Moving Considerations

Abstract

Residential proximity is an important indicator of family members’ availability to provide assistance to each other. We investigate proximity-enhancing moves by older parents and their children and the reasons for such moves. Using the 2000–2004 waves of the Health and Retirement Study, we fit multinomial logit models examining 2-year residential transitions for parents and children living 10 or more miles apart at baseline. Our results show that family members collectively adjust intergenerational proximity to facilitate mutual support. Despite the common assumption that older parents move closer to their children to receive assistance, more than two thirds of all proximity-enhancing moves are made by adult children. While greater anticipated longevity raises the probability that older parents will move closer to their children, parents’ anticipated longevity does not influence children’s moving decisions. Including individual variability in anticipated longevity in the life course framework helps account for relocation that precedes declines in health or increases in the need for support.

Keywords

anticipated longevity caregiving intergenerational relations life course migration residential proximity residential mobility

The majority of older Americans have been leading healthier and more independent lives, and those who need help receive it primarily from family members living with them or nearby (Bengtson, 2001; Clark & Wolf, 1992; Silverstein, 1995; Silverstein & Angelelli, 1998). Over the past 50 years intergenerational co-residence in the United States has declined (Crimmins & Ingegneri, 1990; Ruggles, 1994, 1996, 2001) and geographic proximity among family members has become increasingly relevant as an indicator of family members’ availability to provide assistance to one another (Hiedemann & Stern, 1999; Lee, 1980; Litwak & Kulis, 1987; Rossi & Rossi, 1990; Smith, 1998; Speare & Avery, 1993). This study investigates proximity-enhancing moves by older parents as well as their children and the reasons for such moves. We distinguish proximity-enhancing moves made by parents from those made by children and show that both play key roles in shaping intergenerational proximity in America. Our analysis illustrates the importance of including individual variability in anticipated longevity in the life course framework as a way of accounting for relocation that precedes later-life declines in health or increases in the need for support. It further demonstrates that the factors prompting moves by parents and children differ and that family members collectively adjust their proximity to facilitate mutual support.

Background

The life course model of migration formulated by Litwak and Longino (1987), Longino (1990), and Longino et al. (1991) provides a general framework for studying intergenerational proximity and describes three categories of later-life moves. Amenity moves typically occur around retirement age when older people are in relatively good health, and often involve separation from children and other family members as retirees pursue leisure opportunities. In later years, declining health or widowhood may prompt assistance moves, as older people seek to live closer to adult children or other family members who can provide care. Finally, nursing home moves happen when informal care is no longer sufficient or practical.

Due to the focus on older adults’ life course trajectories, the wider family context has not always received full consideration in studies of residential choices (Mulder, 2007). Clark and Wolf (1992) were among the first to examine intergenerational proximity while considering the characteristics of both older parents and their children. Using the U.S. National Survey of Families and Households (NSFH), they found that proximity had a J-shaped relationship with parent’s age (as anticipated by the life course model), and that parents with fewer resources, or those whose children were younger or in need of support were more likely to live within 10 miles of a child. Notably, however, older respondents’ functional limitations were not associated with intergenerational proximity. Rogerson, Weng, and Lin (1993) likewise reported no significant assoication between older parents’ health status and intergenerational proximity in a cross-sectional analysis.

Lin and Rogerson (1995) and Rogerson, Burr, and Lin (1997) showed that while 24.8% of NSFH respondents reported moving, those living close to their children showed greater residential stability. As before, baseline functional status, age, race, social distance between generations, and income did not influence the probability of a proximity-enhancing move, but older respondents who lost a spouse or experienced an increase in functional limitations were more likely to report greater proximity to children. Chen and Wilmoth (2004) and Wilmoth (2010) also found that worsened health was associated with moves closer to family members.

Silverstein (1995) used the longitudinal study of aging (LSOA) to examine geographic convergence between parents (age 70+) and an adult child. Using travel time to define proximity, he found that a higher number of limitations in instrumental activities of daily living (IADL) at baseline did not increase subsequent intergenerational proximity. However, declines in IADLs did lead to convergence. Among those who did converge, being unmarried or becoming widowed was associated with a larger change in travel time.

Data from Italy (Tomassini, Wolfe, & Rosina, 2003) and the Netherlands (Michielin & Mulder, 2007; Michielin, Mulder, & Zorlu, 2008) have likewise shown that familial characteristics—and especially children’s need for housing assistance—is equally or more influential than parents’ support needs in shaping intergenerational proximity. While changes in intergenerational proximity may thus clearly be a function of moves by either parents or their children in response to changing circumstances, previous studies mainly focused on moves by parents toward their children and rarely considered the influence of parental location on children’s migration decisions (Michielin et al., 2008). Furthermore, much of the data on which past studies have relied could not specify whether changes in proximity were due to moves by older parents or their children.

Though longitudinal research also suggests that only a minority of intended movers actually move (Ferraro, 1981; Longino, Bradley, Stoller, & Haas, 2008), most studies investigating late-life migration have focused on completed, rather than anticipated moves, as data on the former are more readily available. Furthermore, consistent with the “assistance move” orientation, previous studies incorporate information on older parents’ health status and support needs at baseline, without considering their anticipated longevity. However, anticipated longevity may be a key factor in intended and realized moves because it reflects the time horizon associated with a person’s expectation about his or her future health status and potential support needs. While the life course migration framework has proven useful for understanding the mobility behavior of the older population, moves are also influenced by considerations related to the financial and psychological costs of relocation and expectations about future health problems or needs for support. Indeed, studies have shown that anticipated longevity influences economic planning and behavior in later-life even controlling for wealth, chronic conditions, and disability (Gan, Gong, Hurd, & McFadden, 2004; Hurd & McGarry, 1995, 2002; Hurd, Smith, & Zissimopoulos, 2002; Mirowsky & Ross, 2000; Ross & Mirowsky, 2002).

We propose an expanded version of the life course migration framework that explicitly includes an intergenerational perspective and examines the older parent’s anticipated longevity as a potential explanation for relocation that precedes declines in health or increases in the need for financial, instrumental, or other forms of support. We test this idea and explore changes in intergenerational residential proximity via analyses that treat moves by parents and their adult children separately, with the goal of understanding the factors that differentially influence each generation’s behavior.

Method

Data

We use data from three waves (2000, 2002, and 2004) of the Health and Retirement Study (HRS) to examine changes in residential proximity between respondents and their adult children. The HRS is a nationally representative panel study of older persons in the United States. Since 1992, the HRS has collected detailed information about living arrangements and family as well as measures of health, functioning, wealth, and assets (Juster & Suzman, 1995). Notably, it also includes a set of questions assessing respondents’ subjective estimation of future events, such as institutionalization or death. The unit of analysis is the older respondent, who provided information about his or her children. Data for respondents with multiple children were summarized into aggregate measures on the family level.

Our analysis is restricted to respondents who were aged 69 or older in 2000, as the majority of persons are out of the labor market by that age, and it is reasonable to assume that their decisions regarding intergenerational proximity and relocation were unlikely to be driven by job-related factors. The sample also is divided into two HRS cohorts: the “Children of the Depression” or CODA cohort (N = 3,560), born between 1924 and 1930; and the Asset and Health Dynamics Among the Oldest Old (AHEAD) cohort, born in 1923 or earlier (N = 4,554). We estimated models separately for each cohort because, as suggested in the life course model, the meaning of anticipated longevity for planning ahead may differ with age. People in their 80s may be more concerned about family support, for example, while individuals in their 70s may think more about privacy and amenities.

We excluded respondents who were living within 10 miles of an adult child at baseline, because we cannot measure proximity-enhancing moves for those who are already in the closest proximity category measured by the HRS (1,474 co-resided with children; 3,953 lived within 10 miles from children). We also excluded respondents reporting no contact with children because we assume that these respondents’ residential choices will not be influenced by their children (N = 1,653). We also excluded respondents who were institutionalized at baseline since they have already completed a major transition, and many were too cognitively impaired to complete the subjective expectation questions (N = 711). Finally, an additional 608 respondents were omitted from the analysis due to missing data on anticipated longevity, our key independent variable.

We examined two sets of 2-year transition rates (2000–2002 and 2002–2004) using consecutive waves of the HRS. Multiple transitions for the same respondent were pooled and treated as two observations. We accounted for clustering in the data using robust standard errors. The final sample of 2,861 transitions includes 1,438 transitions in the period 2000–2002 (AHEAD = 736; CODA = 702) and 1,423 transitions in the period 2002–2004 (AHEAD = 668; CODA = 755).

Outcome Variables

The primary outcome in this study is a proximity-enhancing move, defined as relocation by either a parent or a child resulting in their living closer together. Respondents in our sample lived 10 miles or more from their nearest child at baseline, and increased proximity was defined as living within 10 miles of a child at follow-up. If there are multiple adult children in the family, only the distance from the nearest children was used. Respondents were asked about their household composition and whether each non co-resident child lives within 10 miles. Ten miles was chosen as the cutoff point by the HRS because it is a distance that can be traveled in about 30 min, and beyond which frequent face-to-face contact and support become difficult. Respondents may have chosen the “10 miles” response when they felt that a child was nearby, but were not sure of the distance (Clark & Wolf 1992). We include co-residence in the “living close” category. While the decision processes leading to co-residence and close proximity are likely different, for the purpose of this study we assumed that nearby children are potential caregivers regardless of their exact residential arrangements.

In order to distinguish proximity-enhancing moves by parents from those by children, we used information on moves by the older respondents. A proximity-enhancing move by a parent was indicated when the older parent moved and ended up within 10 miles of a child. In cases where the parent ended up living within 10 miles of a child at follow-up but did not move, we assumed that one of the children moved closer to the parent. Cases where both parents and children moved and ended up within 10 miles of each other are included with parental moves. The reference group comprises those who remained 10 or more miles away from their nearest child at follow-up, regardless of whether or not they changed residences.

To minimize the potential bias of including only survivors who remain in the community over the 2-year periods, our models consider attrition as a competing risk. Attrition includes institutionalization, death, and other loss to follow up.

Explanatory and Control Variables

The key explanatory variable is older individuals’ anticipated longevity. The HRS asks respondents to assess the probability that they will live another 10–15 years (e.g., if the subject’s age is 65–69 they are asked the probability of living to age 80). Average subjective probabilities of survival have been shown to be very close to survival probabilities predicted by life tables (Hurd & McGarry, 2002).

This subjective probability of survival (SPS) requires transformation for comparison across respondents of different ages. By assuming a constant hazard of mortality across the period between respondents’ age and the target year for survival, we calculated an annual survival probability, and rescaled it into a 10-year survival probability.

Specific values of the SPS merit attention. In our sample, 16% (24% and 9% for the AHEAD and CODA cohorts, respectively) reported no chance of surviving the next 10 years, and 9% (7% and 11% for the AHEAD and CODA cohorts, respectively) reported being 100% sure of their 10-year survival in the 2000 survey. In addition, about 26% (23% and 29% of the AHEAD and CODA cohorts) reported a 50% chance of surviving the next 10–15 years. In preliminary analyses, we tested several different specifications of the variable to determine whether those with perfect certainty or with a “50/50” response differed from others, but these approaches did not change the findings qualitatively. We therefore retained these values, using a continuous distribution of SPS in the analysis.

We use baseline anticipated longevity to predict proximity-enhancing-moves in the next 2 years because the timing, and consequently the causal relationship, between change in anticipation and proximity is unclear and because anticipated longevity is used to represent the time horizon for planning rather than to capture health status. Anticipated longevity does appear to capture information beyond current health: it is negatively and modestly (less than 0.20 in all cases) correlated with the number of ADL and IADL difficulties, and the self-reported number of chronic conditions (authors’ tabulations not shown).

Variables describing older parents’ homeownership and neighborhood tenure represent the costs (economic or psychosocial) of moving to a new place. Respondents are categorized as homeowners, renters, or sharing living spaces with other individuals. Tenure is defined as the duration of residence in a neighborhood, which may reflect the respondent’s social attachment to the community.

Variables describing respondents’ health status, marital status, and possession of long-term care (LTC) insurance are used to reflect the health and socioeconomic needs that might motivate proximity-enhancing moves. Gradients in health are captured by the reported number of difficulties in performing ADLs and IADLs, as well as the number of chronic conditions. Marital status is included as an indicator of support, and coded as currently married, widowed, or divorced. Changes in health and marital status during the 2-year interval were also incorporated into the analysis.

Additional variables in the analysis include sociodemographic characteristics such as age in years, gender, race or ethnicity (classified as non-Hispanic White, non-Hispanic Black, Hispanic, and Asian), educational attainment (categorized as less than high school, high school graduate, some college, and college and above), nonhousing wealth, personal earnings, number of marriages, number of adult children, number of grandchildren, and having any children with annual income below $35,000.

Statistical Approach

We implemented multinomial logit models to compare the predictors of four alternative transitions outcomes: (1) no change in intergenerational proximity, (2) parent moved closer to a child, (3) child moved closer to the parent, and finally, (4) attrition from the sample. The largest category—no change in proximity—was chosen as the reference. The basic estimation equations are as follows:

Ln \frac{Pr (Y_{i} = " parent moved closer ")}{Pr (Y_{i} = " no change in proximity ")} = β_{2} \times X_{i}

Ln \frac{Pr (Y_{i} = " child moved closer ")}{Pr (Y_{i} = " no change in proximity ")} = β_{3} \times X_{i}

Ln \frac{Pr (Y_{i} = " attrition ")}{Pr (Y_{i} = " no change in proximity ")} = β_{4} \times X_{i}

where Y_i represents the outcome for older parent i; β₂–β₄ represent vectors of regression coefficients for outcomes 2–4 (respectively) as compared to the reference group; and X_i represents aforementioned independent variables. The unknown parameters in each vector β₂ through β₄ were jointly calculated via maximum likelihood estimation.

Models were estimated for the AHEAD and CODA cohorts separately. In addition to substantive concerns about differential motivations for moves between these generations, the sampling design for the HRS calls for separate analysis. The AHEAD cohort was enrolled in 1993 and was asked to provide an SPS 4 times by 2002. The CODA cohort was enrolled in 1998 and was asked the same question for the first time in 2000. Since SPS is a key variable in this analysis, we stratify these two cohorts to guard against potential bias in a pooled sample.

For each cohort, three specifications of the multimomial logit regression were separately estimated, with independent variables introduced into the model in three sequential batches. This approach allowed us to understand the explanatory implications of additional information. The first specification includes anticipated longevity and sociodemographic variables, the second specification adds predictors representing health, wealth, and family structure, and the third specification also incorporates variables representing changes in marital status, health, and functional status.

Since observations from the 2000–2002 and 2002–2004 HRS waves are pooled, many respondents contribute two observations to the sample. Furthermore, married couples contribute observations to the sample that are likely to be correlated. To account for such potential interdependence in the data, we assumed a clustered rather than identity variance–covariance matrix in order to get valid standard errors for the model coefficients.

Within the eligible study population, about 15% (19% and 10% among the AHEAD and CODA cohorts, respectively) were missing SPS values. Analyses (not shown) revealed that respondents who were older, less educated, had worse cognition, or had more IADL difficulties, were more likely to be missing a SPS value. Since vulnerable populations were less likely to be included in an analysis of complete cases, we used a two-step Heckman Selection model to control for the sample selection pattern in the multivariate analyses.

We also tested interactions between anticipated longevity, gender, and marital status in order to capture potential differences in the interpretation of SPS by men and women. However, as there were no significant interactions among these variables (results available upon request), we only report results from regression models with main effects. Hausman’s specification tests (Hausman–McFadden tests) were performed to detect any potential violation of the Independence of Irrelevant Alternatives (IIA) assumption (Greene, 2000).

Results

Table 1 shows the unweighted sample distribution of the outcome variable. The majority (73%) of older respondents who lived more than 10 miles away from their nearest child at baseline stayed at that distance, while approximately 14% of respondents ended up living closer to children at the follow-up. Most of the increase in proximity was due to moves by children; about a third of the moves (5%) were by older respondents while about two thirds of the moves (9%) were by children. The transition rates for both types of moves were fairly similar in the two cohorts. Mortality, nursing home admission, and other types of attrition were, however, higher in the older (AHEAD) cohort.

Table 1.

Sample Distribution of Proximity Outcomes for Respondents Who Were Living More Than 10 Miles From Their Children at the Baseline.

	Both cohorts		AHEAD		CODA
Proximity outcome	N	%	N	%	N	%
Stayed further than 10 miles	2,080	72.70	961	68.45	1,119	76.80
Parents moved within 10 miles	155	5.42	77	5.48	78	5.35
Children moved within 10 miles	248	8.67	121	8.62	127	8.72
Institutionalization	55	1.92	44	3.13	11	0.75
Death	217	7.58	145	10.33	72	4.94
Other attrition	106	3.70	56	3.99	50	3.43
Total	2,861	100	1,404	100	1,457	100

Table 2 presents definitions for the explanatory variables and summary statistics for the two HRS cohorts. Notably, the average subjective probability of 10-year survival was 0.40 with fairly large variation (SD 0.32) for the AHEAD cohort and 0.55 (SD 0.31) for the CODA cohort.

Table 2.

Definitions and Descriptive Statistics for Analytic Variables: HRS 2000–2002, 2002–2004, United States.

		Cohort^a,b
Variable name	Coding	AHEAD		CODA
Subjective probability of surviving next 10 years (SPS)	Min = 0; Max = 1	0.399	(0.330)	0.564	(0.300)
Focal points in SPS distribution	SPS = 0	23.58	%	9.06	%
	SPS = 0.5	22.79	%	29.44	%
	SPS = 1	6.84	%	10.84	%
Demographics
Education	Lower than high school	20.23	%	16.06	%
	High school graduate	35.47	%	37.68	%
	Some college	23.50	%	22.10	%
	College and above	20.80	%	24.16	%
Gender	Male	0.45	%	0.43	%
Age group	69–74	—	%	65.68	%
	75–79	29.56	%	34.32	%
	80–84	49.22	%	—	%
	85+	21.23	%	—	%
Mean age in years	Min = 69, max = 90	81.70	(3.34)	73.43	(2.30)
Race or ethnicity	Non-Hispanic White	90.95	%	88.88	%
	Non-Hispanic Black	5.91	%	5.35	%
	Non-Hispanic other	0.50	%	1.58	%
	Hispanic	2.64	%	4.19	%
Family structure
Marital status	Married	53.56	%	73.44	%
	Widowed	40.67	%	17.84	%
	No spouse/divorced/separate/never married	5.77	%	8.72	%
Number of living children	Min = 1, max = 20	2.67	(1.65)	2.94	(1.74)
Number of marriages	Min = 1, max = 5	1.35	(0.61)	1.36	(0.70)
Child with income <35k	Yes/No	12.12	%	12.77	%
Unmarried children (%)	Min = 0; Max = 100	26.96	(33.07)	27.93	(31.46)
Number of grandchildren	Min = 0; Max = 28	2.63	(3.87)	2.62	(4.06)
Housing and wealth
Home ownership and housing wealth	HH value> = 100k	33.83	%	43.45	%
	HH value < 100k	46.30	%	43.99	%
	Rent: null HH value	14.96	%	10.02	%
	Other	4.91	%	2.54	%
Total net housing wealth ($1,000)	$: min = −0.3, max = 9,950	111	(317)	126	(136)
Total nonhousing wealth ($1,000)	$: min = −50, max = 33,300	198	(512)	168	(365)
Yearly earnings ($)	$: min = 0, max = 300,000	517.4	(3,344)	2519	(13,993)
Neighborhood tenure (years)	Min = 0, max = 89	36.97	(24.78)	30.98	(22.56)
Health status
Number of difficulties in ADL^c	Min = 0, max = 5	0.32	(0.78)	0.22	(0.69)
Number of difficulties in IADL^d	Min = 0, max = 5	0.26	(0.70)	0.13	(0.51)
Number of chronic conditions^e	Min = 0, max = 7	2.32	(1.33)	2.07	(1.35)
Self-reported health status	Excellent	9.83	%	11.05	%
	Very good or good	58.05	%	66.44	%
	Good or fair	32.12	%	22.51	%
Health insurance		22.15	%	26.22	%
Have private insurance	Yes/No
LTC insurance	Yes/No	13.32	%	16.61	%
Change in marital status	Stay married	48.12	%	70.08	%
	Stay unmarried	44.86	%	25.19	%
	Married to unmarried	6.18	%	4.06	%
Change in self-reported health status	Better or same	62.53	%	70.83	%
	Worse	33.00	%	26.41	%
	Much worse	4.47	%	2.75	%
Increase in ADL difficulties	Min = −5, max = 5	0.19	(0.94)	0.06	(0.62)
Onset conditions	Yes/No	19.09	%	18.53	%
Number of observation		1,404		1,457

Data source. HRS 2000, 2002 final release.

Percentages are reported for categorical variables.

Means and standard deviations (in parenthesis) are reported for continuous or count variables.

ADL include bathing, dressing, eating, getting in/out of bed, and walking across a room.

IADL include using the phone, managing money, taking medications, shopping for groceries, and preparing hot meals.

Items for self-reported chronic conditions include (1) high blood pressure or hypertension; (2) diabetes or high blood sugar; (3) cancer or a malignant tumor of any kind except skin cancer; (4) chronic lung disease except asthma such as chronic bronchitis or emphysema; (5) heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems; (6) stroke or transient ischemic attack; (7) emotional, nervous, or psychiatric problems; and (8) arthritis or rheumatism.

Tables 3 and 4 summarize results from multinomial logit regressions predicting changes in parent–child proximity relative to the reference category of staying 10 miles apart for the CODA and AHEAD cohorts, respectively. Table 3 presents the coefficients for the “parental move” outcome, and Table 4 shows analogous coefficients for the “children’s move” outcomes. Coefficients for attrition are not shown, though all were jointly estimated relative to the reference group. Each table shows the results from three regression specifications estimated separately for the younger and older cohorts, respectively.

Table 3.

Multinomial Logit Regression Coefficients for the Parental Move Model.

	AHEAD cohort			CODA cohort
Independent variables	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Subjective probability of 10-year survival	0.67*	0.78**	0.77**	0.70	0.78*	0.83*
Education (Ref: high school graduate)
Less than high school	−0.20	−0.16	−0.37	−0.61	−0.58	−0.80
Some college	−0.00	−0.06	−0.02	−0.16	−0.10	−0.06
College and above	−0.03	−0.08	−0.06	−0.12	−0.01	−0.01
Female (Ref: male)	0.53**	0.64**	0.72**	−0.54**	−0.57**	−0.72**
Ethnicity (Ref: non-Hispanic White)
Non-Hispanic Black	−1.46	−1.58	−2.06**	−0.27	−0.70	−0.60
Hispanic and Asian	0.05	−0.25	−0.16	−0.24	−0.67	−0.82
Age in years	−0.04	−0.06	−0.06	0.01	−0.00	−0.03
Self-reported health (Ref: excellent)
Very good or good	—	0.07	0.06	—	0.41	0.44
Fair or poor	—	0.60*	0.55	—	0.84**	0.85**
Health status
Number of ADL difficulties	—	−0.04	−0.00	—	−0.23	−0.38
Number of IADL difficulties	—	−0.09	−0.12	—	−0.08	−0.06
Number of chronic conditions	—	−0.02	−0.02	—	−0.11	−0.10
Health insurance
Has private insurance coverage	—	0.58	0.73**	—	0.22	0.23
Has LTC insurance	−	−0.90*	−0.97**	—	0.22	0.27
Financial
Logged total non-housing assets	—	−0.11	0.03	—	−0.57	−0.59
Logged total earning	—	−0.08	−0.06	—	0.04	0.03
Housing (HH value < 100k)
HH value ≥ 100k	—	0.10	0.10	—	0.17	0.27
Renting	—	0.71*	0.67	—	1.06**	1.05*
Other	—	−0.13	0.03	—	0.65	0.43
Tenure in same place	—	−0.01**	−0.01**	—	−0.02***	−0.02***
Marital status (ref: married )
Widowed	—	0.21	—	—	0.15	—
Divorced	—	−1.24	—	—	−1.51**	—
Number of marriages	—	−0.15	−0.16	—	0.15	0.10
Number of adult children	—	0.20**	0.18*	—	0.04	0.04
Any child with income < 35k	—	−0.11	−0.11	—	0.52	0.58
Percent of children unmarried	—	0.00	0.00	—	0.00	0.00
Number of grandchildren	—	−0.04	−0.04	—	0.00	0.00
Change in marital status (Ref: stay married)
Stay unmarried	—	—	0.05	—	—	−0.18
Married to unmarried	—	—	0.45	—	—	1.47***
Change in reported health (Ref: better or same)
Worse	—	—	−0.26	—	—	0.02
Much worse	—	—	−0.47	—	—	0.37
More ADL difficulties	—	—	0.36**	—	—	0.44**
Onset of conditions	—	—	−0.12	—		0.23
Constant	0.00	2.04	0.61	−3.63	5.23	6.93
Observations	1,404	1,404	1,155	1,457	1,457	1,304
Log pseudolikelihood	−1254.63	−1188.88	−700.65	−1092.96	−1026.54	−623.54
Wald χ²	96.63	231.44	9192.56	65.81	220.90	37938.14
Prob > χ²	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
Pseudo R²	0.041	0.091	0.1373	0.033	0.091	0.1447

Note. Robust standard errors (SE) are used to adjust for potential bias when the sample is clustered within families. Selection bias due to missing information on anticipated longevity has been adjusted using Mills’ ratio from the first stage of the Heckman selection model. Results from the first-stage Heckman model are not reported.

Significance level: *0.1, **0.05, ***0.01. Coefficients for the parental move outcome were jointly estimated with other outcomes: “proximity unchanged” (base category), “children’s move,” and “competing risks” in the same multinomial logit regression.

Table 4.

Multinomial Logit Regression Coefficients for the Children’s Move Model.

	AHEAD cohort			CODA cohort
Independent variables	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Subjective probability of 10-year survival	0.50	0.39	0.30	−0.69**	−0.53	−0.58
Education (Ref: high school graduate)
Less than high school	0.25	0.35	0.23	−0.12	−0.38	−0.34
Some college	−0.33	−0.41	−0.50	0.00	0.10	0.19
College and above	−0.37	−0.45	−0.72**	0.04	0.24	0.24
Female (Ref: male)	−0.18	−0.27	−0.16	0.22	0.23	0.16
Ethnicity (Ref: non-Hispanic White)
Non-Hispanic Black	0.38	0.20	0.23	0.08	−0.78	−0.97*
Hispanic and Asian	1.51***	1.50***	1.51***	0.87***	0.79**	0.70*
Age in years	−0.09*	−0.08	−0.10*	−0.06	−0.06	−0.07
Self-reported health (Ref: excellent)
Very good or good	—	−0.47	−0.67**	—	0.50*	0.58**
Fair or poor	—	−0.41	−0.55	—	0.79**	1.07***
Health status
Number of ADL difficulties	—	0.15	0.17	—	0.00	0.01
Number of IADL difficulties	—	0.00	0.08	—	−0.23	−0.25
Number of chronic conditions	—	0.16*	0.23**	—	−0.08	−0.09
Health insurance
Has private insurance coverage	—	−0.19	0.08	—	0.02	0.04
Has LTC insurance	—	0.18	0.04	—	−0.46	−0.53
Financial
Logged total nonhousing assets	—	−0.40	−0.41	—	−0.17	−0.17
Logged total earning	—	0.00	0.01	—	−0.01	−0.01
Housing (HH value <100k)
HH value ≥100k	—	0.52*	0.52*	—	0.10	0.11
Renting	—	0.27	0.36	—	0.03	0.06
Other	—	−1.45**	−1.48**	—	−1.62	−38.9**
Tenure of residence	—	0.00	0.00	—	0.01***	0.01***
Marital status (Ref: married)
Widowed	—	0.34	—	—	−0.26	—
Divorced	—	−0.43	—	—	−0.36	—
Number of marriages	—	0.04	0.04	—	0.33**	0.30*
Number of adult children	—	0.09	0.13	—	0.20***	0.22***
Any child with income <35k	—	0.60	0.46	—	1.05***	1.19***
Percent of children unmarried	—	0.00	0.00	—	0.00	0.00
Number of grandchildren	—	0.01	0.00	—	−0.01	−0.01
Change in marital status (Ref: stay married)
Stay unmarried	—	—	0.34	—	—	−0.13
Married to unmarried	—	—	−0.10	—	—	0.92**
Change in reported health (Ref: better or same)
Worse	—	—	−0.09	—	—	−0.40
Much worse	—	—	−1.19	—	—	−0.47
More ADL difficulties	—	—	0.04	—	—	0.06
Onset of conditions	—	—	0.26	—	—	0.16
Constant	5.04	9.07	11.1	1.89	2.69	3.16
Observations	1,404	1,404	1,155	1,457	1,457	1,304

Note. Robust standard errors (SE) are used to adjust for potential bias when the sample is clustered within families.

The sample size of the last regression for each subsample is smaller than that in previous regressions because the information needed to construct changes in health and marital status is not available in the waves prior to the each baseline for a number of eligible respondents. Additional analyses show that omitting these respondents from previous regressions had little impact on coefficients and standard errors.

Selection bias due to missing information on anticipated longevity has been adjusted using Mills’ ratio from the first stage of the Heckman selection model. Results from the first-stage Heckman model are not reported.

Significance level: *0.1, **0.05, ***0.01. Coefficients for the parental move outcome were jointly estimated with other outcomes: “proximity unchanged” (base category), “parent’s move,” and “competing risks” in the same multinomial logit regression. Fit statistics for the joint model are reported in Table 3.

Table 3 shows that greater anticipated longevity (measured by SPS) is associated with a higher probability of parents moving closer to children in both cohorts. Women in the older cohort (AHEAD) were more likely to move closer to children than their male counterparts; in contrast, females in the younger cohort (CODA) were less likely than males to make a proximity-enhancing move. Respondents who were divorced at baseline were less likely to move closer to children, though this association is only significant for the younger cohort. In both cohorts, those who experienced martial disruptions during the previous 2 years were more likely to move closer to their children.

Notably, the magnitude of the SPS effect increases in Models 2 and 3 (AHEAD) as well as 5 and 6 (CODA), which account for variation in parental health needs and resources as well as basic sociodemographic variables. Respondents with worse self-reported health at baseline were more likely to move closer to their children, but this association is statistically significant only for the younger cohort (CODA). Other baseline health measures were not significant for either cohort. Changes in disability were significant for both cohorts—those who experienced worsening ADL difficulties over the interval were more likely to move closer to their children.

For the older cohort (AHEAD), assets and income were not significant predictors of parental moves, but having LTC insurance reduced the probability of a proximity-enhancing move. The costs of relocation also appeared to reduce the likelihood of moving: Owning a home and residential tenure were negatively associated with parental moves for both cohorts.

Table 4 presents the coefficients from an analogous set of three regression specifications predicting the “children’s move” outcome, where respondents did not change their residence yet ended up living closer to their children. Most parental characteristics differ in their relationships with children’s moves across the HRS cohorts. Parents’ anticipated longevity is significantly negatively associated with children’s proximity-enhancing moves for the younger group of parents before other covariates are included in the models, but the relationship is not significant for either cohort once health needs and other socioeconomic and demographic variables are included in subsequent models.

For the younger CODA cohort, those who reported fair or poor self-rated health were more likely to have their children move closer, but chronic conditions and disability were not significant predictors of children’s moves. For the older cohort, however, parents with more chronic conditions, rather than those with poor self-rated health, were more likely to have their children move closer. Once changes in health were included (Model 3), better self-rated health at baseline diminished the likelihood that children would move closer in the interval.

Most parental wealth variables do not have an effect on children’s moves for either cohort, although parental housing assets are marginally positively associated with children’s moves for the older cohort. Children’s economic status does, however, have an effect on their likelihood of moving. Respondents who have low-income children are more likely to have a child move closer, an association that is stronger and statistically significant for the younger cohort.

Hispanic and Asian respondents in both cohorts are more likely to have their children move closer compared to other ethnic groups. In the younger cohort, parents who experienced marital disruption in past 2 years are more likely to have their children move closer.

The fit statistics indicate that the larger, more detailed models explain a greater proportion of the variability in proximity outcomes.

Discussion

Fourteen percent of older adults in our sample experienced an increase in intergenerational proximity—due either to their own residential changes or moves by their adult children—over the 2-year study interval. Consistent with previous findings (e.g., Michielin et al., 2008; Rogerson et al., 1997; Tomassini et al., 2003), we show that among families experiencing enhanced intergenerational proximity, children move closer to their parents 1.6 times as often as older parents relocate, and these moves are potentially in response to the children’s own life circumstances as well as their parents’ need for support.

Older parents with longer anticipated longevity were more likely to move closer to their children, controlling for other factors including a broad set of health status variables. A long-standing assumption in much of the gerontological literature has been that living arrangements and intergenerational proximity are responses to the health and LTC needs of the older generation. Our results suggest that these decisions, especially for older parents, may instead reflect the impact of a perceived time horizon: while the financial, physical, and psychological costs of moving may be substantial, high anticipated longevity suggests more time for offsetting the costs of moving and enjoying the benefits of intergenerational proximity. Notably, proximity-enhancing moves by adult children do not seem to be influenced by their parents’ anticipated longevity, suggesting that parents’ anticipated longevity may either be unknown to children, or considered less relevant than factors such as contemporaneous health or economic needs.

The life course migration framework sets forth a typology that characterizes later-life moves in terms of responses to and anticipation of events, such as retirement and declining health. However, individuals’ preparation for and response to such events vary, at least partially, according to individuals’ perceptions of their own time horizons. Anticipated longevity may not equate perfectly with a plan for the future—an older person expecting a longer life span does not necessarily have a strategy for the longer horizon—yet the expected duration of one’s remaining life span does inform planning.

Anticipated longevity may have different meanings for parents at different stages of the life course. For example, younger parents who believe they will remain healthy for a long time may defer retirement moves in order to enjoy more time with children and grandchildren. They may be inclined to help children who are relatively worse-off in the family by moving closer to them. Alternatively, at older ages, greater anticipated longevity may foreshadow a longer period of life with informal care and assistance needs, prompting proximity-enhancing moves.

Our findings also suggest that factors that potentially increase the “cost” of moving diminish parents’ likelihood of moving closer to their children, as those who own their home or have lived in the same neighborhood for longer are less likely to move. Parents’ neighborhood tenure was associated with a higher probability of children moving closer, though the association is significant only for the younger cohort (CODA).

Consistent with prior findings (Rogerson et al., 1997; Silverstein, 1995), we find that life changes that increase needs for family support raise the odds of parents moving closer to their children. For instance, respondents with increases in ADL difficulties and those who experienced marital disruptions (largely due to widowhood) in the 2-year interval were more likely to move closer to their children. Marital status changes were significant only for the younger cohort, a finding that merits further investigation. It may be that couples are more likely to move away from family to pursue amenities, but when a partner is lost, the remaining spouse is not sufficiently attached to the community to stay.

The availability of formal alternatives to family support may also influence migration decisions, especially those related to health care needs. Respondents in the older cohort (AHEAD) who had LTC insurance were less likely to move closer to their children and no more likely to have children move closer to them, even controlling for health variables. This finding suggests older people consider the trade-offs between formal and informal care in determining their living arrangements. It would be interesting to know whether the LTC policies themselves discourage moving due to restrictions on covered care and delivery locations.

Children’s needs also appear to influence their decision to move closer to their parents. Respondents who have any low-income children were more likely to have at least one child move closer to them, though this association is only statistically significant in the younger cohort and the data do not allow us to determine whether it was the poor child who in fact moved. Still, these findings highlight the importance of considering moves by both parents and adult children and are consistent with those reported by Michielin et al. (2008). They support the idea that family members collectively adjust intergenerational proximity to facilitate mutual support.

The relationship of race and ethnicity to intergenerational proximity-enhancing moves is not straightforward. African American parents in the AHEAD cohort are less likely than non-Hispanic Whites to move closer to their children, and coefficients for other minority groups are in the same direction, though not significant. To some extent, this may reflect selection into the more distant category at baseline. Research has shown that persons with higher education and socioeconomic status tend to move more often (and further away) than their counterparts (Malmberg & Petterson, 2007; Michielin & Mulder, 2007; Silverstein, 1995), and minority elders who live further from children may be more selected than their White counterparts. More substantial race and ethnicity effects are seen in children’s moves: Hispanic and Asian children are more likely to move closer to their parents as compared with their White counterparts, even when controlling for needs, resources, and other parent and child characteristics. Since Asian and Hispanic families are more likely than White or African American families to be relatively recent migrants to the United States, these findings raise questions about the relationship of intergenerational proximity to assimilation and about cultural differences in family networks and flows of intergenerational support (see, e.g., Gurak & Kritz, 2010; Peek, Koropeckyj-Cox, Zsembik, & Coward, 2004).

As compared to males, female respondents among the younger cohort (CODA) are less likely to move closer to their children while females in the older cohort (AHEAD) are more likely to do so. Though one might think that these differences are due to the greater likelihood of older women to be widowed or unmarried at baseline, interaction terms between gender and baseline marital status were examined but were not statistically significant. The greater kin-keeping role of women may mean that they more commonly move to be closer to and care for grandchildren, regardless of their own health or children’s socioeconomic needs.

We did not find evidence that respondents’ age is associated with their proximity-enhancing moves, but we found that the likelihood of adult children moving closer to parents decreases with the parent’s age for the older cohort (AHEAD), but not the younger cohort (CODA). This might be explained by differences in the age of the adult children themselves: Those born to AHEAD parents may be in the midst of retirement or amenity activities of their own which pull them away from family, while adult children of CODA parents are likely still working.

While our results are strongly suggestive of important differences in the factors that motivate older parents and their children to relocate, some limitations should be acknowledged.

Studies of residential mobility depend upon the definition and classification of moves (Sergeant, Ekerdt, & Chapin, 2008). Some of our analytic decisions were constrained by the design of the HRS data. We were only able to ascertain proximity-enhancing moves across a distance of 10 miles. As suggested by previous life course studies (Clark & Wolf, 1992; Lin & Rogerson, 1995) and our own analysis, older adults living more than 10 miles from any of their adult children at baseline differ from those who co-reside or live within 10 miles in gender, marital status, wealth and home ownership, and number of children. Furthermore, as proximity was only ascertained at the time of each HRS interview, intervening moves were not captured, leading to an underestimation of the total number of transitions within the 2-year intervals.

Since respondents were asked about their children as a group, we were not able to calculate transition probabilities and examine changes in proximity vis-à-vis individual children. To better understand how specific children are selected to take care of the older generation, future analysis should employ information about specific parent–child dyads to consider residential relocation from the perspective of adult children.

Change in residence could in and of itself lead to a statistical association between change in proximity and the factor leading to relocation, regardless of whether this particular factor is associated with proximity in its own right. Nonetheless, our separate analysis of the probability of older respondents’ relocation during follow-up shows that the observed associations with proximity are not likely to be driven by such factors (results available upon request).

Finally, because our primary interest is in whether adult children live near their older parents and are able to serve as potential caregivers, we do not examine moves that did not change proximity, or that involved moves from living nearby to co-residing. Further research with data that more precisely capture different residential arrangements across time will elaborate the potentially different decision-making processes they involve.

Conclusion

Proximity between older parents and their adult children is a key indicator of the potential availability of family members to provide support to each other. Understanding what factors influence proximity-enhancing moves is important for research and policy related to the provision of informal care to older adults as well as the contributions they make by providing child care and support to adult children in need.

Notably, when there was an increase in intergenerational proximity, it was most often the result of moves by adult children rather than by parents, suggesting that the younger generation plays a more important role in shaping intergenerational proximity than might often be assumed. While previous research described how parents’ needs, resources, and preferences shape decisions regarding living arrangements (e.g., Soldo, Wolf, & Agree, 1990), researchers are only beginning to examine whether and how adult children’s relocation decisions are influenced by their parents’ needs or their own need for support (Michielin et al., 2008). Our findings suggest that children make choices collectively with other family members to create effective living arrangements. Future research on the well-being of older people should thus adopt a more inclusive intergenerational approach and pay more attention to the needs, preferences, resources, and life circumstances of both the younger and the older generations in the family.

Older parents with greater anticipated longevity are more likely to move closer to their children, holding all else equal. The influence of anticipated longevity on parental moves shows that variation in the future planning horizon can influence the timing of migration transitions, even when controlling for current health needs. Although anticipated longevity only serves as a proxy for the planning horizon, its observed association with proximity-enhancing moves provides fresh supporting evidence for the key role of anticipatory behaviors in the decisions families make regarding older generations’ living arrangements.

Further research will need to take into account increasing variability in family forms and relationships in understanding family support in later life (Agree & Hughes, 2012). The present study illustrates the importance of including individual variability in perceived timelines in the life course framework, and provides a starting point for the development of more complex models that accommodate multiple generations. While the life course migration framework has previously rested entirely on the perspective of a single generation, our findings enhance the framework by showing that changes in intergenerational proximity are due to moves by both parents and children and that each type of move is influenced by a different set of factors.

Footnotes

Acknowledgements

The authors would like to thank Douglas A. Wolf for his helpful suggestions on the design of this study.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

Yiduo Zhang is an associate director at MedImmune. His research focuses on health economics and program evaluation.

Michal Engelman is an assistant professor in the Department of Sociology at the University of Chicago. She is a demographer and gerontologist studying the dynamics of population aging and the determinants of longevity and well being at older ages.

Emily M. Agree is the director of the Hopkins Center for Population Aging and Health and the associate director of the Hopkins Population Center. She is an associate professor in the Department of Population, Family and Reproductive Health at the Johns Hopkins Bloomberg School of Public Health. Her current research focuses on the demography of aging, disability trends, family caregiving, and assistive technology.

References

Agree

E. M.

Hughes

M. E.

(2012). Demography of families in later life. In Blieszner

Bedford

(Eds.), Handbook on families and aging (2nd ed.). Westport, CT: Praeger.

Bengtson

V. L.

(2001). Beyond the nuclear family: The increasing importance of multigenerational bonds. Journal of Marriage and the Family, 63, 1–16.

Chen

P. C.

Wilmoth

J. M.

(2004). The effects of residential mobility on ADL and IADL limitations among the very old living in the community. Journal of Gerontology: Social Sciences, 59, S164–S172.

Clark

Wolf

(1992). Proximity of children and elderly migration. In Rogers

(Ed.), Elderly migration and population redistribution: A comparative perspective (pp. 77–96). London, England: Belhaven Press.

Crimmins

E. M.

Ingegneri

D. G.

(1990). Interaction and living arrangements of older parents and their children: Past trends, present determinants, future implications. Research on Aging, 12, 3–35.

Ferraro

K. F.

(1981). Relocation desires and outcomes among the elderly: A longitudinal analysis. Research on Aging, 3, 166–181.

Gan

Gong

Hurd

M. D.

McFadden

(2004). Subjective mortality risk and bequests. National Bureau of Economic Research, Inc. Retrieved from http://ideas.repec.org/p/nbr/nberwo/10789.html

Greene

(2000). Econometric analysis (4th ed.). New Jersey, NJ: Prentice Hall.

Gurak

D. T.

Kritz

M. M.

(2010). Elderly Asian and Hispanic foreign- and native-born living arrangements: Accounting for differences. Research on Aging, 32, 567–594.

10.

Hiedemann

Stern

(1999). Strategic play among family members when making long-term care decisions. Journal of Economic Behavior & Organization, 40, 29–57.

11.

Hurd

M. D.

McGarry

(1995). Evaluation of the subjective probabilities of survival in the health and retirement study. The Journal of Human Resources, 30, S268–S292.

12.

Hurd

M. D.

McGarry

(2002). The predictive validity of subjective probabilities of survival. Economic Journal, Royal Economic Society, 112, 966–985.

13.

Hurd

M. D.

Smith

J. P.

Zissimopoulos

J. M.

(2002). The effects of subjective survival on retirement and social security claiming. National Bureau of Economic Research, Inc. Retrieved from http://ideas.repec.org/p/nbr/nberwo/9142.htm

14.

Juster

F. T.

Suzman

(1995). An overview of the health and retirement study. The Journal of Human Resources, 30, S7–S56.

15.

Lee

G. R.

(1980). Kinship in the seventies: A decade review of research and theory. Journal of Marriage and the Family, 42, 923–934.

16.

Lin

Rogerson

P. A.

(1995). Elderly parents and the geographic availability of their adult children. Research on Aging, 17, 303–331.

17.

Litwak

Longino

C. F.

(1987). Migration patterns among the elderly: A developmental perspective. The Gerontologist, 27, 266–272.

18.

Litwak

Kulis

(1987). Technology, proximity, and measures of kin support. Journal of Marriage and the Family, 49, 649–661.

19.

Longino

C. F.

(1990). Geographical distribution and migration. In Binstock

R. H.

George

L. K.

(Eds.), Handbook of aging and the social sciences (3rd ed., pp. 45–63). San Diego, CA: Academic Press.

20.

Longino

C. F.

Jr Jackson

D. J.

Zimmerman

R. S.

Bradsher

J. E.

(1991). The second move: Health and geographical mobility. Journal of Gerontology: Social Sciences, 46, S218–S224

21.

Longino

C. F.

Bradley

D. E.

Stoller

E. P.

Haas

W. H.

(2008). Predictors of non-local moves among older adults: A prospective study. The Journal of Gerontology B: Psychological Sciences and Social Science, 63, S7–S14.

22.

Malmberg

Pettersson

(2007). Distance to elderly parents: Analyses of Swedish register data. Demographic Research, 17, 679–704.

23.

Michielin

Mulder

C. H.

(2007). Geographical distances between adult children and their parents in the Netherlands. Demographic Research, 17, 655–678.

24.

Michielin

Mulder

C. H.

Zorlu

(2008). Distance to parents and geographical mobility. Population, Space and Place, 14, 327–345.

25.

Mirowsky

Ross

C. E.

(2000). Socioeconomic status and subjective life expectancy. Social Psychology Quarterly, 63, 133–151.

26.

Mulder

C. H.

(2007). The family context and residential choice: A challenge for new research. Population, Space and Place, 13, 265–278.

27.

Peek

C. W.

Koropeckyj-Cox

Zsembik

B. A.

Coward

R. T.

(2004). Race comparisons of the household dynamics of older adults. Research on Aging, 26, 179–201.

28.

Rogerson

P. A.

Burr

J. A.

Lin

(1997). Changes in geographic proximity between parents and their adult children. International Journal of Population Geography, 3, 121–136.

29.

Rogerson

P. A.

Weng

R. H.

Lin

(1993). The spatial separation of parents and their adult children. Annals of the Association of American Geographers, 83, 656–671

30.

Ross

C. E.

Mirowsky

(2002). Family relationships, social support and subjective life expectancy. Journal of Health and Social Behavior, 43, 469–489.

31.

Rossi

A. S.

Rossi

P. H.

(1990). Of human bonding: Parent-child relations across the life course. New York, NY: Aldine de Gruyter.

32.

Ruggles

(1994). The transition of American family structure. American Historical Review, 99, 103–128.

33.

Ruggles

(1996). Living arrangements of the elderly in America: 1880–1980. In Hareven

T. K.

(Ed.), Aging and generational relations over the life course: A historical and cross-cultural perspective (pp. 254–271). Berlin, New York, NY: Walter de Gruyter.

34.

Ruggles

(2001). Living arrangements and well-being of older persons in the past. In Ageing and living arrangements of older persons: Critical issues and policy responses. New York, NY: Population Bulletin of the United Nations, Special Issue Nos. 42/43. Retrieved from http://www.un.org/esa/population/publications/bulletin42_43/bulletin42_43.htm

35.

Sergeant

J. F.

Ekerdt

D. J.

Chapin

(2008). Measurement of late-life residential relocation: Why are rates for such a manifest event so varied? The Journal of Gerontology B: Psychological Sciences and Social Science, 63, S92–S98.

36.

Silverstein

(1995). Stability and change in temporal distance between the elderly and their children. Demography, 32, 29–45.

37.

Silverstein

Angelelli

J. J.

(1998). Older parents’ expectations of moving closer to their children. The Journal of Gerontology B: Psychological Sciences and Social Science, 53, S153–S163.

38.

Smith

G. C.

(1998). Residential separation and patterns of interaction between elderly parents and their adult children. Progress in Human Geography, 22, 368–384.

39.

Soldo

B. J.

Wolf

D. A.

Agree

E. M.

(1990). Family, household, and care arrangements of frail older women: A structural analysis. Journal of Gerontology: Social Sciences, 45, S238–S249.

40.

Speare

Avery

(1993). Who helps whom in older parent-child families. Journal of Gerontology: Social Sciences, 48, S64–S73.

41.

Tomassini

Wolf

D. A.

Rosina

(2003). Parental housing assistance and parent–child proximity in Italy. Journal of Marriage and Family, 65, 700–715.

42.

Wilmoth

J. J.

(2010). Health trajectories among older movers. Journal of Aging and Health, 22, 862–881.