| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Reynolda Gerontology Program and 3 Department of Sociology, Wake Forest University, Winston-Salem, North Carolina.
2 Department of Sociology, East Carolina University, Greenville, North Carolina.
4 Department of Sociology, University of North Carolina at Asheville.
Address correspondence to Charles F. Longino, Jr., Reynolda Gerontology Program, Box 7808, Wake Forest University, Winston-Salem, NC 27109. E-mail: longino{at}wfu.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. We employed data from the Health and Retirement Study, 1994 to 2002, based on a representative sample of households containing at least one member aged 51 to 61 in 1992. We employed measures available in the Health and Retirement Study to construct a series of Cox proportional hazards models that examined the causes of non-local moves.
Results. Community and person ties emerged as important predictors of non-local moves.
Discussion. Travel experience, when measured by regular vacationing and second homes, may increase community ties to a destination. The life-course model must be modified in its explanation of the importance of community and person ties, and of life transitions, as motivators of migration.
Key Words: Migration • Predictors • Prospective • Older Adults
TWENTY years ago, Litwak and Longino published an article laying out the rationale for what they called a life-course model of later life migration (Litwak & Longino, 1987
). Because the life course includes a life-long accumulation of experiences and some major transitions, their use of the concept was perhaps overly ambitious. The life course is only glimpsed in census data. Litwak and Longino argued that people of retirement age are free to move long distances from their family members because transportation and electronic communication technology make it possible to keep in close touch over the miles while pursuing some ideal retirement lifestyle. They called these amenity or lifestyle-driven moves. They continued with the assertion that family ties, especially relationships with children, tend to become more important when one's disability level increases. Health problems, therefore, particularly in the face of widowhood, can motivate movement toward kin. These are called assistance moves. A third type of move comes when health care needs overwhelm the abilities of close family members and others to provide care. These are called nursing home moves. Litwak and Longino did not assume or imply that all or even most older persons would move. Most stay rooted in places that they love (Longino, 2006).
One can see evidence supporting this life-course theory of migration in the tendency for migrants in the major migration streams to popular destination states like Florida from major origin states like New York to be younger, more often married, and residentially independent with lower levels of disability. Counterstream migrants moving in the opposite direction (e.g., to major northeastern and midwestern states from Florida) are older and more often widowed. More are disabled, and again more are living dependently with relatives. These observations fit the model, although without a time dimension it is not possible to link the persons in the two streams.
During the decade that followed the publication of Litwak and Longino's (1987) article, several other studies, using longitudinal data from the Longitudinal Study on Aging, supported the theory, particularly concerning the assistance move. A very narrow range of causal factors was available in the data set. In addition, its respondents were aged 70 at baseline, so it missed the full range of the life-course model (Bradsher, Longino, Jackson, & Zimmerman, 1992; Clark & Wolf, 1992; G. Lin & Rogerson, 1995; Longino, Jackson, Zimmerman, & Bradsher, 1991; Miller, Longino, Anderson, James, & Worley, 1995; Silverstein, 1995; Speare & McNally, 1992).
Taking a functionalist point of view, Litwak and Longino (1987) asserted that only family ties, particularly those of children, can sustain the demands of routine care of a disabled older relative in the community, and the strength of these personal ties tends to motivate assistance moves. Rowles (1990) demonstrated that enduring close friendships among older people in small towns substitute well for kinship ties. Unspoken by Litwak and Longino was the assumption that in the absence of disabling health problems, these personal ties would not strongly motivate or deter long-distance moves. This is an assumption that we test in this article. The image that emerges from the life-course model is that people move away from their children and other kin when making amenity moves, and toward them when making assistance moves.
In reality, early moves, which take place when people are in their 50s and 60s, are also affected by individuals' cumulative life experiences and attributes, including demographic characteristics (such as age, gender, marital status, and race), resource level (such as education, wealth, income, and health), earlier mobility experience, as well as ties to the community and persons on both ends of the move. We are deliberately breaking no new ground in suggesting that these particular factors predict migration. These factors derive from the earliest studies of migration (Lee, 1966; Ravenstein, 1885) in demography and were applied to old-age migration by Roseman and Oldakowski (1984), Rowles (1990), McHugh and Mings (1996), and Cuba (1989). The life-course model tends to ignore this cluster of causal factors and focuses instead on the pull of lifestyle-related forces at the destination of the first move, and kinship ties at the destination of the second move, a perspective that is far too narrow.
Even if older people have accumulated the economic and other resources needed to make the move and the travel experience that would enable them to imagine moving, strong community and person ties may hold them back, mooring them to their place of origin, or pull them toward a destination. Similarly, strong ties to the destination and weakening ties at origin may constitute a "pull," in Ravenstein's (1885) sense, to the destination.
Longino, Perzynski, and Stoller (2002) interviewed older "permanent" migrants to Florida and found that some of them remained ambivalent about staying in Florida because old community and person ties to their origin still pulled them back. Vacationing "back home" was one way of resolving that tension. No prospective study of older movers has demonstrated the relative power of these various factors in motivating a move.
McHugh and Mings (1996) identified three archetypes of seasonal migrants. The still-rooted have strong ties to their home place. The suspended feel equally at home in both places. Finally, the footloose have weak place ties back home due primarily to the loss of person ties there. Accordingly, the strength of place ties varies among seasonal migrants at both their origin and their destination.
Theoretically based research on later life geographic mobility has lagged in large part because appropriate data sets have not been available for testing dynamic behavioral models. In the past, the U.S. Census mobility item, for example, has asked where one lived 5 years before the census. However, because most other information is solicited at the time of the census (i.e., after the move), it is inappropriate to test causal models of moving using these data (e.g., Longino, 1995; Longino & Bradley, 2003).
Local surveys, although cross-sectional, can tailor retrospective questions to test causal models (Haas & Serow, 1993). However, the problem with local surveys is that where migration has already occurred, some element of rationalization may be present in responses. For these reasons, we argue that retrospective studies are inferior to prospective studies in predicting outcomes.
Some influential studies have sought to predict migration expectations. Without knowing who actually moved, however, these pseudo-prospective studies are difficult to interpret (Roseman & Oldakowski, 1984). At best they reinforce theoretical links to mobility that should be tested prospectively for actual movers.
Discovering and affirming the causal mechanisms underlying geographic mobility among older adults has awaited the availability of longitudinal data with appropriate measures. The present research makes use of panel data collected between 1994 and 2002 from a nationally representative sample of households containing at least one member between the ages of 51 and 61 in 1992. As such, what follows represents the first effort to test a causal model of non-local moves among people in their 50s and 60s using appropriate, time-sequenced data. We addressed the following research question: How are demographic covariates, resources, travel experience, and especially community ties and family propinquity related to non-local mobility?
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
), the core sample for the HRS is based on a stratified multistage area probability sampling of households containing at least one person born between 1931 and 1941. Within selected households, all persons who were age eligible (i.e., born between 1931 and 1941) were interviewed along with spouses or partners whether the latter were age eligible or not. The present analysis included only age-eligible respondents who provided complete interviews in 1992 and 1994. Our data extract included 8,280 individual respondents representing 6,601 households. The number is both pre and post weighting.
In addition to the core sample, African Americans and Hispanics were oversampled at approximately twice the rate of non-Hispanic Whites. Florida residents were also oversampled in order to give attention to an area where older residents are especially dense and numerous. Our analyses employed person-level weights that (a) adjusted for differential probabilities of selection, arising in part from the oversampling of selected populations, including Florida residents; and (b) included a poststratification adjustment based on population totals from the 1990 decennial census across census region by race, gender, and age category. Analytic models employed normalized weights. This means that the weights were rescaled to add up to the actual sample size of 8,280.
Face-to-face interviews were conducted with panel members in 1992. Follow-up interviews by telephone have been conducted every other year since. The present study exploited five waves of data collected between 1992 and 2002. Notably, about one third of the original HRS sample members were lost to attrition. Existing research using diverse panel study data has suggested that attrition effects on sample representativeness tend to be mild (Alderman, Behrman, Kohler, Maluccio, & Watkins, 2001; Lillard & Panis, 1998; McCurdy, Mroz, & Gritz, 1998), in part because most attrition appears to be random (e.g., Fitzgerald, Gottschalk, & Moffitt, 1998). Moreover, Cao and Hill's (2005) analysis of the HRS data suggested that (a) attrition has not undermined the representativeness of the sample, and (b) panel members lost for reasons other than mortality, active attritors, appear to be quite similar to those remaining in the study.
It may be that non-local moves increase the chances that a respondent will be lost at follow-up, leaving the panel as an active attritor. Bias may result to the extent that lost non-local movers are systematically different from non-local movers who remain in the study. Available data did not permit us to distinguish lost non-local movers from other active attritors. As an alternative, we conducted a sensitivity analysis, replicating our full model under a worst case scenario. Models that assumed all active attritors made a non-local move immediately after the last completed interview supported the same substantive conclusions as those reported in the Results section, with only a handful of exceptions. These results suggest that bias generated by lost movers was unlikely to be a serious problem.
Analytic Strategy
We explored non-local moves among persons in their 50s and 60s by using event history analysis techniques. We estimated the monthly hazard of non-local moves across respondents at risk for making such a move. Respondents entered the risk set at the date of the 1994 interview and remained until they made a non-local move, died, or dropped out of the panel. We chose 1994 as the baseline because limited information was available to characterize moves made between 1992 and 1994.
The analysis estimated Cox's proportional hazards models, which take the following form:
|
|
0(t), is a constant and maximizes a partial likelihood function in order to generate estimates of the influence of covariates (Allison, 1995).
Model coefficients β1 ... βk indicate the proportionate adjustment to the baseline hazard rate implied by changes in selected independent variables x1 ... xk. Exponentiated coefficients, termed hazard ratios, are comparable to odds ratios in their interpretation. With respect to ordinal or interval measures, for example, 100 x (exp βi – 1) provides the anticipated percentage change in the estimated hazard of some event for each unit change in a given covariate (Allison, 1995). By way of illustration, Equation 1 incorporates one time-dependent covariate, β1x1i(t), indicating that the hazard of non-local mobility at month t is dependent on the value of x1 at month t. In this analysis we employed a number of time-dependent covariates, described in detail below.
Given the complex multistage sampling process used to select HRS panel members, statistical analyses that fail to account for homogeneity within sampling clusters will yield standard errors that are downwardly biased, increasing the risk of Type I errors (e.g., Kalton, 1983). With respect to mobility, we followed a procedure outlined by Boudreau and Lawless (2006) to construct proportional hazards models stratified by sampling cluster using PROC PHREG within SAS, whereby we based hypothesis tests on D. Y. Lin and Wei's (1989) robust sandwich estimates of the covariance matrix.
Measures
Dependent variable
Event history analysis models the occurrence and timing of events. As to occurrence, non-local moves were measured by a series of questions available at each wave from 1996 to 2002. We classified persons as having changed residence if they no longer had the same address as reported at the previous wave. A non-local move occurred when an individual (a) changed residence and (b) no longer lived "in or around" the town or city reported at the previous wave. Between 1994 and 2002, 1,071 respondents, or about 13% of the sample, reported having made at least one such move. This figure was roughly comparable to estimates from the 2000 census suggesting that about 11% of adults aged 55 to 64 had made an intercounty or interstate move within the previous 5 years (He & Schacter, 2003). This non-local move measure was not the same as that used by the Census Bureau, which defines internal migration as movement across political boundaries (county or state lines). Rather, this was a measure of non-local moves that could extend to a town a few miles away or to a municipality on the other side of the continent. We assumed that the vast majority of these moves were across political boundaries, although we use the more conservative term, non-local moves, here.
As to timing, we assigned non-local movers a time-to-event value reflecting the number of months between the date of interview in 1994 and the first subsequent non-local move. We right-censored time-to-event scores for those not making a local move, indicating that observation ended before a non-local move occurred.
Independent variables
The strength of this project lay in the availability of measures prior to the move. The panel data also allowed the opportunity to consider change. Analyzing the implications of such significant life-course events as retirement and widowhood on the decision to make a non-local move allowed us to examine the dynamic process of geographical mobility.
Demographic covariates
Table 1 describes each of the predictors used in the analysis. The covariates represented master statuses such as gender, age, race/ethnicity, and marital status that influence a wide range of outcomes and are standard controls. We used baseline marital status and subsequently reported changes (i.e., marital dissolution, new marriages) to construct a series of dichotomous variables indicating marital status at any given month. What kind of impact did the loss of a spouse have on non-local mobility? In order to address this question, we created three time-dependent items indicating the recency of widowhood. To illustrate, barring remarriage, a respondent widowed at Month 40 was coded as widowed 6 months or lesst for Months 40 to 46, as widowed 7 to 12 monthst at Month 47, and as widowed more than a year at Month 53 and following. Incorporating the widowhood transition items required that analysis begin at the 13th month subsequent to baseline.
|
). Given that wealth and income measures are typically skewed, we employed the natural log of both household income and household wealth in our models. Better education, more financial resources, and better health were resources expected to facilitate a move.
Travel experience
Table 1 defines measures of travel experience.
Community ties
These were measured in several ways. If a person knows more of his or her neighbors by name, one can also assume him to be more rooted in or tied to the place. It is the social landscape of the neighborhood that is more important to this concept, and less the individuals. If a person has made a recent local move to a new home, or owns rather than rents, community ties also should be stronger because of the investment in housing. If a person is an area native, that is, he or she was born in the area, one can assume community ties to be strong. Finally, severing employment ties is a major life-course event that should increase the likelihood of a non-local move. It may be that the effect of retirement at any given point in time depends on how recently the transition out of the labor force occurred. In order to evaluate this idea we constructed a series of dichotomous time-dependent items that captured retirement status at any given month. We used the earliest date when a respondent reported being completely retired or partly retired to identify those at any given month as retired more than a yeart, retired 7 to 12 monthst, or retired 6 months or lesst. We coded persons out of the labor force (e.g., homemakers) who did not provide retirement date information as having been retired at baseline.
Person ties
We measured these in two ways: by the number of children or children-in-law, and parents or parents-in-law, who lived either with or within 10 miles of the respondent. The 10-mile boundary was arbitrary and thereby created certain data-imposed limitations. Moreover, parent and child proximity does not measure the amount of contact or the closeness of relationships. It was not possible to include person ties at destination in the event history analysis, but we considered this factor separately.
Time-dependent covariates are denoted with the subscript 94-00, indicating that at any given month we estimated the influence of a particular covariate using the most recent measurement of that variable. Except where otherwise noted, measures reflect individual respondent reports. On selected items, designated respondents answered specific items for all members of a household. The family respondent, for example, was asked to provide information on the residency status of parents and children. Household-level measures are designated with the subscript h (e.g., Children Near94-00 h).
We tested independent variables for multicollinearity. The analysis found that across independent variables in no case did the proportion of variability explained by the other variables in the model rise above 44%; therefore, we concluded that multicollinearity was not a problem.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
). The effect of being female was not significant. Also, note that age was negatively associated with the likelihood of moving. Concerning marital status, however, formerly married persons were more likely than married or partnered individuals to make an out-of-area move, whereas never-married persons were less likely to move. With respect to widowhood, those widowed 6 months or less at any given month were more than twice as likely as their married or partnered counterparts to make a non-local move. The estimated hazard ratio of widowed 7 to 12 months was of similar magnitude, though only marginally significant (p =.068). By contrast, respondents widowed for more than a year were no different from married or partnered individuals with respect to the hazard of making a non-local move.
|
Travel Experience
Model 3 contained an anomaly. Net of covariates, having a place in which one regularly vacations and a second home or condominium increased the likelihood of a non-local move by about 38% and 31%, respectively. However, having a recreational vehicle suppressed mobility, and a past history of military service had no influence.
Community Ties
Model 4 told an interesting story. Community ties, such as having been born in or near one's current community, suppressed the hazard of making a non-local move. In addition, households in which the designated respondent knew a large portion of the neighbors by name were also less likely to move. In addition, a recent local move to a new home enhanced the risk of a subsequent non-local move. As expected, being rooted in the community, in these ways at least, did make a difference, and that difference was in most cases substantial. Homeowners, for example, were approximately 55% less likely to make a non-local move than those who did not own their home.
Moreover, severing the work tie to the community was influential in predicting a move. Retirement made a difference in mobility when the covariates and resource variables were taken into account, and the effect size was larger the more recent the retirement. Findings presented in Model 4 suggest that, at any given month, those retired 6 months or less were nearly 5 times as likely to make a non-local move as compared to respondents who were not retired.
Person Ties
Model 5 argued for the strong influence of person ties at origin in suppressing migration, when measured as family propinquity. The more children who lived with or within 10 miles of the respondent, the more suppressed the hazard of a non-local move. The same could be said for the residential proximity of one's parents or parents-in-law, although its effect was less than that of the proximity of children.
Model 6 included all independent variables. Though space limitations prohibit an exhaustive discussion, we highlight selected differences between Model 6 and previous models. With respect to demographic covariates, findings were relatively stable. One exception to this pattern, divorced or separated was dramatically reduced in Models 6 and 4 as compared to Model 1. Higher levels of non-local mobility among the formerly married appear to reflect relatively weak community and person ties. Also noteworthy, widowed 7 to 12 months was positive and significant in Models 5 and 6 after accounting for family propinquity and other mobility predictors.
Of the resource variables, in Model 6 as compared to Model 2, the effect of household wealth was substantially reduced and no longer significant. Ancillary analysis suggested that this change reflected the inclusion of home ownership in Model 6. Wealthier persons were more likely to be homeowners, and homeowners were less likely to make a non-local move.
With respect to measures of travel experience, recreational vehicle was no longer significant at the.05 level in Model 6 as compared to Model 3. However, this was not a substantively important change inasmuch as there was only a modest difference in coefficient magnitude. Moreover, recreational vehicle in Model 6 continued to be marginally significant (p =.07).
The robust influence of community and person ties was evident here inasmuch as most of the relevant measures continued to influence migration despite the range of controls introduced in the final model. Notably, area native was somewhat weaker and no longer significant in Model 6 as compared to Model 4. Further analysis suggested that this change was due to family propinquity measures included in Model 6. In short, area natives were less likely to move, largely because they were more likely to have children or parents living nearby.
Our analysis highlighted the importance of person ties at origin, but how often are non-local moves motivated by person ties at destination? In order to address this question, we exploited reason-for-move items that were asked of respondents who had changed their main residence since the previous data collection wave. Multiple mentions were allowed, but most respondents provided only one reason for moving. Results from these retrospective assessments suggested that person ties at destination often motivated non-local moves among the young old.
Of the 1,071 respondents who had made a non-local move, 14.8% mentioned the desire to be near or with children as a reason for moving. Proximity to other relatives or friends was mentioned as an important motivating factor by 11.8% of non-local movers. Taken together, about 26% mentioned either proximity to children or friends and other relatives as a factor motivating relocation. A small number (n = 14) mentioned both.
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Also, one should be aware that the decade when the respondents were repeatedly interviewed, 1992 to 2002, may have left its mark on this cohort. Panel members were born before the baby boom, 1931 to 1941, and most of them were interviewed repeatedly during a time of relative peace internationally, and prosperity at home, during the Clinton years. This respondent cohort may not be ideal as a model for the more anxious post-9/11 world. In addition, baby boom aging will increase mobility in absolute numbers, even if there is no change in the proportion who move (Rogers & Rajbhandary, 1997). Finally, taking a longer term perspective, the United States may be approaching the final stages of its elderly mobility transition, which may usher in a period of reduced mobility in the future and perhaps a gradual change in the factors that motivate moves (Rogers, Watkins, & Woodard, 1990). Short-term aspects of climate change, whether or not they are connected to hypothesized long-term changes, may operate as a wild card. Hurricanes come to mind.
Furthermore, this analysis begs the question of what it "means" to retire early (i.e., between ages 51 and 61) in the context of geographical mobility. These issues are discussed in earlier analyses (Haas, Bradley, Longino, Stoller, & Serow, 2006; Longino et al., 2002).
Finally, one aspect of person ties not considered in this analysis is the couple tie. We presumed the persons studied here to be isolated individuals, making their choices independently. We know that this is an unrealistic assumption for married people. The spouse must be considered if the move is to succeed. Community and person ties may not operate identically for both husband and wife.
The major innovative aspect of this study is its prospective nature. We deliberately drew the variables included in the model categories from the established research literature. They are not novel. The purpose of the study was to take established concepts and measures commonly expected to increase the hazard of mobility and put them to the test in the kind of analysis in which it is actually possible to observe this process across time. We accomplished this goal.
Another goal was to include some important life events in the model. The life-course theory of retirement migration has actually been used as a very static model. The life course is dynamic (Rubinstein & Parmalee, 1992). It involves the accumulation of human experience and the effect of major life transitions. We included two such transitions in this study, retirement and widowhood, and found that both were robust and survived to Model 6 in Table 2. The life course is an important way of organizing explanations of non-local moves.
To synthesize and interpret some of the key findings, people in their 50s and 60s who have moved around in their work, and whose children, if they have children, have moved away for education and employment, are at higher risk for making an out-of-area move. In addition, those who have moved into their homes more recently than other respondents, perhaps because they are more mobile, tend to have higher education levels and are at higher risk for making non-local moves.
Travel experience is a robust category. Except for military service, the variables have an independent affect (i.e., they maintain their power to the final full model). Having a regular place to take one's vacations and a second home tend to strengthen community ties at the destination, at least for those who choose those places to move when they retire. Having a recreational vehicle may, in some ways, be a substitute for making a more permanent move. In addition, having a mobile home also may reflect an interest in exploring different, perhaps multiple, locations rather than committing to a single destination (McHugh & Mings, 1996).
Viewed in this way, community and person ties are the core of the theoretical constructs influencing the hazard of moving both in the years of one's late-life health decline and in one's late-work and early-retirement years. Ties both to origin and destination are important, and person ties seem to be as important in mobility considerations as community ties.
Friends, or at least neighbors, become anchors to one's origin, along with the close proximity of parents and children. The friends and family members who live in or near the potential destination are important, too, as Lee (1966) suggested and as we have shown. Relative attachment to property as measured by recent residence and tenure (owning or renting) also plays its part in mooring individuals to a community.
Finally, geographic mobility may have been studied longer in demography than geography. Nonetheless, the rich tradition of conceptually based research on place and person ties in geography (McHugh, 1984; Roseman & Oldakowski, 1984; Rowles, 1978, 1983, 1990) gives this academic enterprise special relevance for later life mobility research. Framing research from the point of view of only one of these traditions tends to disenfranchise the other and impoverish both.
|
Acknowledgments |
|---|
The idea for the article and the first draft of the introductory and Discussion sections were contributed by Charles F. Longino, Jr.; the Methods and Results sections were initially contributed by Don E. Bradley; and Eleanor P. Stoller and William H. Haas, III, contributed their suggestions for the first and successive drafts. The order of authorship reflects the level of contribution to the finished manuscript.
|
Footnotes |
|---|
Received for publication January 11, 2007. Accepted for publication September 26, 2007.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
