آمد و رفت: توضیح اثرات تحرک اقامتی و مدرسه در بزهکاری نوجوانان

Abstract Over the past half century, a large body of theoretical and empirical work in sociology and other social sciences has emphasized the negative consequences of mobility for human development in general, and youth outcomes in particular. In criminology, decades of research have documented a link between residential mobility and crime at both the macro and micro levels. At the micro level, mobility is associated with delinquency, substance use, and other deviant behaviors among adolescents. However, it is possible that the relationship between mobility and delinquency may be due to selection on pre-existing differences between mobile and non-mobile youth in their propensity for delinquency, and prior studies have not adequately addressed this issue. Specifically, the families that are most likely to move are also the most disadvantaged and may be characterized by dynamics and processes that are conducive to the development of delinquency and problem behavior in their children. This study uses data from the National Longitudinal Survey of Youth 1997 to assess the impact of residential and school mobility between the ages of 12 and 17 on delinquency and substance use. Random effects models control for selection on both observed and unobserved differences. Results show that mobility and delinquency are indeed spuriously related. Implications for future research on mobility and outcomes are discussed.

Results 5.1. Descriptive findings Fig. 1 shows average residential and school mobility rates for NLSY97 sample members from ages 13 to 18. Since the data are weighted, these are nationally representative estimates of mobility rates for youth ages 12–16 on December 31, 1996. Fig. 1 shows that the rate of school mobility is considerably higher than the rate of residential mobility at every age. About 8–10% of 13 and 14 year olds in the sample have experienced a residential move and over 20% have changed schools. Less than 10% of 15 and 16 year olds have moved, and about 17% of them have changed schools. Adolescents change schools more frequently than they move, supporting the idea that school change often occurs for reasons other than a change of residence. With the exception of the period from ages 13 to 14, the rate of residential mobility declines with age.12 School mobility also exhibits a general decline with age.13 Residential and school mobility rates by age. Fig. 1. Residential and school mobility rates by age. Figure options Table 1 presents descriptive statistics (means and standard deviations) for all variables used in the analysis by mobility status. Although the unit of analysis in the statistical models is the person-year, the descriptive statistics in Table 1 give a cross-sectional slice of youth in the first year they appear in our analytic sample in order to ease interpretation. For the majority of youth (over 97%), these data refer to round 2 (1998). Because the sample sizes are slightly different for the delinquency and substance use outcomes, Table 1 uses the delinquency sample. Table 1 shows that about 11% of the sample report moving, while 19% report changing schools. It is also worth noting that there is a connection between the two types of mobility: over half (58%) of residential movers change schools, and one third (33.3%) of school changers also change addresses. As shown in previous research, youth who report ever moving or changing schools are more likely to be delinquent and use substances. Youth who report residential and school mobility are also more disadvantaged and come from households where parents have lower levels of education and lower incomes. Black and Hispanic youth are more likely to change schools, as are students with poor academic performance and lower test scores. Students who report changing schools are more likely to have been bullied as children and to have had sex at earlier ages. Table 1. Descriptive statistics for all variables used in the analysis in 1998, by residential and school mobility. Variable All Youth Residential Mobility School Mobility (N = 4890) Yes (N536) No (N = 4354) Yes (N = 1001) No (N = 3889) Mean or % S.D. Mean or % S.D. Mean or % S.D. Mean or % S.D. Mean or % S.D. Problem Behavior Delinquency 27.9% 37.0% 26.7% 34.9% 26.2% Substance use 53.3% 62.2% 52.1% 61.7% 51.2% Residential and School Mobility Residential move 11.2% 100.0% 0.0% 33.3% 5.9% School change 19.6% 58.1% 14.8% 100.0% 0.0% Demographics Age 14.99 0.91 15.05 0.92 14.99 0.91 14.92 0.93 15.01 0.91 Race/ethnicity White 71.0% 74.2% 70.6% 66.7% 72.1% Black 15.3% 13.7% 15.5% 17.6% 14.7% Hispanic 12.5% 10.9% 12.7% 13.9% 12.1% Other 1.3% 1.2% 1.3% 1.9% 1.1% 1996 Household income (Log) 10.39 1.46 10.08 1.55 10.43 1.44 10.15 1.44 10.45 1.45 Parental education (1997) 13.83 3.95 13.36 3.73 13.89 3.97 13.41 4.30 13.93 3.85 Residence Urban 69.7% 66.8% 70.0% 74.8% 68.4% Rural 30.3% 33.2% 30.0% 25.2% 31.6% Family structure Both biological parents 53.7% 30.9% 56.6% 37.8% 57.6% Other living situation 46.3% 69.1% 43.4% 62.2% 42.4% Household size 4.41 1.43 4.45 1.71 4.40 1.40 4.50 1.60 4.39 1.39 Unemployment rate in county 1.99 0.79 1.98 0.77 1.99 0.80 2.01 0.83 1.99 0.78 School Performance, Experiences, and Characteristics ASVAB math-verbal percentile score 50.06 29.06 44.59 28.06 50.75 29.11 40.89 27.78 52.30 28.93 8th grade GPA, 8-point scale 5.75 1.75 5.28 1.79 5.81 1.73 5.14 1.79 5.90 1.70 Track in high school General 71.8% 80.6% 70.7% 82.5% 66.2% College prep 25.4% 18.2% 26.3% 14.8% 28.0% Vocational 2.8% 1.3% 3.0% 2.6% 2.8% Grade in school relative to same-age peers −0.02 0.69 0.18 0.80 −0.05 0.67 0.17 0.74 −0.07 0.67 Victim of bullying as a child 19.5% 25.7% 18.7% 23.8% 18.4% School size 4.94 1.41 4.83 1.40 4.96 1.41 4.84 1.43 4.97 1.40 Student–teacher ratio 2.23 1.04 2.27 1.03 2.23 1.04 2.28 1.04 2.22 1.04 Individual Heterogeneity Years since first sexual intercourse 0.50 1.38 0.78 1.63 0.47 1.34 0.80 1.69 0.43 1.28 Survey Structure Weeks since last interview 85.15 12.63 89.79 12.50 84.56 12.52 87.98 12.10 84.45 12.65 S.D. = standard deviation. Note: Descriptive statistics are based on the delinquency sample of 4,890 for 1998. Table options 5.2. Predicting delinquency and substance use Table 2 shows results from the logistic regression models predicting delinquent behavior. These models indicate whether mobility in one year leads to a change in delinquency in the following year. Models 1–3 are not random effects models but rather conventional logistic regression models, which attempt to equate mobile and non-mobile youth by including controls for between-person differences in the form of observed covariates. Model 1 includes only a measure of residential mobility, and shows the expected positive and significant relationship between residential mobility and delinquency. In Model 2, we add the measure of school mobility and find that the coefficient for residential mobility is reduced but is still positive and significant. Therefore, both residential and school mobility are independently associated with delinquency. Table 2. Logistic regression models predicting delinquency (N = 4947, person-years = 14438). Model 1 Logit Move Only Model 2 Logit Move + Change Only Model 2 Logit Move + Change + Controls Model 3 Random Effects Logit Between Within Residential and School Mobility Residential move .55⁎⁎⁎ .37⁎⁎⁎ .21⁎⁎ .29⁎ .14 (.06) (.07) (.08) (.14) (.10) School change .42⁎⁎⁎ .20⁎⁎⁎ .37⁎⁎⁎ .09 (.05) (.06) (.11) (.07) Demographics Male .26⁎⁎⁎ .31⁎⁎⁎ (.04) (.06) Black −.26⁎⁎⁎ −.31⁎⁎⁎ (.07) (.08) Hispanic −.05 −.06 (.07) (.08) Other .03 .03 (.18) (.23) Log household income .02 .03 (.02) (.02) Highest parental education .01 .01 (.01) (.01) Urban .07 .16⁎ −.16 (.05) (.07) (.19) Living with both biological parents −.16⁎⁎ −.22⁎⁎⁎ −.03 (.05) (.07) (.18) Household size −.02 −.05⁎ .01 (.02) (.02) (.03) Unemployment rate .00 −.03 .03 (.03) (.04) (.06) School Performance, Experiences, and Characteristics ASVAB percentile score .00 .01⁎⁎⁎ (.00) (.00) 8th grade GPA, 8-point scale −.14⁎⁎⁎ −.18⁎⁎⁎ (.02) (.02) College prep track in h.s. −.02 −.11⁎⁎⁎ .06 (.05) (.09) (.07) Vocational track in h.s. .13 .06 .06 (.11) (.19) (.13) Grade in school relative to peers −.01 −.05 .03 (.03) (.05) (.06) Bullied by age 12 .32⁎⁎⁎ .50 (.05) (.07) School size .02 −.01 .07⁎ (.02) (.03) (.03) Student/teacher ratio −.02 −.02 −.07 (.03) (.04) (.05) Individual Heterogeneity Years since first intercourse .10 .17 −.24⁎⁎⁎ (.01) (.02)⁎⁎⁎ (.05) Lagged delinquency 1.63⁎⁎⁎ (.04) Constant −1.00⁎⁎⁎ −1.07⁎⁎⁎ −.54 −1.07 (.05) (.06) (1.04) (1.84) Standard errors shown in parentheses. Standard errors are survey corrected. Two-tailed tests. Note: All models also include dummy variable controls for survey year, age, region of residence, and birth year. All models also include a control for weeks elapsed since the last interview. Coefficients have been omitted to simplify presentation. ⁎⁎⁎ p < .001. ⁎⁎ p < .01. ⁎ p < .05. Table options Model 3 adds all of the observed controls, including the lagged measure of delinquency, and the coefficients for residential and school mobility are reduced but still highly significant. The odds of delinquency for youth who change residences are 23% [(exp(.21)) − 1) × 100] higher than for youth who do not experience residential mobility. The relationship between school mobility and delinquency is similarly strong, although the coefficient was reduced by about half. The odds of delinquency for youth who change schools are 22% [(exp(.20)) − 1) × 100] higher than for those who do not change schools. Therefore, both residential and school mobility still appear to have an influence on delinquency even in the face of a host of controls designed to capture preexisting differences between mobile and non-mobile youth, including a lagged measure of offending. These results largely replicate those of prior studies. However, Model 3 is still vulnerable to criticisms of selection bias, as there may be unobserved characteristics driving the relationship between mobility and delinquency. Model 4 is a random effects logistic regression model predicting delinquency. This model provides the key test of whether the relationship between mobility and delinquency is attributable to preexisting differences.14 We begin by examining the “between” coefficients, which give the between-person differences in delinquency for mobile and non-mobile youth averaged across all time periods. Consistent with the findings from Model 3, the between-person coefficients for both residential school mobility are positive and significant, suggesting an association between both moving residences and changing schools and delinquency at the between-person level. That is, the positive and significant between-person coefficients for both residential and school mobility tell us that youth who move residences or change schools are more likely to be involved in delinquency, on average, than their non-mobile counterparts. The between-person coefficients, however, do not tell us anything about whether moving residences or changing schools puts youth at risk for delinquency. Our main interest, therefore, is in the “within-person” coefficients, which give the change in delinquency (from no delinquency to any delinquency or vice versa) that follows from a change of residence or school and provides the best evidence of a causal effect of mobility. The within-person coefficient for residential mobility is half the size of the between-person coefficient and not significantly different from zero. This means that changing residences does not lead to a change in delinquency. Rather, the association between residential mobility and delinquency is attributable to unobserved between-person differences in delinquency. Youth who move are more likely to be delinquent, but not because of their mobility. A more formal test of the importance of individual unobserved heterogeneity can be performed by a statistical test of the difference between the deviation and mean coefficients. The difference between these two coefficients for school mobility is significant (p < .05), indicating that there are person-specific unobservables driving the school mobility–delinquency relationship. However, the difference between the two coefficients for residential mobility does not reach conventional standards for statistical significance (p < .31). While not our main focus, it is worth noting the within-person results for some of the time-varying control variables in Model 4. Few time-varying control variables have significant within-person effects on delinquency. While living with both biological parents is negatively associated with delinquency at the between-person level (p < .001), changes in family structure do not lead to delinquency at the within-person level. Family structure, aspects of parenting from before elementary school, such as the level of supervision and disciplinary style, and other sources of unobserved heterogeneity may be more important for delinquency in adolescence than changes in family structure during adolescence. Also, while the between-person coefficient for years since sexual intercourse is positive and significant (p < .001), the within-person coefficient is actually negative and significant (p < .001). This indicates that the association of sexual precocity and delinquency is driven by selection. Youth who have a predisposition to engage in delinquency are likely to be drawn into sex at earlier ages. However, once these delinquency-prone youth have sex, their delinquency is inhibited. While it is unclear why this would be the case, one possibility is that sex may lead to having a child, which in turn may increase social control and reduce delinquency ( Hope et al., 2003). Finally, the within-person coefficient for school size is positive and significant, indicating that increases in school size lead to delinquency (p < .05). 15 Turning to the effects of mobility on substance use, Table 3 presents the results from the logistic regression models. Model 1 shows the expected positive and significant relationship between residential mobility and substance use. For youth who move, the odds of using tobacco, alcohol, or marijuana are about 71.6 [(exp(.54) − 1) × 100]% higher than for non-movers. When school mobility is added in Model 2, the coefficient for residential mobility remains positive and highly significant, and school mobility is also strongly associated with substance use. The coefficients for both residential and school mobility remain significant in Model 3 when observed controls for selection are introduced, although both are reduced by roughly half. When we decompose our measures of residential and school mobility into their between-person and within-person components in the random effects model (Model 4), the results are consistent with those obtained when delinquency was the outcome: while the between-person coefficient is positive and statistically significant, indicating that youth who move residences or change schools are more likely to engage in substance use than their more non-mobile counterparts, the within-person coefficient is non-significant, indicating that changes in mobility are not associated with changes in substance use. In other words, the observed positive relationship between mobility and substance use is due to differences before moving residences or changing schools, not to residential or school mobility.16 Table 3. Logistic regression models predicting substance use (N = 4986, person-years = 14542). Model 1 Logit Move Only Model 2 Logit Move + Change Only Model 2 Logit Move + Change + Controls Model 3 Random Effects Logit Between Between Residential and School Mobility Residential move .54⁎⁎⁎ .41⁎⁎⁎ .22⁎⁎ .37⁎⁎ .06 (.06) (.07) (.08) (.14) (.08) School change .31*** .13* .28** .06 (.05) (.06) (.11) (.06) Demographics Male −.19⁎⁎⁎ −.26 (.04) (.06) Black −.78⁎⁎⁎ −1.09 (.07) (.08) Hispanic −.21** −.26 (.07) (.08) Other −.28 −.53 (.18) (.28) Log household income .05** .07 (.02) (.03) Highest parental education .01 .01 (.01) (.01) Urban .09 .14* −.08 (.05) (.07) (.14) Living with both biological parents −.17⁎⁎⁎ −.21** −.03 (.05) (.06) (.13) Household size −.07⁎⁎⁎ −.14⁎⁎* .03 (.01) (.02) (.03) Unemployment rate .03 −.01 .03 (.03) (.04) (.04) School Performance, Experiences, and Characteristics ASVAB percentile score .00 .00 (.00) (.00) 8th grade GPA, 8-point scale −.11⁎⁎⁎ −.19 (.01) (.02) College prep track in h.s. .06 −.13 .14** (.05) (.08) (.05) Vocational track in h.s. −.11 −.18 −.07 (.11) (.18) (.10) Grade in school relative to peers −.08* −.17⁎⁎⁎ −.02 (.03) (.04) (.05) Bullied by age 12 .10* .24 (.05) (.07) School size .00 −.05 .05 (.02) (.03) (.03) Student/teacher ratio −.03 −.04 −.04 (.02) (.04) (.04) Individual Heterogeneity Years since first intercourse .11⁎⁎* .25⁎⁎⁎ −.18⁎⁎⁎ (.02) (.03) (.04) Lagged substance use 2.20⁎⁎⁎ (.04) Constant −.26⁎⁎⁎ −.31⁎⁎⁎ −.49 −.94 (.05) (.05) (.98) (1.78) Standard errors shown in parentheses. Standard errors are survey corrected. Two-tailed tests. Note: All models also include dummy variable controls for survey year, age, region of residence, and birth year. All models also include a control for weeks elapsed since the last interview. Coefficients have been omitted to simplify presentation. ⁎⁎⁎ p < .001. ⁎⁎ p < .01. ⁎ p < .05. Table options The within-person results for the time-varying control variables are similar to those when delinquency was the outcome. One exception is that the within-person coefficient for college preparatory track in high school is positive and significant (p < .05), indicating that changing from a regular to a college prep curriculum leads to substance use. One possible reason for this finding is that while youth in a college prep track are no more or less likely to engage in substance use, for a substance use-prone youth, who is likely at risk for academic failure, being placed in a college track is likely to lead to stress and strain. This psychological distress may lead to alcohol or marijuana use as a way of coping.