The Gerontologist Vol. 48, No. 2, 190–202 Copyright 2008 by The Gerontological Society of America Actuation of Mobility Intentions Among the Young-Old: An Event-History Analysis Don E. Bradley, PhD,1 Charles F. Longino, Jr., PhD,2 Eleanor P. Stoller, PhD,3 and William H. Haas, III, PhD4 Purpose: Although migration decision making is central to understanding later-life migration, the critical step between migration intentions and mobility outcomes has received only limited empirical attention. We address two questions: How often are intended moves actuated? What factors condition the likelihood that mobility intentions will be actuated? Design and Methods: We employ data from the 1994–2002 Health and Retirement Study, which is a nationally representative panel targeting households containing persons aged 53 to 63 years at baseline. Event-history techniques are used to examine the link between reported mobility intentions at baseline and mobility outcomes across the study period, net of relevant controls. We conduct separate household-level analyses for couple and noncouple households and recognize three types of moves: local, family oriented, and nonlocal. Results: Findings confirm the utility of mobility expectations as a predictor of future mobility. More importantly, results highlight the complex nature of later-life mobility. The actuation of mobility intentions appears to operate differently in couple than in noncouple households. Moreover, our findings suggest that the role of several key variables depends on the type of move under consideration. Implications: The ability to identify potential ‘‘retirement migrants’’ may be of practical importance for state and local government officials as well as developers interested in recruiting or retaining young-old residents. Our study We gratefully acknowledge support under Grant R03-AG023813 from The National Institute on Aging, which that made this research possible. D. E. Bradley extends additional thanks for support from East Carolina University’s Faculty Senate in the form of a Research and Creative Activity Grant. We also appreciate Ms. Peggy Beckman for the administrative assistance she has provided. Address correspondence to Don E. Bradley, PhD, Department of Sociology, East Carolina University, 442-A Brewster Building, Greenville, NC 27858. E-mail: bradleyd@ecu.edu 1 Department of Sociology, East Carolina University, Greenville, NC. 2 Reynolda Gerontology Program, Wake Forest University, WinstonSalem, NC. 3 Department of Sociology, Wake Forest University, Winston-Salem, NC. 4 Department of Sociology and Anthropology, University of North Carolina at Asheville. 190 offers insight on the interpretation of stated mobility intentions. Moreover, consistent with early theoretical work in the field, our analysis suggests that empirical studies must account for heterogeneity among older movers in order to avoid misleading results. Key Words: Migration, Residential mobility, Retirement migration An established and vibrant research literature examines the causes, patterns, and consequences of later-life migration (for reviews, see Longino & Bradley, 2006; Walters, 2002). Indeed, the geographic mobility of older adults is an important topic worthy of attention for at least two reasons. First, available evidence suggests that the number of older adults who move is large and likely to increase in the foreseeable future. Data from the 2000 U.S. Census suggests that about 23% of those persons 65 years and older and 26% of those aged 55 to 64 had moved within the previous five years, most relocating to a new residence within the same county (He & Schachter, 2003). Moreover, as a result of the aging of the ‘‘baby boom’’ generation, the absolute number of older movers should increase substantially over coming decades, assuming that the rate of spatial mobility holds relatively steady, as it has with respect to interstate moves since 1960 (Longino, 2006). Second, later-life in-migration may be an important driver of structural change in receiving communities. For example, elder in-migrants appear to stimulate local economies, in part by purchasing local goods and services with earnings generated outside the region (e.g., Reeder, 1998; Sastry, 1992). At the same time, economic and social impacts depend on in-migrant characteristics (Stallmann, Deller, & Shields, 1999) and are not expected to be uniformly positive (e.g., McHugh, Gober, & Borough, 2002; Reeder; Rowles & Watkins, 1993). Why do older people move? Migration decision making is commonly portrayed as a dynamic multistage process involving three moments: (a) an initial The Gerontologist desire to move, (b) the formation of migration intentions, and (c) mobility behavior (Rossi, 1955). Despite the fact that later-life migration decision making has occupied a central place in the research literature (e.g., Cuba, 1991; Haas & Serow, 1993; Wiseman, 1980), the critical step between migration intentions and mobility behavior has received only limited empirical attention. As a result, scholars know little about the actuation of migration intentions among older adults. This is an important gap in the literature, and not merely for academic reasons. Developers as well as state and local planners have a vested interest in identifying potential movers. Migration intentions manifest a complex decision process, but they are both relatively easy to measure and the most proximate predictor of voluntary migration behavior (DeJong, 1999, 2000). This would suggest the utility of concentrating marketing, retention, and recruitment efforts on seniors that anticipate moving. Unfortunately, existing literature provides little guidance on interpreting the migration intentions of seniors. In the present study we focus on the ‘‘young-old’’ (see Neugarten, 1974), and we ask two questions. First, how often are intended moves actuated? Second, what factors condition the likelihood that intended moves will be actuated? We examine these questions by using data from the Health and Retirement Study, which is based on a nationally representative sample of households containing at least one person born between 1931 and 1941, that is, aged 53 to 63 years at study baseline in 1994. Event-history techniques are used to examine the link between measured mobility intentions in 1994 and migration outcomes over the study period ending in 2002. Background Drawing from the general migration literature (Brown & Moore, 1970; Rossi, 1955; Yee and Van Arsdol, 1977), Wiseman (1980) posits a range of ‘‘triggering mechanisms’’ (e.g., retirement, death of a spouse) that prompt seniors to consider moving. Of critical importance, the character of these triggering mechanisms and the subsequent migration decision varies substantially. For this reason later-life migration is characterized by heterogeneity. The variety of later-life movers is typically categorized along two dimensions: distance and motivation (see Wiseman & Roseman, 1979). As to distance, local moves are especially likely to reflect dissatisfaction with housing or neighborhood conditions, whereas long-distance moves are additionally influenced by remote pull factors, such as lower cost of living. As to motivation, life-course transitions are thought to generate three broad types of older movers (see Litwak & Longino, 1987; Walters, 2000; Wiseman & Roseman). Amenity-oriented movers, nearing retirement or newly retired, relocate to improve their lifestyle: either locally to a nicer Vol. 48, No. 2, 2008 191 neighborhood or nonlocally to a community offering an agreeable climate, scenic environs, and an array of recreational and cultural opportunities. Assistance movers relocate to be closer to younger family members in response to negative life circumstances (e.g., the death of a spouse, declining health, low income). Finally, for other movers, severe disability may trigger a move into an institutional setting where constant care can be provided. The Actuation of Mobility Intentions Review of Research.—What does the existing literature tell us about the actuation of mobility intentions among older adults? Ferraro (1981) examines the fulfillment of relocation desires, though this is a qualitatively different question than the one posed here. Only a handful of studies employ the sort of longitudinal design necessary to link migration intentions at baseline and subsequent mobility (Colsher & Wallace, 1990; Goldscheider, 1966; Lawton, Kleban, & Carlson, 1973; Robison & Moen, 2000; Teaford, 1992; Yee & Van Arsdol, 1977). Although it generally suggests that a minority of intended movers actually move, the relevant literature is limited in several respects. First, all of the studies in question are based on data from regional or community samples, most of which are quite small. Both Goldscheider (1966) and Yee and Van Arsdol (1977) rely on data from a probability sample of households in the Los Angeles area that included 386 respondents aged 50 or older. In two rural Iowa counties, Colsher and Wallace (1990) surveyed 3097 residents aged 65 or older. Teaford (1992) followed 277 widows she recruited in Franklin County, Ohio, whereas Lawton and colleagues (1973) studied 115 older residents in one of Philadelphia’s inner-city neighborhoods. Robison and Moen (2000) employed two waves of data from the Cornell Retirement and Well-Being Survey of 762 retirees and workers aged 50 to 72 randomly chosen from six selected employer organizations in upstate New York. Consequently, the generalizability of reported findings from existing studies cannot be taken for granted. Second, nearly all studies linking migration intentions and mobility behavior among older adults conduct follow-up interviews after only 1 or 2 years. In those instances in which migration intentions at baseline are expressed in terms of an unbounded future (Robison & Moen, 2000), findings are difficult to interpret. However, even when migration intentions are specified over some time period (e.g., ‘‘How likely are you to move within the next 2 years?’’), ending observation after that time period has elapsed ignores the possibility that some elders may delay expected moves. Relatively short follow-up periods may underestimate the extent to which migration intentions are actuated. Third, available studies that examine the actuation of migration intentions among seniors do not adequately capture variation in strength of mobility expectations; it is typically operationalized as a dichotomous variable identifying those planning a move (Colsher & Wallace, 1990; Goldscheider, 1966; Lawton et al., 1973; Yee & Van Arsdol, 1977). However, the formation of migration intentions is a dynamic process, as a part of which individuals may entertain ‘‘remote thoughts’’ of relocating for many years prior to actually moving (e.g., Haas & Serow, 1993). Treating mobility expectations as dichotomous ignores this process, aggregating those with modest to weak expectations of moving together with those not actually considering a move. Fourth, the relevant literature largely overlooks the heterogeneity of older movers, described by early scholarship; with one exception (Colsher & Wallace, 1990), existing studies examine change of residence irrespective of the type of move completed. This is an important shortcoming, because the process of migration decision making appears to operate differently for different types of moves (Wiseman & Roseman, 1979). Regarding the decision to move, for example, personal resources such as health and income may facilitate an amenity-oriented move but reduce the likelihood of an assistance move (see Meyer & Speare, 1985). The problem here is that where characteristics predicting the actuation of intentions are examined, results are difficult to interpret. Goldscheider (1966), for example, shows an inverse relationship between socioeconomic status and the likelihood of successfully anticipating a move. However, these findings may be driven by a greater preponderance of local or within-area moves motivated by neighborhood or housing dissatisfaction, thus obscuring the processes governing the actuation of out-of-area moves. Failure to distinguish between mover types may be less problematic in those studies that focus on population subgroups especially likely to complete family- or assistanceoriented moves (i.e., Lawton et al., 1973; Teaford, 1992), but there is no information provided regarding the actual character of the move completed. Fifth and finally, existing studies regarding the actuation of mobility intentions among older adults give little attention to conceptualizing the link between migration intentions and migration outcomes. Empirical analyses do not appear to be guided by conceptual models that specifically address the capacity of older adults to translate mobility expectations into actual moves. Limited theoretical development makes it difficult to generate reasonable expectations and interpret results. The present study offers several contributions. First, our analysis employs a nationally representative panel of households. Second, we link migration intentions at baseline to migration outcomes over an 8-year study period. Third, we employ measures that capture substantial variation in the strength of 192 mobility intentions. Fourth, our analysis accounts for heterogeneity among older movers, constraining analysis to a young-old panel and recognizing three types of movers. Fifth, our analysis is guided by a specific conceptual framework regarding factors that may condition the likelihood that households will actuate mobility intentions. Conceptual Framework.—Among intended movers, are there characteristics that might predict the likelihood of actually moving? Migration intentions reflect expectations that a move will facilitate the attainment of one or more valued goals. According to value-expectancy theory, the greater the value placed on a particular goal or set of goals and the higher the expectancy that moving to a particular location will lead to attaining that goal or set of goals, the stronger the intention to move (DeJong, 1999, 2000; DeJong & Fawcett, 1981). All other factors being equal, at any given point in time those with stronger mobility intentions should be more likely than others to actually move. At the same time, individuals form migration intentions based on imperfect knowledge as to future conditions, as is obvious. Value-expectancy theory argues that the actuation of mobility intentions depends, in part, on ‘‘unanticipated constraints and facilitators’’ (DeJong & Fawcett, 1981). Intended movers, for example, may underestimate the costs of moving, both financial and otherwise. To the extent that expected moves are voluntary and desired, the likelihood of actually moving should depend on exposure to unanticipated constraints and the capacity to overcome unanticipated constraints. Consistent with value-expectancy theory (DeJong, 1999), financial resources may enhance a household’s capacity to overcome unanticipated constraints. Among those individuals considering a move, wealth and income should be positively linked with the likelihood of actually moving. Moreover, intended movers in relatively good health may be better prepared to perform the tasks necessary to complete a move and less likely to experience the sort of reversals that would constrain them from actually moving (see Wiseman, 1980). At the same time, the difficulties associated with moving, and consequently the importance of financial resources and health, should depend on the character of the move contemplated. The effective cost of a move is partly a function of distance between origin and destination (e.g., Lee, 1966; Speare, Goldstein, & Frey, 1975). Irrespective of distance, friends and relatives in the receiving area often sponsor migrants, thereby allowing persons with modest financial resources and limited human capital to make long-distance moves (e.g., Massey, Alarcon, Durand, & Gonzalez, 1987; Massey et al., ´ ´ 1993). Financial resources and health may be particularly salient to nonlocal moves completed without the support of family at destination. This The Gerontologist expectation makes sense because local movers and movers with family at destination, irrespective of distance moved, are likely to have better information concerning the destination and thus should be less likely to encounter unanticipated constraints. Moreover, local movers and family-supported movers should be well positioned to rely on social resources at destination in order to overcome unanticipated constraints that do arise. Methods Data The present study exploits data from the Health and Retirement Study (HRS). Supported in large part by the National Institute on Aging and conducted by the University of Michigan, the HRS is an ongoing panel study designed to follow respondents and their spouses through the transition to retirement (see Juster and Suzman, 1995). Our data extract is partly derived from the RAND HRS data file, a cleaned user-friendly version of the HRS produced with funding from the National Institute on Aging and the Social Security Administration (RAND, 2004). The core sample for the HRS is based on a stratified multistage area probability design, similar to that employed in several federal government-sponsored data-collection efforts (e.g., the Health Interview Survey and the Current Population Survey). It is representative of U.S. households containing at least one person born between 1931 and 1941. At study baseline in 1994, age-eligible panel members were 53 to 63 years old. Since the first wave of data collection in 1992, about one third of the HRS sample members had been lost to attrition by 2002. However, the analysis by Cao and Hill (2005) of the HRS data suggest that (a) attrition across waves has not undermined the representativeness of the sample and (b) panel members lost for reasons other than mortality appear to be quite similar to those remaining in the study. Migration decision making is properly conceived as a household-level process (Brown & Moore, 1970; DeJong & Fawcett, 1981; Rossi, 1955; Stark & Bloom, 1985). Moreover, decision making in couple households is likely to be qualitatively different from decision making for single individuals. In a couple household, the anticipated costs and benefits of a move should depend on the characteristics of both household members (e.g., a wife’s retirement may be less likely to trigger a long-distance move when the husband is still working). Further, reported mobility expectations for members of a couple are likely to inform one another. Within couples, reported mobility intentions for any individual are likely to include an assessment of the spouse or partner’s desire or willingness to move. Individual-level analyses, in this context, would be problematic because correlated residuals within households Vol. 48, No. 2, 2008 193 generate downwardly biased estimates of the standard errors (e.g., Berry, 1993). In light of these considerations, in the analysis to follow we treat households as the unit of analysis and disaggregate couple households (composed of two married or partnered age-eligible persons or an age-eligible person and his or her age-ineligible spouse) and single households (composed of a single age-eligible respondent). We apply household weights in order to account for oversampling among African Americans, Hispanics, and Florida residents, effectively excluding a trivial number of households with nonpositive weights (i.e., 1 single household and 20 couple households). The couple household analysis is limited to heterosexual couples in which both members (a) considered themselves married or partnered and (b) were interviewed in 1994 (n = 4,302). In the analysis and discussion to follow, the term husband refers to the male member of a married or partnered couple and wife refers to the female member of a married or partnered couple. Respondents in single households are included, if they responded at the initial wave of data collection in 1992 and provided a full interview in 1994 (n = 2,144). Analytical Strategy We examine the actuation of mobility intentions by using Cox’s proportional hazards models. Analysis estimates the hazard or risk that a move will occur within a given month, conditional on the household being ‘‘at risk’’ of experiencing a move at the beginning of the month. Households are considered at risk of moving from the baseline interview in 1994 until moving or being censored. Households are censored upon dissolving as a result of death or divorce, or attrition, or otherwise at the end of the study period in 2002. We chose 1994 as the baseline because limited information is available to characterize moves made between 1992 and 1994. A simple proportional hazards model with no time-dependent covariates takes the following form: hi ðtÞ ¼ k0 ðtÞ 3 expðb1 x1i þ b2 x2i þ . . . þ bk xki Þ: ð1Þ For our purposes, let hi(t) represent the hazard of mobility for household i at month t, given that the individual is still in the risk set at the beginning of month t. Proportional hazards models assume that the baseline function of time, k0(t), is a constant and maximize a partial likelihood function in order to estimate the influence of covariates. Model coefficients, b1, . . ., bk, indicate the proportionate adjustment to the baseline hazard rate that is implied by changes in selected independent variables, x1, . . . , xk. Exponentiated coefficients, termed hazard ratios, are comparable with odds ratios. Thus, with respect to ordinal or interval measures, 100 3 (exp bi À 1) provides the anticipated percentage change in the estimated hazard of some event for each unit change in a given covariate (Allison, 1995). We conduct the analyses by using PROC PHREG within the SAS program. Following a procedure outlined by Boudreau and Lawless (2006), we adjust estimated standard errors to account for the complex multistage sampling process used to select HRS panel members. Theory and previous research suggest heterogeneity among later-life movers. We partly constrain heterogeneity by focusing on the young-old, who are likely to be transitioning into retirement during the study period. This constraint minimizes the danger of obscuring relationships by aggregating households where the migration decision-making process is fundamentally dissimilar. As compared with the young-old, persons aged 70 or older may be especially likely to be pushed into an assistance move by some combination of widowhood, disability, and financial constraints (see Choi, 1996; Litwak & Longino, 1987; Meyer & Speare, 1985; Walters, 2000). Under these circumstances, migration decision making may be shared with or transferred to family members at alternative destinations (Wiseman & Roseman, 1979) and depend on the capacity and willingness of others to provide assistance and support (e.g., Miller, Longino, Anderson, James, & Worley, 1999). In addition, our analyses recognize that households are at risk for different types of moves. Specifically, we distinguish between local moves, nonlocal moves, and family-oriented moves, as described in the following paragraphs. Following Allison (1995), we estimate a series of separate models that focus on each type of move in turn, treating alternative event types as a form of censoring. Measures Dependent Variable.—Spatial mobility is measured by a series of questions, available at each wave from 1996 to 2002. A move occurs when the household has changed residence since the previous wave interview. Date of move information is used to compute the number of months between the date of interview in 1994 and the first subsequent move. Our analysis identifies three types of moves. Local moves involve a change of residence in which (a) the new residence is ‘‘in or around’’ the primary address city reported at the previous wave and (b) there are no children or parents living with or within 10 miles of the household at destination. Family-oriented moves involve any change of residence, in which children or parents are living with or within 10 miles of the household at destination. A change of residence in which (a) the new residence is no longer ‘‘in or around’’ the primary address city reported at the previous wave and (b) there are no parents or children living with or within 10 miles of the household at destination is coded as a nonlocal move. 194 Family-oriented moves include both within-area moves and between-area moves in which immediate family members are resident at the destination. This approach is justified on two grounds. First, the destination for a move toward family members is constrained by the location of eligible family members (Wiseman & Roseman, 1979), so that distance moved may not discriminate distinct types of family-oriented moves. Second, models estimating the hazard of a within-area family-oriented move would likely be misspecified. Only those individuals and couples having parents or children living in the area would be at risk of making a within-area family-oriented move. However, it is not possible to properly construct the risk set, because the HRS measures parents and children within 10 miles, which will not reliably be coterminous with the boundaries of the address city or area of residence. The operationalization of mover types we employ is an important improvement over previous studies that do not account for the heterogeneity of older movers. Even so, we do not claim to fully represent the full range of variability among later-life movers. We constructed mover types that aggregate households sharing a reasonably similar underlying causal structure with respect to the actuation of mobility intentions. For this reason, family-oriented moves include those individuals and couples moving toward family and others who may be moving with family or even moving from one set of family to another. Theoretical arguments already presented suggest that overcoming unanticipated constraints and actualizing mobility intentions depends on the support of family at destination rather than whether or not the members of a household are moving closer to family. In addition, the reader should be aware of certain data-imposed limitations that constrain our ability to measure the supply of supportive social ties at destination. Nonlocal moves, for example, include movers with siblings or other extended family members in the destination area along with those who move to a community where they have no existing social ties. Along these same lines, we were unable to disaggregate movers returning to their community of origin (Longino & Serow, 1990; Wiseman & Roseman, 1979), some of whom will be motivated by assistance concerns and others by amenity considerations. Furthermore, measures available in the HRS identify those households with children or parents living within 10 miles but fail to discriminate among households with immediate family living 11 or more miles away. Some local and nonlocal movers identified here may be living close to their children but not within 10 miles. Independent Variables.—We have argued that the actuation of mobility intentions should depend on the strength of mobility intentions, health, and financial resources. The HRS includes suitable measures for each of these concepts. Expected The Gerontologist mobility at baseline is captured with the following item from the 1994 interview: ‘‘Now, using the same scale as before where ‘0’ is absolutely no chance and ‘100’ means that it is absolutely certain, please tell me what you think are the chances that you will move in the next 2 years?’’ Self-assessed health is measured by using the following item: ‘‘Next I have some questions about your health. Would you say your health is excellent, very good, good, fair, or poor?’’ Responses were coded 1 to 5, where 5 indicates excellent perceived health. Although it is not included in the present study, functional disability appears to be an important predictor of geographic mobility in older age cohorts (e.g., Miller et al., 1999). The vast majority of HRS panel members were free of functional health problems at baseline, as few reported any difficulty with either basic activities of daily living (bathing, dressing, and eating) or instrumental activities of daily living (using the telephone, managing money, or taking medications). Not surprisingly, exploratory analyses in both noncouple and couple households failed to suggest any link between functional disability measures and geographic mobility. Household wealth measures the total value of diverse assets, such as real estate, vehicles, a business or farm, retirement accounts, bonds, and other savings vehicles. Household income assesses total earnings from a variety of sources, including salary or wages, unemployment compensation, retirement pensions and annuities, and social security income. Data on household economic resources were collected from a designated financial respondent within each household. We exploit RAND-constructed measures of household wealth and household income. For both measures, RAND used regression modeling to impute values where the original HRS data were incomplete (RAND, 2004). Given that wealth and income measures are typically skewed, we employ the natural log of both household wealth and household income in our models. In those cases in which reported household wealth or household income was zero, we coded the corresponding logged variable as zero. Where household income or household wealth values were negative, we multiplied by À1 in order to reverse the sign, took the natural log, and then multiplied by À1 again. Our analysis includes a range of additional control variables suggested by value-expectancy theory (e.g., DeJong, 1999) and previous migration research. We include the following demographic measures: race and ethnicity, age, gender (in noncouple households), marital status, and education. Accounting for demographic characteristic is important, in part because awareness of available destination opportunities and access to alternative destinations may vary across population subgroups. Education, for example, may expand an individual’s ‘‘awareness space,’’ thereby enhancing the likelihood Vol. 48, No. 2, 2008 195 of successfully finding a suitable destination (e.g., Speare et al., 1975). Moreover, single women and minorities may be especially likely to face housing discrimination as an unanticipated constraint, thus blocking efforts to actualize mobility intentions (e.g., Crowder, 2001). Consistent with previous later-life migration research (e.g., DeJong, Wilmoth, Angel, & Cornwell, 1995; Miller et al., 1999; Speare, Avery, & Lawton, 1991), our research also includes statistical controls accounting for the influence of person-ties and placeties at origin. As moorings hold boats in the harbor, so person-ties and place-ties at origin (i.e., familiarity and associations with a given place generated through residential experience) are expected to decrease the likelihood of a later-life move (see Longino, Perzynski, & Stoller, 2002 for a review). Specifically, we measure familiarity with neighbors, home ownership, duration of residence (i.e., whether household members are native to the community of origin or not), labor force participation, and family propinquity. Finally, previous migration experience may allow households to form migration intentions based on better information as to (a) the difficulties associated with moving and (b) the range of available destination opportunities. Consistent with valueexpectancy theory (e.g., DeJong, 1999), this would lead us to suspect that intended migrants with higher levels of previous migration experience should experience fewer unanticipated constraints. We control for veteran status because older veterans are likely to have substantial migration experience and may be especially likely to move (see Cowpers et al., 2000). Descriptive statistics and details regarding the construction of included measures for noncouple and couple households are presented in Tables 1 and 2. As noted in Table 2, among couple households we employ a number of couple-summary measures. With respect to mobility intentions, for example, we use the couple-average expected mobility value, unless only one member provided a valid response, in which case we code the single valid response for the household. For each construct where husband and wife values were available, we compared models employing two terms (e.g., highest expected mobility within couple and the difference between high and low expected mobility within couple) with models using a single couple-summary value. We use couplesummary measures for those constructs where we find no significant improvement in model fit associated with the use of two terms. These preliminary analyses lead us to exclude the item Wife Retired or Not in Labor Force from the full models presented in the following paragraphs. Although unanticipated, this result makes sense in light of the literature suggesting that, within couples, the partner with greater earning capacity will exert disproportionate influence on household decisions, Table 1. Descriptive Statistics for Single Households Variable Expected mobility Logged household wealth Logged household income Self-assessed health African American Hispanic origin Age (years) Female Formerly marrieda Newly marriedtb Education (years) Know neighborsc Homeowner Area natived Retiredte Children nearf Parents nearg Veteran M SD 23.2 7.730 9.077 3.1 0.308 0.090 58.0 0.709 0.858 0.044h 11.8 1.5 0.572 0.400 0.547h 0.590 0.194 0.156 34.134 6.172 2.293 1.246 3.212 3.296 0.876 Table 2. Descriptive Statistics for Couple Households No. Missing 44 0 0 1 0 1 0 0 0 Variable Average expected mobility Logged household wealth Logged household income Wife’s self-assessed health Husband’s self-assessed health African-American couplea Hispanic coupleb Average age (years) Partnered couple Average education (years) Know neighborsc Homeowner Area native coupled Husband retired or out of labor forcete Children nearf Parents nearg Husband veteran 0 7 19 0 0 0 0 Notes: Standard deviations are omitted for binary variables. Mean value for binary variables is the proportion of households with that attribute. a 1 = divorced, separated, or widowed; 0 = otherwise. b 1 = married by the beginning of month t; 0 = otherwise. c Measured with ‘‘Among your nearby neighbors, how many of the adults would you know by name if you met them on the street?’’ Responses coded from 0 to 3 (0 = none of them to 3 = all of them). d 1 = responded ‘‘yes’’ to this question: ‘‘Are you currently living in the same general area where you were born?’’ (0 = otherwise). e 1 = retired or not in the labor force by the beginning of month t; 0 = otherwise. f 1 = any child living in the household or within 10 miles of the household; 0 = otherwise. g 1 = any parent living in the household or within 10 miles of the household; 0 = otherwise. h value at 24th month. including the decision to make a long-distance move (see Smits, Mulder, & Hooimeijer, 2003 for a review), and the fact that across couple households, husbands report substantially higher earnings than do their wives (average husband advantage = $13,915; median husband advantage = $4,000). Findings Figures 1 and 2 graph the probability of having moved at least once, across months since the 1994 baseline interview, for noncouple and couple households respectively. Figures are based on Kaplan– Meier estimates of the survivor function, and results are broken out by mobility expectations at baseline. Not surprisingly, the probability of moving differs substantially depending on expected mobility. In those households certain of an upcoming move (expected mobility = 100), the estimated probability of having moved 2 years after the initial interview is .48 in single households and .55 for couple households. By contrast, in households with low mobility 196 M SD 17.3 11.069 10.563 3.5 3.4 0.157 0.093 57.4 0.029 12.2 1.7 0.885 0.856 27.659 3.746 1.202 1.104 1.149 0.686h 0.694 0.308 0.578 4.854 2.832 .847 No. Missing 14 0 0 0 2 0 0 0 0 28 8 8 0 0 0 22 Notes: Standard deviations are omitted for binary variables. Mean value for binary variables is the proportion of households with that attribute. a 1 = at least one member of couple African American; 0 = otherwise. b 1 = at least one member of couple of Hispanic origin; 0 = otherwise. c Measured with ‘‘Among your nearby neighbors, how many of the adults would you know by name if you met them on the street?’’ Responses coded from 0 to 3 (0 = none of them to 3 = all of them). d Sum of couple members who responded ‘‘yes’’ to this question: ‘‘Are you currently living in the same general area where you were born? e 1 = Husband retired or not in the labor force by the beginning of month t; 0 = otherwise. f 1 = any child living in the household or within 10 miles of the household; 0 = otherwise. g 1 = any parent living in the household or within 10 miles of the household; 0 = otherwise. h Value at the 24th month. expectations at baseline (expected mobility = 1–50), the estimated probability of having moved in the subsequent 24 months is .12 for noncouple households and .05 for couple households. Notably, among both noncouple and couple households assessing the chances of mobility at or above 50%, the rate of change in the probability of having moved appears to flatten somewhat 2 years after the baseline interview. This makes intuitive sense given that the expected mobility item specifically asks about the chances of moving within the next 2 years. Tables 3 and 4 report results from a series of proportional hazards models for noncouple and couple households, respectively, considering a move. As we already noted, the impact of expected mobility appears to be nonconstant over time, weakening after 2 years. In order to model this dynamic, we create a time-dependent covariate after 2 years (1 = more than 24 months since initial interview; 0 = 24 or fewer months since initial interview), and we enter a multiplicative interaction term, expected mobility 3 after 2 The Gerontologist Figure 1. Noncouple households categorized by expected mobility, estimated probability of moving at least once by months since baseline interview. years, into the proportional hazards models. We represent mobility expectations by using two terms, as follows: exp½b1 expected moblity þ b2 expected moblity 3 after 2 yearsðtÞ: The effect of mobility expectations during the first 2 years is represented by the hazard ratio for expected mobility (i.e., exponentiated b1). After the first 2 years, the estimated effect of mobility expectations is the product of the hazard ratios for expected mobility and expected mobility 3 after 2 years (i.e., exponentiated b1 3 exponentiated b2). The calculated value is presented at the bottom of both Tables 3 and 4. The p value for expected mobility 3 after 2 years evaluates the null hypothesis that there is no difference between the slope of expected mobility on the hazard of moving after the first 2 years as compared with that during the first 24 months. In Table 3, results are presented for noncouple households considering a move at baseline (i.e., expected mobility . 0). Model 1 assesses the net effect of covariates on the hazard of moving, regardless of the type of move reported. Models 2, 3, and 4 analyze three types of moves: local moves, familyoriented moves, and nonlocal moves, respectively. Vol. 48, No. 2, 2008 197 Among noncouple households considering a move, how does the strength of mobility expectations impact mobility behavior? In Model 1, each unit increase in expected mobility is associated with a 2% increase in the hazard of making some type of move during the first 2 years. However, after the first 2 years, expected mobility is significantly weaker and no longer significant (hazard ratio = 1.002, p = .867). A similar pattern is apparent across types of moves. Expected mobility predicts all three types of moves during the first 2 years after the baseline interview, but there is no evidence of an effect in the 25th and following months. There is some apparent variation in the impact of expected mobility. Each unit increase in expected mobility is associated with an estimated 3.9% increase in the hazard of a nonlocal independent move during the first 24 months after the initial interview. This is a substantial effect, and it implies that every 20-point increase in expected mobility more than doubles the hazard of a nonlocal independent move during the first 24 months. By contrast, each unit increase in expected mobility is associated with an estimated 1.6% increase in the hazard of a family-oriented move. Each 20-point increase in expected mobility according to these results implies an approximate 37% increase in the hazard of a family-oriented move. We calculated Wald Figure 2. Couple households categorized by average expected mobility, estimated probability of moving at least once by months since baseline interview. chi-square values comparing the estimated net effect of expected mobility on family-oriented as compared with nonlocal moves, and we found that the difference is statistically significant (v2 = 3.884, p = .0489). Do personal resources increase the likelihood of moving among noncouple households considering a move? Net of covariates, neither household wealth, household income, nor self-assessed health increase the overall hazard of moving, in Model 1. However, household wealth is negatively associated with the hazard of a nonlocal move in Model 4, where results suggest that each percentage increase in household wealth is associated with an estimated 5.8% decrease in the hazard of making a nonlocal move. It is noteworthy that education in Model 1 increases the hazard of moving in single households considering a move at baseline. Education’s influence, however, appears to be limited to Model 4, where each additional year of schooling is associated with an estimated 29.4% increase in the hazard of a nonlocal independent move. An effect of this size implies, for example, that net of covariates, a college graduate with 16 years of education is about 2.8 times more likely to make a nonlocal move as a comparable single individual with a high school degree with 12 years of education. Wald chi-square tests suggest that education is a significantly stronger predictor of nonlocal independent moves as compared with both 198 family-oriented moves (v2 = 9.122, p = .003) and local moves (v2 = 4.943, p = .026). Results for couple households considering a move at baseline (i.e., average expected mobility . 0) are presented in Table 4. Findings in Model 1 suggest that, during the first 2 years after baseline, expected mobility is strongly associated with the overall hazard of moving. Moreover, the fact that expected mobility 3 after 2 years is significant indicates that the effect of expected mobility on the overall hazard of moving weakens after the first 2 years. Even so, Model 1 implies that in the 25th and subsequent months, expected mobility remains significant and positive (hazard ratio = 1.013, p , .0001). Models 2, 3, and 4 present proportional hazards regression results for alternative move types. The estimated effect of expected mobility in the first 2 years is substantial and consistent across models. By way of illustration, Model 4 results indicate that, in the initial 24 months, the expected hazard of a nonlocal independent move for couples at the 75th percentile of the distribution of expected mobility (i.e., expected mobility = 50) is approximately four times greater than that of comparable couples at the 25th percentile of the distribution (i.e., expected mobility = 10). Expected mobility effects of similar magnitudes are observed for the hazards of a local move and the hazards of a family-oriented move. The Gerontologist Table 3. Results of Proportional Hazards Models Predicting First Move for Noncouple Households Considering a Move at Baseline (expected mobility > 0), 1994 to 2002 (n = 890) Model 1 Variable Expected mobility Exp. mobility 3 after 2 years Logged household wealth Logged household income Self-assessed health Hispanic origin African American Age (years) Female Formerly married Newly marriedt Education (years) Know neighbors Homeowner Area native Retiredt Children near Parents near Veteran À2 log likelihood Likelihood ratio v2 Model-implied effect of expected mobility in the 25th and following months Model 2 Model 3 Model 4 HR p HR p HR p HR p 1.020 0.981 0.979 0.993 1.033 0.970 0.885 1.020 1.191 0.991 3.112 1.085 0.964 0.361 0.785 1.401 1.038 0.909 0.524 1,094.4 168.1 .000 .000 .121 .857 .594 .914 .625 .377 .402 .958 .000 .001 .591 .000 .159 .036 .804 .585 .013 1.030 0.972 0.978 0.965 1.100 0.691 0.875 1.023 1.111 0.547 1.923 1.038 1.308 0.235 .955 1.937 0.192 0.360 0.438 174.0 75.8 .002 .035 .434 .564 .501 .624 .798 .742 .826 .077 .518 .641 .095 .005 .884 .097 .002 .111 .124 1.016 0.983 0.987 1.011 1.025 1.039 0.934 1.011 1.259 1.848 3.293 1.049 0.847 0.315 0.703 1.288 2.034 1.105 0.570 626.0 131.2 .001 .035 .444 .801 .729 .928 .810 .692 .379 .048 .000 .170 .101 .000 .171 .256 .001 .665 .069 .000 .005 .022 .724 .225 .408 .673 .370 .786 .633 .001 .000 .441 .295 .688 .491 .020 .082 .354 ,.0001 1.039 0.969 0.942 0.977 1.180 0.581 0.807 1.055 1.127 0.790 5.217 1.294 1.118 0.644 0.868 1.273 0.464 0.480 0.607 200.9 54.6 0.999 0.8119 1.007 1.001 ,.0001 0.867 1.001 ,.0001 0.93 ,.0001 0.259 Note: Model 1, 2, 3, and 4 = any type of move, local move, family-oriented move, and nonlocal move, respectively. HR = hazard ratio. After the first 2 years, the net effect of expected mobility weakens significantly with respect to family-oriented and nonlocal moves, but not local moves. Nevertheless, across types of move, expected mobility retains positive significance even 2 years after the initial interview. Thus, in couple households there is evidence suggesting a residual effect of expected mobility beyond the 2-year period specified by the question item. Do resources enhance the likelihood of mobility among couples considering a move at baseline? Household wealth is significantly associated only with the hazard of a nonlocal independent move in Model 4. Here each percentage change in household wealth is associated with a 13.1% increase in the hazard of a nonlocal independent move. Comparing couple households at the 75th and 25th percentiles of household wealth, Model 4 results imply that net of covariates those reporting a net worth of $332,500 are approximately 26% more likely to make a nonlocal independent move as compared with those reporting $52,900 in household wealth. According to Wald chi-square tests, the effect of household wealth is significantly more positive with respect to nonlocal as compared with family-oriented moves (v2 = 5.611, p = .018), but not significantly different between nonlocal and local independent moves (v2 = 2.854, p = .091). Similarly, wife’s self-assessed health predicts only nonlocal moves. Each unit change in wife’s subVol. 48, No. 2, 2008 199 jective health is associated with a 29.3% increase in the hazard of a nonlocal move according to Model 4. According to these results, among couples considering a move, those in which the wife reports excellent health are approximately 2.8 times more likely to make a nonlocal move as compared with couples in which the wife assesses her health as poor, net of expected mobility and other covariates. Finally, the slope coefficient for wife’s self-assessed health is significantly different with respect to nonlocal independent moves as compared with both familyoriented (v2 = 5.881, p = .015) and local (v2 = 4.315; p = .038) independent moves, according to Wald chi-square tests. We estimated a number of ancillary models in order to evaluate the possibility that, in noncouple households, migration decision making operates differently for men and women. Findings suggest that among those considering a move at baseline, divorced men may be significantly less likely than divorced women to report a local move. Analyses failed to support the notion that the proximity of children and parents affects single men and women in different ways. Discussion and Conclusion This study examines the actuation of migration intentions among the young-old. Longitudinal studies of older adults that link mobility intentions to future mobility are few in number and characterized by Table 4. Results of Proportional Hazards Models Predicting First Move for Couple Households Considering a Move at Baseline (i.e., avg. expected mobility > 0), 1994 to 2002 (n = 2,010) Model 1 Variable Average expected mobility Avg. exp. mobility 3 after 2 years Household wealth Household income Wife’s self-assessed health Husband’s self-assessed health Hispanic couple African-American couple Average age (years) Partnered couple Average education (years) Know neighbors Homeowner Area native couple Husband retired or out of labor forcet Children near Parents near Husband veteran À2 log likelihood Likelihood ratio v2 Model-implied effect of expected mobility in the 25th and following months Model 2 Model 3 Model 4 HR p HR p HR p HR p 1.033 0.981 0.991 1.055 1.057 1.035 0.495 0.807 0.982 1.275 0.986 0.977 0.505 1.018 1.123 0.834 1.065 0.911 2849.4 380.3 .000 .000 .544 .448 .255 .474 .005 .257 .149 .383 .542 .694 .000 .793 .345 .095 .570 .378 .000 .172 .652 .949 .757 .215 .658 .901 .585 .278 .914 .827 .001 .035 .414 .003 .016 .703 ,.0001 1.035 0.981 1.135 0.927 1.293 1.020 0.081 0.674 1.009 1.355 0.938 0.927 0.792 0.794 0.937 0.504 0.670 1.393 723.7 147.9 .000 .004 .036 .592 .003 .834 .007 .358 .696 .660 .173 .536 .412 .096 .799 .001 .083 .131 ,.0001 1.035 0.978 0.977 1.103 0.984 1.003 0.691 0.898 0.966 1.058 0.996 0.999 0.441 1.042 1.164 1.323 1.477 0.816 1534.1 244.0 .000 .000 .189 .317 .823 .966 .172 .628 .043 .880 .913 .987 .000 .618 .327 .043 .006 .103 ,.0001 1.029 0.988 1.013 1.014 0.967 1.136 0.745 0.944 0.984 1.726 1.006 0.970 0.334 1.352 1.330 0.477 0.483 0.916 475.6 104.4 ,.0001 1.014 ,.0001 1.017 .0073 1.011 ,.0001 1.016 .0001 Note: Model 1, 2, 3, and 4 = any type of move, local move, family-oriented move, and nonlocal move, respectively. HR = hazard ratio. several limitations. We seek to improve on these studies in the following ways. First, we use panel data from a large nationally representative sample. Second, our study follows panel-member households over an 8-year study period. Third, we measure mobility intentions with greater precision than that offered by discrete and ordinal items used in previous research. Fourth, our study accounts for heterogeneity among older movers by restricting the analysis to young-old households and by modeling three types of moves (i.e., local, family oriented, and nonlocal). Findings reported here address two questions: How often are intended moves actuated? What factors condition the likelihood that intended moves will be actuated? With respect to the first question, among those households at baseline certain of moving within 2 years, the estimated probability of actually completing a move during that interval is .48 for single households and .55 for couple households. Lower rates of mobility were observed as migration intentions weakened. Consistent with existing research, the link between expected mobility and actual mobility is strong but imperfect; this is not surprising, given that migration intentions shift in response to changing conditions and perceptions (DeJong, 1999). What factors shape the chances of actually moving for those individuals and couples considering a move at baseline? First, our findings suggest that the strength of mobility expectations is of substantial importance, with one exception. In single house200 holds, expected mobility is significantly weaker as a predictor of family-oriented moves as compared with nonlocal moves. This result is not entirely counterintuitive. It may be that the actuation of family-oriented migration intentions among single elders is especially dependent on support from family at destination, support that may be volatile and difficult to judge in advance as compared with the kinds of conditions facilitating other types of moves (e.g., warm climate, nicer neighborhood). Second, we argued that among those persons and couples considering a move, economic resources and health should facilitate the actuation of migration intentions, especially with respect to nonlocal moves. Contrary to expectations, for single households our analysis suggests that, net of expected mobility and other covariates, increasing household wealth diminishes the likelihood of a nonlocal move among those individuals considering a move at baseline. The most obvious explanation is that for many young-old singles, expected moves are not necessarily desired moves and wealthier individuals are better prepared to avoid moving. This interpretation makes sense to the extent that unmarried elders appear to be especially likely to move toward kin (Cuba, 1992; DeJong et al., 1995; Litwak & Longino, 1987), motivated in part by assistance needs, whether perceived or anticipated (Wiseman, 1980; Meyer & Speare, 1985). It may be that, for many single elders, anticipated assistance needs serve as a motivation to move toward kin, but those with greater wealth are The Gerontologist better able to delay or avoid actually moving. More specifically, wealthier singles may be better equipped to purchase or otherwise secure access to formal services. Findings with respect to couple households considering a move at baseline are broadly consistent with expectations and suggest that household wealth and wife’s self-assessed health increase the likelihood of a nonlocal move, net of average expected mobility and other covariates. Moreover, the effects of both household wealth and wife’s self-assessed health are significantly more positive with respect to nonlocal as compared with family-oriented moves. Our results confirm the utility of migration intentions as a predictor of future mobility, at least among young-old households. At the same time, findings suggest important differences in the role of economic resources and health between types of households and across different types of moves. Of course, there are several limitations that readers should keep in mind. Our analysis and interpretation assumes that the actual moves across the study period are of the same character as the moves under consideration at baseline. This is not an unreasonable assumption, especially for moves completed soon after the baseline interview, given evidence to suggest that elders may contemplate moves many years in advance (e.g., Gober & Zonn, 1983; Haas & Serow, 1993). Even so, future research might generate additional insight by exploiting more specific measures of migration intentions than we employ here (McHugh, 1984). There are a range of additional questions raised by our results which will have to await future research. Specifically, our findings apply to youngold individuals and couples, but how might migration decision making may play out differently in age cohorts who are older? Future research might profitably exploit data from the Asset and Health Dynamics panel, which targets persons born before 1923 and includes measures of migration intentions and outcomes. Moreover, there may be substantial heterogeneity among single elders with respect to the nature and quality of family relationships and the kinds of needs or demands that may motivate a move. Nevermarried elders, for example, may be especially likely to move toward nieces, nephews, and siblings in light of perceived or anticipated assistance needs. Moreover, divorce may have implications for ties to one’s children, thus altering the migration calculus. In addition, it is not immediately clear why only the wife’s or female partner’s subjective health directly impacts the hazard of making a nonlocal move, though this finding may not be entirely counterintuitive. Executing a long-distance move without family support at destination obviously requires a substantial amount of planning and work. Within couples it may be that a disproportionate share of the planning and work of moving is shouldered by Vol. 48, No. 2, 2008 201 women, which, if true, would help us to understand the singular importance of the wife’s self-assessed health. The search for a new residence, for example, may typically fall to wives and female partners, and healthier women may be more likely to have sufficient energy to successfully execute a search. Future research should explore the way in which the work of moving is typically divided in couple households. Future studies should also pay attention to other kinds of intracouple dynamics that may condition migration decision making and mobility behavior. Couples that enjoy spending time together, for example, may be more likely to make a long-distance move involving the disruption of existing social network ties as compared with couples sharing a less supportive less intimate relationship. References Allison, P. D. (1995). Survival analysis using the SAS system: A practical guide. Cary, NC: SAS. Berry, E. G. (1993). Understanding regression assumptions. London: Sage. Boudreau, C., & Lawless, J. (2006). Survival analysis based on the proportional hazards model and survey data. The Canadian Journal of Statistics, 34, 1–14. Brown, L. A., & Moore, E. G. (1970). The intra-urban migration process: A perspective. Geografiska Annaler: Human Geography, 52, 1–13. Cao, H., & Hill, D. (2005). Active versus passive sample attrition: The Health and Retirement Study. Econometrics Series: Economics Working Paper Archive. Retrieved January 10, 2007, from http://ideas.repec.org/p/wpa/ wuwpem/0505006.html Choi, N. G. (1996). Older people who move: Reasons and health consequences. Journal of Applied Gerontology, 15, 325–344. Colsher, P. L., & Wallace R. B. (1990). Health and other social antecedents of relocation in rural elderly persons. Journal of Gerontology: Social Sciences, 45, S32–S38. Cowpers, D. C., Longino, C. F., Jr., Kubal, J. D., Manheim, L. M., Dienstfrey, S. J., & Palmer, J. M. (2000). The retirement migration of U.S. veterans, 1960, 1970, 1980, and 1990. The Journal of Applied Gerontology, 19, 123–137. Crowder, K. D. (2001). Racial stratification in the actuation of mobility expectations: Microlevel impacts of racially restrictive housing markets. Social Forces, 79, 1377–1396. Cuba, L. (1991). Models of migration decision making reexamined: The destination search of older migrants to Cape Cod. The Gerontologist, 31, 204–209. Cuba, L. (1992). Family and retirement in the context of elderly migration. In M. Szinovacz, D. J. Ekerdt, & B. H. Vinick (Eds.), Families and retirement (pp. 205–221). Newbury Park, CA: Sage. DeJong, G. F. (1999). Choice processes in migration behavior. In K. Pandit & S. D. Withers (Eds.), Migration and restructuring in the United States: A geographic perspective (pp. 273–293). New York: Rowan & Littlefield. DeJong, G. F. (2000). Expectations, gender, and norms in migration decisionmaking. Population Studies, 54, 307–319. DeJong, G. F., & Fawcett, J. T. (1981). Motivations for migration: An assessment and a value-expectancy research model. In G. F. DeJong & R. W. Gardner (Eds.), Migration decision making: Multidisciplinary approaches to microlevel studies in developing and developed countries (pp. 13–58). New York: Pergamon Press. DeJong, G. F., Wilmoth, J. M., Angel, J. L., & Cornwell, G. T. (1995). Motives and the geographic mobility of very old Americans. Journal of Gerontology: Social Sciences, 50B, S395–S404. Ferraro, K. F. (1981). Relocation desires and outcomes among the elderly. Research on Aging, 3, 166–181. Gober, P., & Zonn, L. E. (1983). Kin and elderly amenity migration. The Gerontologist, 23, 288–294. Goldscheider, C. (1966). Differential residential mobility of the older population. Gerontology, 21, 103–108. Haas, W. H., III, & Serow, W. J. (1993). Amenity retirement migration process: A model and preliminary evidence. The Gerontologist, 33, 212–220. He, W., & Schacter, J. P. (2003). Internal migration of the older population: 1995 to 2000. Washington, DC: U.S. Census Bureau. Juster, F. T., & Suzman, R. (1995). An overview of the Health and Retirement Study. Journal of Human Resources, 30, S7–S56. Lawton, M. P., Kleban, M. H., & Carlson, D. A. (1973). The inner-city resident: To move or not to move. The Gerontologist, 13, 443–448. Lee, E. S. (1966). A theory of migration. Demography, 3, 147–157. Litwak, E., & Longino, C. F., Jr. (1987). Migration patterns among the elderly: A developmental perspective. The Gerontologist, 27, 266–272. Longino, C. F., Jr. (2006). Retirement migration in America (2nd ed., R. J. Manheimer, Ed.). Houston: Vacation. Longino, C. F., Jr., & Bradley, D. E. (2006). Internal and international migration. In R. H. Binstock & L. K. George (Eds.), Handbook of aging and the social sciences (6th ed., pp. 76–93). San Diego, CA: Elsevier. Longino, C. F., Perzynski, A. T., & Stoller, E. P. (2002). Pandora’s briefcase: Unpacking the retirement migration decision. Research on Aging, 24, 29–49. Longino, C. F., Jr., & Serow, W. J. (1990). Regional differences in the characteristics of elderly return migrants. Journal of Gerontology: Social Sciences, 47, S38–S43. Massey, D. S., Alarcon, R., Durand, J., & Gonzalez, H. (1987). Return to ´ ´ Aztlan: The social processes of international migration from Western Mexico. Los Angeles: University of California Press. Massey, D. S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., & Taylor, J. E. (1993). Theories of international migration: A review and appraisal. Population and Development Review, 19, 431–466. McHugh, K. (1984). Explaining migration intentions and destination selection. Professional Geographer, 36, 315–325. McHugh, K., Gober, P., & Borough, D. (2002). The Sun City wars: Chapter 3. Urban Geography, 23, 627–648. Meyer, J. W., & Speare, A., Jr. (1985). Distinctively elderly mobility: Types and determinants. Economic Geography, 61, 79–88. Miller, M. E., Longino, C. F., Jr., Anderson, R. T., James, M. K., & Worley, A. S. (1999). Functional status, assistance, and the risk of a communitybased move. The Gerontologist, 39, 187–200. Neugarten, B. L. (1974). Age groups in American society and the rise of the young-old. Annals of the American Academy of Political and Social Science, 415, 187–198. RAND. (2004). RAND HRS data documentation, version D. Santa Monica, CA: RAND Center for the Study of Aging. Reeder, R. J. (1998). Retiree-attraction policies for rural development. Washington, DC: Economic Research Service, U.S. Department of Agriculture. 202 Robison, J. T., & Moen, P. (2000). A life-course perspective on housing expectations and shifts in late midlife. Research on Aging, 22, 499–532. Rossi, P. H. (1955). Why families move: A study in the social psychology of urban residential mobility. Glencoe, IL: The Free Press. Rowles, G. D., & Watkins, J. F. (1993). Elderly migration and development in small communities. Growth and Change, 24, 509–538. Sastry, M. L. (1992). Estimating the economic impacts of elderly migration: An input–output analysis. Growth and Change, 23, 54–79. Smits, J., Mulder, C. H., & Hooimeijer, P. (2003). Changing gender roles, shifting power balance and long-distance migration of couples. Urban Studies, 40, 603–613. Speare, A., Jr., Avery, R., & Lawton, L. (1991). Disability, residential mobility, and changes in living arrangements. Journal of Gerontology: Social Sciences, 46, S133–S142. Speare, A., Jr., Goldstein, S., & Frey, W. H. (1975). Residential mobility, migration, and metropolitan change. Cambridge, MA: Ballinger. Stallmann, J. I., Deller, D. C., & Shields, M. (1999). The economic and fiscal impact of aging retirees on a small rural region. The Gerontologist, 39, 599–610. Stark, O., & Bloom, D. E. (1985). The new economics of labor migration. The American Economic Review, 75, 173–178. Teaford, M. H. (1992). Predictors of older widows’ intentions to move and actual relocation. Unpublished doctoral dissertation, The Ohio State University. Walters, W. H. (2000). Types and patterns of later-life migration. Geografiska Annaler, 82B, 129–147. Walters, W. H. (2002). Later-life migration in the United States: A review of recent research. Journal of Planning Literature, 17, 38–66. Wiseman, R. F. (1980). Why older people move: Theoretical issues. Research on Aging, 2, 141–154. Wiseman, R. F., & Roseman, C. C. (1979). A typology of elderly migration based on the decision making process. Economic Geography, 55, 324– 337. Yee, W., & Van Arsdol, M. D. (1977). Residential mobility, age, and the life cycle. Gerontology, 32, 211–221. Received January 16, 2007 Accepted July 11, 2007 Decision Editor: William J. McAuley, PhD The Gerontologist