

0.062**

−0.130**

0.023

0.017**

0.011**

 Period 3

0.121**

0.085**

0.217**

0.188**

−0.089**

0.060

0.036**

0.073**

 Period 4

0.082**

0.055**

0.150**

0.129**

−0.087**

0.049

0.035**

0.036**

 Period 5

0.083**

0.056**

0.158**

0.136**

−0.076**

0.065

0.015**

0.042**

 Period 6

0.136**

0.086**

0.244**

0.208**

−0.062**

0.108

0.020**

0.125**

Household composition

 Spouse not in labour forcea (single-earner household)

1

1

1

1

  

1

1

 Spouse nonea (single household)

1.316

1.455

1.270

1.167

0.011

0.351

2.428*

0.593

 Spouse also retireda (dual-earner household, spouse retired)

1.738*

1.673*

1.766*

1.670*

0.043

0.486

2.328

2.085*

 Spouse in labour forcea (dual-earner household, spouse not retired)

1.163

0.935

0.850

0.778

−0.016

0.269

2.269

0.491

Spouse characteristics

 Retiree good healthb

 

1

1

1

  

1

1

 Retiree bad healthb

 

1.845**

1.776**

1.569*

0.035*

0.335

1.334

2.279*

 Spouse good healthb

 

1

1

1

  

1

1

 Spouse bad healthb

 

0.806

0.793

0.797

−0.015

0.236

0.544

0.422

 Retiree high schoola

 

1

1

1

  

1

1

 Retiree post-secondarya

 

1.135

1.179

1.123

−0.013

0.275

1.097

1.199

 Retiree postgraduatea

 

0.889

0.871

0.823

0.060

0.770

0.450

0.900

 Spouse high schoola

 

1

1

1

  

1

1

 Spouse post-secondarya

 

1.545

1.695*

1.687*

0.008

0.267

1.692

1.560

 Spouse postgraduatea

 

2.063*

2.093*

1.944*

0.043

0.466

2.304

1.774

Household characteristics

 Home renterb

  

1

1

  

1

1

 Home ownerb

  

0.401**

0.419**

−0.085*

0.115

0.614

0.535

 High household incomeb

  

1

1

  

1

1

 Low household incomeb

  

0.758

0.754

−0.021

0.182

1.191

0.640

Retirement and occupation

 Retirement age at or above 60a

   

1

  

1

1

 Retirement age below 60a

   

1.734**

0.042*

0.290

2.055*

1.344

 Occupation, high workplace activitya

   

1

  

1

1

 Occupation, low workplace activitya

   

0.999

0.000

0.234

0.729

1.803

Log likelihood (final value)

−359.59

−352.65

−346.80

−343.01

  

−152.95

−134.68

df

9

15

17

19

  

18

18

Wald Chi square

578.64

562.05

549.3

541.31

  

321.02

283.68

No. of observations

1,218

1,218

1,218

1,218

  

868

852

No. of migration events

111

111

111

111

  

42

38

No. of households

315

315

315

315

  

283

279

aTime-constant covariate

bTime-varying covariate

cMarginal effects calculated as dy/dx for discrete change of dummy variable from 0 to 1

P ≤ 0.05, ** P ≤ 0.01

A binary (dummy) variable and a series of interaction terms were added to Models 5 and 6 to indicate whether predictors of local and long-distance moves differ between non-attrited and attrited households. We added the binary indicator and interaction terms between the attrition dummy and all other explanatory variables and conducted a Wald test for the equality of the coefficients for non-attritors and attritors (results not shown). Testing the collective contribution of all interaction terms simultaneously, the differences were not statistically significant at the 0.05 level (for local moves: df = 17, Wald’s statistics = 17.19, P > 0.05; for longer distance moves: df = 18, Wald’s statistics = 13.98, P > 0.05), suggesting that sample attrition did not seriously bias the results of our event history analysis. Wald tests conducted separately for each interaction term revealed a significant difference in the effect of retirement age on local moves (df = 1, Wald’s statistics = 5.25, P < 0.05), and a significant difference between non-attritors and attritors in the effect of retirees’ post-secondary education on longer-distance moves (df = 1, Wald’s statistics = 4.22, P < 0.05). All other Wald tests returned non-significant P values. Significant differences in the effect of retirement age between attritors and non-attritors may be a direct reflection of age being significantly related to attrition in the HILDA survey. Watson and Wooden (2004) show that Wave 2 attrition rates were highest among 20–24-year-olds (23.4 %) and declined with age to 8 % among retirees aged 65–74.

General determinants of retirement-related migration

The literature suggests that retirement migrants are on average more likely to be married, have higher incomes, better education levels and fewer health problems than non-migrants at that age (see Biggar 1980; Speare and Meyer 1988; Walters 2000). The results presented here confirm that household composition, health status, education level, tenure and age at retirement influence retirement migration intensities (see Table 2), but, for some characteristics, the strength and direction of these effects varied by type of move. Looking first at the results for all moves, it is apparent that household composition and educational attainment have the most significant effects.

Dual-earner households in which both partners retired during the survey period had a significantly higher likelihood of migrating than single-earner households. Single-person households also displayed a slightly higher probability of moving than male single-earner couples, while dual-earner households in which the female spouse had not retired by the end of the survey had a lower risk. Thus it appears that the presence of a female spouse reduces the probability of migration, unless that spouse retires at the same time as her partner, in which case the chance of migration rises significantly.

Consistent with previous findings, education too elevates the probability of migrating around retirement, but our results show it is the educational attainment of the spouse, rather than the retiree, that has the most significant effect on the hazard. Male retiree households had twice the chance of moving if the female spouse had a postgraduate degree compared to spouses without secondary or tertiary education. Post-secondary qualifications on the part of the retiree also lifted the probability of moving, but the chance was actually reduced by 11 % if the retiree had a postgraduate degree. Perhaps the high mobility associated with well-educated female spouses is related to their higher labour-force participation than less-educated spouses, which in turn has a positive effect on the household’s financial resources, while highly educated retirees prefer to stay in close touch with their prior vocation, even after formal retirement.

Occupation before retirement had no significant effect on the risk of migration but lower household income reduced the likelihood of moving, though the effect was not statistically significant. Not surprisingly, people who owned their homes before moving had a lower hazard of migration compared to those who rented, while those who retired below age 60 had a 70 % greater chance of migrating than those who retired later, confirming the finding reported above. Health status, however, generated more complex outcomes. Poor health on the part of a spouse depressed the likelihood of a move at retirement, lending support to earlier research that associates retirement migration with good health. In the case of the male retirees themselves, however, the relationship was reversed: retirees who experienced bad health were more likely to move than those reporting good health. Previous research has concluded that the determinants of local mobility are different from those shaping longer-distance moves (see Biggar 1980; Wiseman and Roseman 1979), and our findings confirm and elaborate these differences.

Determinants of short-distance retirement moves

Poor health is commonly associated with assistance-related moves, which tend to be short-distance, and Model 5 confirms that poor health on the part of the retiree, though not on the part of the female spouse, elevated the probability of a local move. Households with low incomes also displayed a higher risk of moving locally, possibly reflecting less secure tenure or loss of autonomy, but such moves were not simply the preserve of older people: early retirement was strongly associated with local mobility, more so even than with long-distance migration. Household composition also emerged as a significant factor: single-person households, and those with a retired female spouse or one who was still working, were more likely to make a local move. Thus it was two-person households with a male retiree who had been on a single income that were least likely to move locally, perhaps reflecting financial constraints.

Determinants of longer-distance retirement moves

Longer distance moves around retirement are commonly conceived as amenity-led migration, and our results lend conditional support to this view. Comparing the results from Models 5 and 6 reveals that the effects of household composition, health status and household income differ markedly between short and longer-distance migration (see Table 2). While short-distance moves were common except among two-person households reliant on one income, longer-distance moves were largely the preserve of two-income households in which both partners retired from the labour force at, or near, the same time. Single-person households and those with a spouse still in the labour force had an increased risk of making a local move, but a reduced probability of longer-distance migration. Thus, limited financial resources and female spousal job ties appear to drive or at least facilitate local moves, probably underpinned by housing or assistance-related motives. However, a working female spouse reduces the likelihood of long-distance migration, until that spouse in turn retires, at which point the risk of migration rises sharply. In part at least this is almost certainly a product of the greater assets which dual-income households are able to accrue to finance such a move. It is also consistent with results for the income variable in Model 6: as found in previous research, high household income increased the risk of long-distance migration around retirement (Biggar 1980; Clark and Davies 1990; Haas and Serow 1993; Sommers and Rowell 1992).

In general, retirement migrants tend to be healthier than other types of elderly migrants (Litwak and Longino 1987; Walters 2000, 2002). However, if the male retiree suffered from bad health, households in the HILDA sample had an 80 % higher risk of making a longer-distance move, while if the female spouse had bad health, the risk of moving was reduced by 20 %. In the Australian context, one possibility is that health problems raise the incentive to move from temperate southern latitudes to the subtropical north to ameliorate chronic respiratory problems. Subtropical Queensland has long been a prime destination for retirement migrants from the southern states, especially Victoria (Bell and Hugo 2000).

Conclusion

Analysis of older-age migration commonly positions retirement-related moves around the conventional retirement age of 60–65. In Australia, however, available evidence demonstrates that the transition to retirement occurs over a much broader age-span, extending from 40 to 70 years, around a median age of 60 for men and 57 for women. Ninety per cent of respondents to the nationally representative HILDA survey had withdrawn from the labour force before 65, but only half retired between the ages of 55 and 63. Retirement-related migration shows a similar wide-ranging spread across the age profile, extending across an age span of some 25 years, rather than the 5-year window within which it is commonly conceived. One consequence is to underline the importance of differentiating moves that are associated with retirement from those that occur around the conventional age of retirement. The focus here was on the former, and our principal aim was to establish the link between these two life-course events, and to identify the characteristics that shape this connection.

We computed hazard rates and applied discrete-time event history models to pooled data from a nationally representative panel survey of Australian households which captured the location, labour force participation, household composition and individual characteristics of respondents across six annual waves of the HILDA survey from its inception in 2001. Retirement migrations were defined as those events occurring between 2 years before and 4 years after retirement from the workforce.

The results provide clear evidence that retirement acts as a trigger to migration, and that the incidence and timing of such moves is closely related to the retirement transition (Bures 1997; Frey 1986; Longino et al. 2008). The propensity to migrate peaks in the year of retirement and declines with increasing time following withdrawal from the labour force. Fully 40 % of all moves occurred in the retirement year. Age at retirement has a strong conditioning effect on the probability of making a retirement-related move, in that early retirees have a higher propensity to migrate than those who leave the workforce later in life. As retirement age rises, the propensity to migrate falls, but the migration hazard becomes increasingly concentrated around the retirement event. We interpret these patterns as a product of shifts in the balance of place ties, opportunities and constraints as individuals age.

Previous research has identified health and income as key factors facilitating retirement migration, and these also emerge from analysis of the HILDA data. However, it is household characteristics, in particular the presence and labour-force attachment of the spouse, that stands out as the most significant arbiter of migration around retirement in the Australian context, especially over longer distances. Households with a working spouse who retired from the labour force around the same time as his or her partner had a significantly higher likelihood of making a long-distance retirement move. In contrast, a partner who continued to work reduced the chance of long-distance migration, and the same was true for households in which no spouse was present. Short-distance moves displayed a quite different profile, in which it was two-person households on a single income that had the lowest chance of moving. For households with a spouse who continued to work, local ties and income appear as the likely intermediate variables underpinning these differences, facilitating residential mobility by increased financial capacity but constraining longer-distance migration. Household income mediated migration among the HILDA sample in the same way as reported by Clark and Davies (1990) and by Sommers and Rowell (1992): high-income levels increase the risk of longer-distance migration around retirement while short-distance retirement moves were more common among low-income households. The effects of health status, on the other hand, were inconsistent with prior findings that good health facilitates migration on retirement (Biggar 1980; Clark and Davies 1990; Sommers and Rowell 1992). Poor health on the part of a spouse reduced long and short-distance mobility, but retirees in poor health had a significantly increased chance of a long-distance move; a pattern that is consistent with the long standing northwards drift of retirees away from the harsher winters of southern Australia.

The need for a clear understanding of retirement migration is increasingly important given the impending retirement of the baby-boom cohorts. Compared to earlier generations of retirees, the baby boomers are more affluent, better educated, experienced in travel, and more likely to have dual incomes, reflecting the rise in labour-force participation of married women. Based on the results presented here, this suggests an increase in the probability of migration among the baby-boom generation. However, increasing retirement age, postponement of the empty-nest transition and a rise in the proportion of single-person households could exert a counteracting effect on the propensity to move.

Beyond the substantive findings reported here, one important implication of this analysis is to underline the need for caution when identifying retirement migration using age as a proxy measure, as is commonly the case when using Census data. Variations between respondents in the timing of retirement were such that only a quarter of all moves at ages 55–69 were made within 3 years before or after retirement. Most moves in this age group were undertaken by those who were still working or were outside the labour force. Ultimately all migration decisions are multifaceted, and distilling the specific causes of residential change around retirement calls for datasets that tease out the links between key events in a life course framework, which recognizes the household context as well as the individual characteristics of the migrants. What is further needed, and not commonly available even from panel data sources, is a nuanced understanding of the particular mix of motives that drive decisions to relocate around these critical life course events.

Footnotes

Notes

Acknowledgments

This paper was supported financially by an Australian Research Council Discovery Grant DP 0451399, Understanding the Structure of Internal Migration in Australia. This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or the Melbourne Institute. The authors would like to thank Jutta Gampe and Temesgen Kifle for their comments on an earlier version of the paper.

References

  1. Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data. Newbury Park: Sage.Google Scholar
  2. Bell, M., Blake, M., Boyle, P., Duke-Williams, O., Rees, P., Stillwell, J., & Hugo, G. (2002). Cross-national comparison of internal migration: Issues and measures. Journal of the Royal Statistical Society A, 165(3), 435–464.CrossRefGoogle Scholar
  3. Bell, M., & Hugo, G. (2000). Internal migration in Australia 1991–1996: Overview and the overseas-born. Canberra: Department of Immigration and Multicultural Affairs.Google Scholar
  4. Biggar, J. C. (1980). Who moved among the elderly, 1965–1970: A comparison of types of older movers. Research on Ageing, 2, 73–91.CrossRefGoogle Scholar
  5. Boyle, P., Cooke, T. J., Halfacree, K., & Smith, D. (2001). A cross-national comparison of the impact of family migration on women’s employment status. Demography, 38, 201–213.CrossRefGoogle Scholar
  6. Boyle, P. J., Kulu, H., Cooke, T., Gayle, V., & Mulder, C. H. (2008). Moving and union dissolution. Demography, 45, 209–222.CrossRefGoogle Scholar
  7. Bradley, D. E. (2011). Litwak and Longino’s developmental model of later-life migration: Evidence from the American Community Survey, 2005–2007. Journal of Applied Gerontology, 30(2), 141–158.CrossRefGoogle Scholar
  8. Bradley, D. E., Longino, C. F., Stoller, E. P., & Haas, W. H. (2008). Actuation of mobility intentions among the young-old: An event-history analysis. The Gerontologist, 48(2), 190–202.CrossRefGoogle Scholar
  9. Bures, R. M. (1997). Migration and the life course: Is there a retirement transition? International Journal of Population Geography, 3, 109–119.CrossRefGoogle Scholar
  10. Clark, W. A. V., & Davies, S. (1990). Elderly mobility and mobility outcomes. Research on Aging, 12, 430–462.CrossRefGoogle Scholar
  11. Courgeau, D., & Lelièvre, E. (1992). Event history analysis in demography. Oxford: Clarendon Press.Google Scholar
  12. Flowerdew, R., & Al-Hamad, A. (2004). The relationship between marriage, divorce and migration in a British data set. Journal of Ethnic and Migration Studies, 30, 339–351.CrossRefGoogle Scholar
  13. Frey, W. H. (1986). Lifecourse migration and redistribution of the elderly across US regions and metropolitan areas. Economic Outlook USA, 13, 10–16.Google Scholar
  14. Grundy, E. (1985). Divorce, widowhood, remarriage and geographic mobility among women. Journal of Biosocial Science, 17, 415–435.CrossRefGoogle Scholar
  15. Haas, W. H., & Serow, W. J. (1993). Amenity retirement migration process: A model and preliminary evidence. The Gerontologist, 33, 212–220.CrossRefGoogle Scholar
  16. Himes, C. L. (2001). Elderly Americans. Population Bulletin, 56, 3–40.Google Scholar
  17. Hugo, G., & Bell, M. (1998). The hypothesis of welfare-led migration to rural areas: The Australia case. In P. Boyle & K. Halfacree (Eds.), Migration into rural areas: Theories and issues. London: Wiley.Google Scholar
  18. Jenkins, S. P. (1995). Easy estimation methods for discrete-time duration models. Oxford bulletin of economics and statistics (vol. 57, pp. 129–138). Oxford: Department of Economics, University of Oxford.Google Scholar
  19. Kitching, R. (1990). Migration behaviour among the unemployed and low-skilled. In J. J. Johnson & J. Salt (Eds.), Labour migration: The internal geographical mobility of labour in the developed world (pp. 172–190). London: David Fulton.Google Scholar
  20. Law, C. M., & Warnes, A. M. (1975). Life begins at sixty: The increase in regional retirement migration. Town and Country Planning, 43, 531–534.Google Scholar
  21. Law, C. M., & Warnes, A. M. (1982). The destination decision in retirement migration. In A. M. Warnes (Ed.), Geographical perspectives on the elderly (pp. 53–82). New York: John Wiley.Google Scholar
  22. Litwak, E., & Longino, C. F. (1987). Migration patterns among the elderly: A developmental perspective. The Gerontologist, 27, 266–272.CrossRefGoogle Scholar
  23. Longino, C. F., Jr, Bradley, D. E., Stoller, E. P., & Haas, W. H. (2008). Predictors of non-local moves among older adults: A prospective study. Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 63, S7–S14.CrossRefGoogle Scholar
  24. Lutz, W., Sanderson, W., & Scherbov, S. (2008). The coming acceleration of global population ageing. Nature, 451, 716–719.CrossRefGoogle Scholar
  25. Mulder, C., & Hooimeijer, P. (1999). Residential relocations in the life course. In L. van Wissen & P. Dykstra (Eds.), Population issues: An interdisciplinary focus (pp. 159–186). New York: Kluwer Academic/Plenum Publishers.CrossRefGoogle Scholar
  26. Organisation for Economic Coooperation and Development (OECD) (2011). Trends in retirement and in working at older ages, in OECD, Pensions at a glance 2011: Retirement-income systems in OECD and G20 countries, OECD Publishing.Google Scholar
  27. Prentice, R., & Gloeckler, L. (1978). Regression analysis of grouped survival data with application to breast cancer data. Biometrics, 34, 57–67.CrossRefGoogle Scholar
  28. Rogers, A. (1988). Age patterns of elderly migration: An international comparison. Demography, 25, 355–370.CrossRefGoogle Scholar
  29. Rogers, A., & Castro, L. J. (1981). Model migration schedules. In Research report 81–30. Laxenburg, Austria: International Institute for Applied Systems Analysis.Google Scholar
  30. Rogers, A., Raquillet, R., & Castro, L. J. (1978). Model migration schedules and their applications. Environment and Planning A, 10, 475–502.CrossRefGoogle Scholar
  31. Singer, J. D., & Willett, J. B. (1993). It’s about time: Using discrete-time survival analysis to study duration and the timing of events. Journal of Educational and Behavioral Statistics, 18, 155–195.CrossRefGoogle Scholar
  32. Smith, D. B., & Moen, P. (1998). Spousal influence on retirement: His, her and their perception. Journal of Marriage and the Family, 60, 734–744.CrossRefGoogle Scholar
  33. Sommers, D. G., & Rowell, K. R. (1992). Factors differentiating elderly residential movers and nonmovers. Population Research and Policy Review, 11, 249–262.CrossRefGoogle Scholar
  34. Speare, A., & Meyer, J. W. (1988). Types of elderly residential mobility and their determinants. Journal of Gerontology, 43, S74–S81.CrossRefGoogle Scholar
  35. Thomas, D. S. (1938). Research memorandum on migration differentials. New York: Social Science Research Council.Google Scholar
  36. van Solinge, H., & Henkens, K. (2005). Couples’ adjustment to retirement: A multi-actor panel study. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 60, S11–S20.CrossRefGoogle Scholar
  37. Walters, W. H. (2000). Types and patterns of later-life migration. Geografiska Annaler: Series B, Human Geography, 82, 129–147.CrossRefGoogle Scholar
  38. Walters, W. H. (2002). Later-life migration in the United States: A review of recent research. Journal of Planning Literature, 17, 37–66.CrossRefGoogle Scholar
  39. Warnes, A. M., & Law, C. M. (1984). The elderly population of Great Britain: Locational trends and policy implications. Transactions of the Institute of British Geographers, 9(1): 37–59.Google Scholar
  40. Warnes, A. M., Friedrich, K., Kellaher, L., & Torres, S. (2004). The diversity and welfare of older migrants in Europe. Ageing and Society, 24, 307–326.CrossRefGoogle Scholar
  41. Watson, N., & Wooden, M. (2002). The Household, Income and Labour Dynamics in Australia (HILDA) Survey: Wave 1 survey methodology. HILDA Project Technical Paper Series. Melbourne: University of Melbourne.Google Scholar
  42. Watson, N., & Wooden, M. (2004). Sample attrition in the HILDA Survey. Australian Journal of Labour Economics, 7, 293–308.Google Scholar
  43. Wiseman, R. F. (1980). Why older people move: Theoretical issues. Research on Aging, 2, 141–154.CrossRefGoogle Scholar
  44. Wiseman, R., & Roseman, C. (1979). A typology of elderly migration nased on the decision making process. Economic Geography, 55, 324–337.CrossRefGoogle Scholar
  45. Wroe, D. (1973). The elderly. Social Trends, 4, 23–24.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 2013

Authors and Affiliations

  1. 1.Wittgenstein Centre (IIASA, VID/ÖAW, WU)Vienna Institute of Demography, Austrian Academy of SciencesViennaAustria
  2. 2.Queensland Centre for Population Research, School of Geography, Planning and Environmental ManagementThe University of QueenslandBrisbaneAustralia

Personalised recommendations