Individual- and Family-Level Mobility Effects

Abstract

This Chapter explores whether or not household mobility has negative effects for individuals, particularly children, and their families. Drawing on a large, interdisciplinary body of research on “mobility effects,” this chapter also explores several reasons why research on the effects of moving on children has led to mixed conclusions. Nationally representative data from the National Longitudinal Survey of Youth 1997 (NLSY97) are used to explore the short and long-term effects of local and distance mobility, including the frequency of moving, across a number of different outcome domains.

Appendix: Additional Details for Data and Analysis in Chap. 5

NLSY97 Sample

The NLSY97 is a nationally representative sample of 8984 adolescents in 1997 who were born from 1980 through 1984 and were 12 through 17 years of age during the initial 1997 round. The annual multi-topic survey collects extensive information on child and family interactions and relationships. It also contains an array of information about parenting practices, parent–child closeness, and other environmental characteristics. Furthermore, the survey collects information on parents and siblings of the respondents. The studies in Chap. 5 draw on data from the 1998, 1999, and 2006, 2008, and 2010 waves.

A chained multiple imputation procedure was utilized to handle missing data. The dependent and independent variables were used to construct the imputations but imputed values for dependent variables were excluded from the analyses. The imputation procedure produced ten imputed datasets and the imputed estimates were subsequently combined. Descriptive statistics and parameter estimates for each imputed dataset were virtually identical.

The measures and analytic procedures are discussed below as they relate to each of the three studies presented in Chap. 3. Study 1 explored short-term behavioral and health outcomes of household mobility on adolescents (N = 8140). Study 2 explored the long-term outcomes of childhood mobility on young adults (N = 6944). Study 3 explored the effects of household mobility on family processes for a subsample of adolescents (N = 4223).

Study 1: Short-Term Adolescent Outcomes of Household Mobility (N = 8140)

Dependent Variables

Adolescent Delinquency

The adolescent delinquency index is based on the respondent child’s response to ten questions about delinquent behaviors between 1998 and 1999. The behaviors were related to running away, carrying a handgun, belonging to a gang, destroying or damaging property, minor stealing, major theft, other property crimes, fighting and physical assault, selling drugs, and being arrested by the police for a delinquent offense. Summing the affirmative responses to the delinquent acts yields an index, ranging from 0 to 10, with previously established predictive and internal validity (Center for Human Resource Research 1999). A log-transformed measure of adolescent delinquency was modeled because the variable was highly positively skewed (kurtosis = 15.6). Distance mobility was still a significant predictor of increased behavior problems using a number of different transformations of the variable. However, models 1.1 and 1.2 use the log of adolescent delinquency in 1999 as the dependent variable.

Adolescent Self-Reported Health

The child respondents were asked to report on their overall health as being (0) poor, (1) fair, (2) good, (3) very good, or (4) excellent.

Independent Variables

Household mobility indicates whether the respondent reported (0) not moving, (1) moving locally, or (2) moving to another city, county, or state between 1998 and 1999. Move frequency was a separate measure that explored whether (0) no move, (1) a single move, or (2) multiple moves took place between the 1998 and 1999 waves.

In order to tease out the short-term effects of household mobility, a control variable was included for a number of childhood moves. This variable documented the number of adolescent household relocations that occurred prior to age 12; the measure was top-coded at 10. An additional measure for number of schools attended was included to control for potentially confounding effects of school mobility. However, as discussed in Chap. 5, this variable measured any school moves, even promotional changes (e.g., from middle school to high school).

Additional variables for individual and family characteristics included age, gender (male/female), and number of siblings. Race was classified as non-Black/non-Hispanic, Black, Hispanic, and mixed race. A dichotomous variable indicated whether or not either parent had a college degree. Household structure identified children as residing with both biological parents (omitted for reference), a biological parent and stepparent, a single parent, or another household structure. An additional household-level variable identified whether or not any household structure change occurred between 1998 and 1999. A measure for urbanicity assessed whether or not children relocated to an urban or suburban (1) as opposed to rural (0) area. The models included a measure for logged household income.

Analysis Notes

Descriptive statistics for each of the variables discussed above are presented in Table 5.5. Multilevel logistic regression models explored the effects of household mobility on adolescent delinquency. Additionally, multilevel ordered logistic regression models were used to examine household mobility effects on self-reported health. For both outcomes, separate models explored the effects of move type (distance versus local) and move frequency (one move versus multiple moves).

Table 5.5 Descriptive statistics for adolescent outcomes, age 12–16

In order to control for child outcomes before and after the move took place, each model included a lagged control for the dependent variable in 1998 and a measure of the other dependent variable in 1999. For example, the model that predicts delinquent behavior in 1999 includes a control for delinquent behavior in 1998 as well as a control for self-reported health in 1999. Variance inflation factors were checked in order to assess any severe multicollinearity in the models (average VIFs < 1.1). Analysis of the correlation matrix (not shown) indicated that none of the observed relationships between the independent variables in the models were very strong—the strongest correlation (0.32) was between household income and parental education.

Study 2: Long-Term Outcomes of Childhood Mobility on Young Adults (N = 6944)

Dependent Variables

Young Adult Educational Attainment

Information for young adults’ educational attainment was collected in 2010. Individuals were categorized as having (0) no academic degree, (1) a high school diploma or equivalent, (2) some college or AA degree, (3) a college degree, or (4) a graduate degree. As noted in Chap. 5, an additional model included a collapsed dichotomous measure to indicate whether or not an individual had a college degree.

Young Adult Life Satisfaction

A global measure of subjective well-being asked respondents to rate their current life satisfaction. Respondents were presented with the following question, “All things considered, how satisfied are you with your life as a whole these days?” Respondents were asked to provide an answer from 1 to 10, where 1 means extremely dissatisfied and 10 means extremely satisfied.

Young Adult Happiness

In addition to a global measure of overall satisfaction, respondents are asked biennially about their happiness in the last month. Individuals are presented with the following question: “How much of the time during the last month have you been a happy person?” The response options were (1) none of the time, (2) some of the time, (3) most of the time, and (4) all of the time. The responses were averaged across the years 2006, 2008, and 2010 to provide an average young adult happiness score (range = 1–4).

Young Adult Self-Reported Health

Similar to the measure of adolescent self-reported health in study 1 (which was included as a control variable in these analyses), the young adult respondents were asked to report on their overall health as being (0) poor, (1) fair, (2) good, (3) very good, or (4) excellent.

Independent Variables

Childhood Household Mobility

To tease out the long-term versus short-term effects of household mobility, a variable denoting moves that occurred between 2006 and 2010 was included in the models. A number of additional independent variables were included in the models in order to control for individual and family effects in childhood and young adulthood. At the individual-level, and similar to the models for study 1, these models included respondents’ age and gender (male/female).

Race was classified as non-Black/non-Hispanic, Black, Hispanic, and mixed race. In order to control for physical and emotional well-being in adolescence, a sadness and depression measure indicated whether the adolescent rated the statement “You are unhappy, sad, or depressed” as being (0) not true, (1) sometimes/somewhat true, or (2) always true in 1997. Additionally, a measure for adolescent self-reported health in 1997 was included, with the same response options as those reported above.

At the adolescent household level, the model included measures for number of siblings. A retrospective measure of household structure at age 12 (collected starting in 2002) indicated whether respondents resided with both biological parents (omitted reference), a single parent or a blended family, or another household structure. A measure was also included for logged household income in adolescence (1997). Similar to the measure in study 1, the number of schools the respondent attended by 1999 was also included.

To control for the effects of individual and household characteristics in young adulthood, a variable was included for logged family income in 2010. A dichotomous measure assessed whether or not the respondent was married in 2010. Number of children in 2010 was also included as a young adult household-level control variable. Lastly, a measure for urbanicity indicated whether the young adult lived in an urban or suburban (1) as opposed to rural (0) area in 2010.

Analysis Notes and Limitations

Descriptive statistics for each of the variables discussed above are presented in Table 5.6. In Table 5.3of Chap. 5, models 1–4 include each of the controls for individual and family characteristics in both adolescence and adulthood discussed above. Each model also includes controls for the dependent variables used in each of the other models.

Table 5.6 Descriptive statistics for long-term mobility outcomes, age 24–32

Examination of variance inflation factors did not indicate there was any severe multicollinearity in the models (average VIF = 1.4). Analysis of the correlation matrix (not shown) indicated that none of the observed relationships between the independent variables in the models was very strong—the strongest correlation (0.44) was between life satisfaction in 2008 and the average happiness score for 2006–10. Multilevel ordered logistic regressions were modeled for each of the outcomes.

Study 3: Effects of Household Mobility on Family Processes (N = 4223)

Dependent Variables

Adolescent Family Routines

A family routine’s scale sums responses to how many days in a typical week (1) the respondent had dinner with the family; (2) did something fun as a family, such as played a game, went to a sporting event, went swimming, and so forth; (3) did something religious as a family, such as going to church, praying, or reading the scriptures together; or (4) the housework got done when it is supposed to, for example, cleaning up after dinner, doing dishes, or taking out the trash. The resulting scale, which ranges from 0–28 with a higher score indicating a higher level of routine activities, has been shown to be high in predictive validity (Center for Human Resource Research 1999).

Parental Monitoring

For four items with response categories ranging from 1 to 4, the youth reported on the monitoring practices of his or her mother: (1) how well she knows her child’s close friends; (2) how well she knows her child’s friend’s parents; (3) if she knows who her child spends time with when her child is not at home; and (4) how well she knows her child’s teachers and school. These items were summed, creating a 16-point parental awareness scale, with higher scores indicating greater levels of parental awareness (α = 0.71).

Parenting Style Changes

Based on the commonly used four-quadrant typology of parenting style (Maccoby and Martin 1983), adolescents responded about whether or not they considered their mothers “very supportive, somewhat supportive, or not very supportive.” A separate item asked whether they considered their mothers “permissive or strict about making sure you did what you were supposed to do.” For responsiveness, “very supportive” responses are coded “1,” else “0.” For demandingness, “demanding,” responses are coded “1,” else “0.” Combined, the variables create a two-by-two typology of parenting style: authoritative (demanding and supportive), authoritarian (demanding and not very supportive), permissive (nondemanding and very supportive), and uninvolved (nondemanding and not supportive). These parenting style measures are high in construct and predictive ability (Center for Human Resource Research 1999). Based on the parenting style categories, a dichotomous variable indicates whether the mother’s parenting style changed between 1998 and 1999 (change = 1, else = 0).

Independent Variables

Similar to study 1, household mobility indicates whether the respondent reported (0) not moving, (1) moving locally, or (2) moving to another city, county, or state between 1998 and 1999. Move frequency was a separate measure that explored whether (0) no move, (1) a single move, or (2) multiple moves took place between the 1998 and 1999 waves.

Additional individual and household level variables were included for age, gender (male/female), and the presence of siblings. Race was categorized as non-Black/non-Hispanic, Black, Hispanic, and mixed race. At the household and family level, a dichotomous variable for parental education indicated whether or not either parent had a college degree. Household structure identified children as residing with both biological parents (omitted for reference), a biological parent and stepparent, a single parent, or another household structure. An additional household-level variable identified whether or not any household structure change occurred between 1998 and 1999.

The models included a measure for logged household income. Additionally, a measure for urbanicity indicated whether or not the household relocated to an urban or suburban (1) as opposed to rural (0) area. To tease out the effects of earlier household mobility, childhood mobility indicated the number of household relocations of any type that occurred in adolescence (top-coded at 10).

Additional controls included parental behaviors and characteristics known to influence family processes, including parenting. Mother’s self-reported health in 1997, the only year for which the information was available, used the same response options as the self-reported health measures reported above. Parent religiosity is based on six questions that describe how the mother felt about religion and religious practices in 1997. The items were summed to produce a scale ranging from 0 (not religious) to 6 (very religious) (α = 0.60).

To assess the quality of the mother–child relationship after the move takes place, mother–child closeness is based on the adolescent’s report on their relationship with their mother in 1999. The emotional dimension includes “She is a person I want to be like” and “I really enjoy spending time with her.” The behavioral dimension is based on two 5-point Likert items ranging from 0–4 (never to always): “How often does she praise you for doing well?” and “How often does she help you do things that are important to you?” The items were summed to create a scale, ranging from 0–16, with the highest scores indicating a stronger mother–child relationship (α = 0.77).

Analysis Notes and Limitations

Models 1 and 2 in Table 5.4 used multilevel OLS regression to model the effects of moving on changes in family routines and parental monitoring. Model 3 used logistic regression to examine the effects of moving on any change in parenting style between 1998 and 1999. Descriptive statistics for each of the variables discussed above are presented in Table 5.7.

Table 5.7 Descriptive statistics for family outcomes, age 12–14

A collective account of the chapter’s limitations is below. However, one data limitation specific to study 2 is that the childhood mobility measure is imperfect for a number of reasons. First, the variable is based on retrospective assessment of mobility behavior in childhood; therefore, the data could reflect some recall bias. Second, the variable does not have information on whether, or how many of, these moves were long-versus short-distance moves. As Chap. 5 demonstrates, there are different effects associated with different types of moves. Third, the variable assesses the number of moves made during childhood, but nothing is known regarding the frequency with which moves were made. For example, four moves before age 12 with 2-year intervals between moves might have very different effects than four moves before age 12 where all occurred within the span of 2 years.

In order to control for family processes before and after the move took place, each model controlled for family processes in 1998 (before the move took place). Additionally, the measure for respondents’ reported closeness to their mother was included to control for the quality of the family relationship after the move took place.

Variance inflation factors were checked in order to assess any severe multicollinearity in the models (average VIFs < 1.2). Analysis of the correlation matrix (not shown) indicated that none of the observed relationships between the independent variables in the models was very strong—the strongest correlation (0.42) was between mother’s monitoring score in 1998 and closeness to mother in 1999.

There are several limitations particular to study 3. First, the parenting style typology consists of only two measures (i.e., demandingness and responsiveness). Although this typology has been validated in recent research (Bronte-Tinkew et al. 2006; Bronte-Tinkew et al. 2010; Baumrind et al. 2010), this two-measure typology may be less stable than a continuum-based scale for parenting style and change. However, analyses conducted by the Center for Human Resource Research (1999) confirm that the parenting style typology is a high-quality measure with both construct and predictive validity. Another limitation involves the fact that adolescents provide information about the parent–child relationship. Mothers might perceive themselves to be more demanding and more responsive than their adolescents perceive them to be. Along the same lines, they could also report more monitoring and involved parenting than their children report. Therefore, single-source bias may affect the validity of the outcome measures.

Overall Study Limitations

The NLSY97 does not contain information on the reason for moving. This is an important limitation since the circumstances of a move (e.g., upward mobility following a promotion versus downward mobility following divorce) can influence individual and family outcomes. An additional limitation is that the NLSY97 does not include information about whether or not childhood mobility (before age 12) occurred over long or short distances. Thus, the analyses did not explore whether or not there were particularly pronounced effects for adolescents and families that frequently moved across long distances versus those who made frequent local moves.

Also, individual and family outcomes were only assessed for children starting at age 12. As such, the findings may not hold for younger children. Of course, development through adolescence and young adulthood introduces a complex issue. For instance, is it mobility that leads to change in parenting behavior or is it just autonomy associated with the transition to adulthood? At the same time, parents who move frequently may already be the type of parents to exhibit inconsistent parenting styles.