Journal of Family Psychology 2014, Vol. 28, No. 6, 855-866
© 2014 American Psychological Association 0893-3200/14/$ 12.00 http://dx.doi.org/10.1037/a0038242
Income, Neighborhood Stressors, and Harsh Parenting: Test of Moderation by Ethnicity, Age, and Gender R. Gabriela Barajas-Gonzalez
Jeanne Brooks-Gunn
New York University School of Medicine
Teachers College, Columbia University
Family and neighborhood influences related to low-income were examined to understand their associ ation with harsh parenting among an ethnically diverse sample of families. Specifically, a path model linking household income to harsh parenting via neighborhood disorder, fear for safety, maternal depressive symptoms, and family conflict was evaluated using cross-sectional data from 2,132 families with children ages 5-16 years from Chicago. The sample was 42% Mexican American, 41% African American, and 17% European American. Results provide support for a family process model where a lower income-to-needs ratio is associated with higher reports of neighborhood disorder, greater fear for safety, and more family conflict, which is in turn, associated with greater frequency of harsh parenting. Our tests for moderation by ethnicity/immigrant status, child gender, and child age (younger child vs. adolescent) indicate that although paths are similar for families of boys and girls, as well as for families of young children and adolescents, there are some differences by ethnic group. Specifically, we find the path from neighborhood disorder to fear for safety is stronger for Mexican American (United States born and immigrant) and European American families in comparison with African American families. We also find that the path from fear for safety to harsh parenting is significant for European American and African American families only. Possible reasons for such moderated effects are considered. Keywords: parenting, neighborhood disorder, fear, poverty, stress
Low income and poverty strain the well-being of families in many ways: the worry associated with financial hardship, the struggle to make ends meet, and the inability to provide material goods for one’s children are just some of the stressors parents face (Duncan & Brooks-Gunn, 1997; Gershoff, Aber, Raver, & Len non, 2007). Another way that income is tied to family well-being and parenting is through the restriction it places on where families can afford to live, as the neighborhood in which a family resides is determined—to some degree— by a family’s resources as well as segregation (Massey, Gross, & Eggers, 1991). Families with limited incomes are more likely to live in low-income neighbor hoods (Fiy & Taylor, 2012), which are more likely to offer lower-quality resources and have higher rates of crime and disor der (Sampson, Raudenbush, & Earls, 1997) in comparison with higher income neighborhoods. The purpose of this study is to contribute to the discussion of how income influences families, by examining neighborhood characteristics and perceptions of these characteristics in the link between income and parenting across a diverse group of families. This work is informed by two theoretical frameworks: (a) the family stress model (Conger & Elder, 1994; Elder, 1999), which
posits that the psychological stress associated with economic hard ship strains family relations and disrupts parenting; and (b) social disorganization theory, which suggests that neighborhoods with a high proportion of poor residents, high residential mobility, and/or high ethnic heterogeneity are disadvantaged in comparison with other neighborhoods, as these characteristics are hypothesized to weaken residents’ attachment to the neighborhood and limit re sources with which to support local institutions (Shaw & McKay, 1969). These conditions in turn are associated with the emergence of disorder (Massey & Denton, 1993; Sampson et al., 1997; Wilson, 1987) which is related to higher levels of mistrust and fear among residents (Ross & Mirowsky, 2001; Wallace, 2012). There fore, we integrate these two theories and test a model that builds on some of the links in the family stress model by including an indirect path from income-to-needs ratio to family conflict via neighborhood disorder and fear for safety (see Figure 1). We also test whether parental report of fear is directly associated with harsh parenting, as mothers under stress (e.g., living with fear for safety) may be more likely to use punitive parenting practices (WebsterStratton, 1990). We examine whether, in addition to maternal depressive symptoms and family conflict, low income is associated with living in more disordered neighborhoods, which in turn is associated with increased levels of fear, family conflict and ulti mately, harsh parenting. We test our model with a large (N = 2,132), diverse sample of Mexican American, African American, and European American families with children (ages 5-16 years) who participated in the Project of Human Development in Chicago Neighborhoods (PHDCN). Although research examining the pro cesses Unking low-income to harsh parenting has been conducted with different ethnic groups, as discussed below, most studies have
This article was published Online First November 10, 2014. R. Gabriela Barajas-Gonzalez, Department of Population Health, New York University School of Medicine; Jeanne Brooks-Gunn, Department of Human Development, Teachers College, Columbia University. Correspondence concerning this article should be addressed to R. Ga briela Barajas-Gonzalez, One Park Avenue, 7th floor, Center for Early Childhood & Development, Department of Population Health, New York, NY 10016. E-mail: [email protected]
855
856
BARAJAS-GONZALEZ AND BROOKS-GUNN
Figure 1. Model of indirect influence of income-to-needs ratio on harsh parenting via neighborhood disorder and fear for safety.
been based on smaller, nonrepresentative samples that do not include multiple ethnic/immigrant groups in one study, making it difficult to make comparisons about the processes operating for each group. The diversity of the PHDCN sample allows us to address this.
Low Income and Multiple Sources of Stress Considerable progress has been made in documenting the im pact of economic hardship on family functioning in European American and African American families (Conger, Wallace, Sun, Simons, McLoyd, & Brody, 2002; McLoyd, 1998), and more recently, in Mexican American families as well (Barrera et al., 2002; Loukas, Prelow, Suizzo, & Allua, 2008; Parke et al., 2004). Studies indicate that low-income and economic hardship strain parent mental health, increasing the likelihood of family conflict and reducing parents’ ability to interact with family members in a responsive and nurturing manner. Families with low incomes are more likely to live in disadvan taged neighborhoods (Fry & Taylor, 2012; Sampson & Sharkey, 2008) and face daily stressors associated with living in such neighborhoods— over and above having low family income— such as crime, harassment, and danger (Hill, Ross, & Angel, 2005; Wandersman & Nation, 1998). Perceptions of the neighborhood environment as violent, unsafe, or highly disordered can increase feelings of distress, both directly and indirectly, through increased feelings of fear, powerlessness, and a diminished sense of control (Brunton-Smith & Sturgis, 2011; Hill & Maimon, 2013; Ross & Jang, 2000). The stress imposed by such adverse neighborhood conditions has been found to generate angst above and beyond the effects of the individual’s own personal stressors, straining mental health (Cutrona, Wallace, & Wesner, 2006; Hill et al., 2005; Ross & Mirowsky, 2001; Wallace, 2012). Hill and colleagues (2005), for instance, used data from the Welfare, Children, and Families project (a household-based, stratified random sample of 2,402 poor mothers in low-income neighborhoods in Boston, Chicago, and San Antonio) to test whether neighborhood conditions influenced women’s health over and above individual characteristics. The authors distinguished individual disadvantage from neighborhood problems by including sociodemographic characteristics of indi viduals (e.g., education, economic well-being, and employment status) that correlated with the likelihood of living in a neighbor hood with high levels of disorder and with poor health. Women who lived in neighborhoods where they reported a lot of problems with drugs, crime, abandoned buildings, and unsafe streets had
significantly worse self-reported health than those who reported more order in their neighborhoods. Notably, the authors found no evidence of subjectivity bias— as people who were more distressed were not likely to report more disorder. In addition to being associated with mental health, neighbor hood conditions have also been linked to parenting. Studies, though varied with respect to the number and quality of family level characteristics accounted for, have found that parents who reside in dangerous neighborhoods report more controlling parent ing practices, more need for supervision, and greater use of harsh disciplinary techniques in comparison to similar parents in less dangerous neighborhoods (Gutman, McLoyd, & Tokoyawa, 2005; Kotchick, Dorsey, & Heller, 2005; Pinderhughes, Nix, Foster, Jones, & The Conduct Problems Prevention Research Group, 2001). Thus, it is possible that neighborhood disorder undermines the ability of families to function effectively, as parents who live in dangerous neighborhoods may be more likely to experience chronic tension and stress, increasing the likelihood of disciplining and parenting reactively (Hill & Herman-Stahl, 2002; McLoyd, 1990). Recently, several studies have tested the relations among lowincome, neighborhood stress, and family functioning with African American families (Gutman et al., 2005; Kotchick et al., 2005), Mexican American families (Gonzalez et al., 2011; White, Roosa, & Zeiders, 2012; White, Zeiders, Gonzales, Tein, & Roosa, 2013), and European American families (Deng et al., 2006). Kotchick and colleagues (2005), for example, interviewed 123 African Ameri can single mothers at several time points to understand whether exposure to neighborhood stress (e.g., presence of gangs, drug dealing, and shootings) was related to psychological distress (e.g., depression, anxiety) and, in turn, compromised parenting. Findings revealed that in fact, neighborhood stress was related to greater psychological distress, predicting less positive parenting practices 15 months later. Similarly, Deng et al. (2006) found that perceived neighborhood quality mediated the association between neighbor hood disadvantage and parent-child conflict among 189 lowincome European American and Mexican American mothers. Of interest to the authors, Gonzalez et al. (2011) and White et al. (2012) only found moderated associations between perceptions of neighborhood danger/risk and harsh parenting in their sample of 750 Mexican American families. This may have been, in part, because no measures of mother’s psychological well-being or family conflict were included in the models they tested. We build on the previous literature by including an indicator of maternal depressive symptoms and an assessment of family conflict—two key constructs that have been theorized to influence parenting, but that have not been consistently included in previous studies. More over, because we have a large sample of African American, European American, and Mexican American families, we are able to test for possible moderation by ethnicity and immigrant status, as some studies have suggested that that there may be there may be race-ethnic differences in how disorder is perceived (ChristieMizell et al., 2003; Sampson & Raudenbush, 2004). Sampson and Raudenbush (2004) for example, found that European Americans reported more disorder than African Americans on the same block group of a neighborhood. They posited that African Americans may have a higher threshold for perceiving disorder in comparison with European Americans (assuming African Americans had been exposed to more disorder in the past). Relatedly, immigrant moth-
NEIGHBORHOOD STRESSORS AND PARENTING
ers, especially those bom in countries where poverty is more common, may have been raised with expectations of neighbor hoods that differ from those of people born in the United States. Mothers raised in countries where poverty is more common may be less stressed by adverse neighborhood conditions in the United States, especially if the current conditions are better than the ones in which they were raised. Baccallao and Smokowski (2007) for example, conducted qualitative interviews with undocumented Mexican families and found that these immigrant families consid ered the most basic living conditions in the United States a marked improvement in comparison with the chronic poverty they endured in Mexico. We test for the possibility that neighborhood charac teristics influence mothers differentially by disaggregating our Mexican American sample by immigrant status and comparing first (foreign-bom mother and foreign-born child) and second (foreign-born mother, United States bom child) generation status families (“Mexican immigrant”) to third generation and beyond (United States bom mother, United States born child) families (“Mexican American”) and testing for moderation across the Mex ican immigrant, Mexican American, African American, and Euro pean American subgroups. Based on the literature reviewed above, we hypothesized that the association between neighborhood dis order and fear for safety may be stronger for European American and Mexican American families than for African American and Mexican immigrant families. In addition to moderation by ethnic/immigrant group, we test for moderation by gender and examine whether the model holds for mothers of boys versus mothers of girls. A handful of studies examining the association between economic hardship, maternal depression, and parenting have tested for moderation by gender (Conger et al., 1992, 1993; Dennis, Parke, Coltrane, Blacher, & Borthwick-Duffy, 2003; Gutman, McLoyd, & Tokoyawa, 2005; Mistry, Vandewater, Huston, & McLoyd, 2002) with some finding no evidence of moderation (Mistry et al., 2002) and others finding differences for mothers of boys versus girls (Conger et al., 1992; Conger et al., 1993; Dennis et al., 2003). For example, in a series of studies analyzing cross-sectional data from 7th Grade European American boys (Conger et al., 1992) and then girls (Conger et al., 1993), Conger and colleagues found a direct association between caregiver depressed mood and hostile parenting among adolescent boys, but not among girls. Conger et al. (1993) posited that one explanation for this may be that girls, being less oppositional and disruptive than boys, were perhaps less likely to elicit harsh parenting from a distressed caregiver. Similarly, in their study with 56 immigrant Latina mothers of children ages 4-13 years, Dennis and colleagues (2003) found measures of maternal economic hard ship and depression to be significantly correlated for mothers of boys but not for mothers of girls. Mistry et al. (2002) however, found no evidence of moderation by gender in their test of the family stress model with 419 African American, Hispanic, and European American families of children ages 5 to 12 years. With regards to the links between disorder, fear for safety, and parent ing, there is a scarcity of studies examining gender differences in the association between neighborhood context and parenting. The limited literature that does exist suggests that boys may be more vulnerable to the detriments of living in impoverished neighbor hoods than girls, in part because girls may be socialized to spend more time at home rather than in their neighborhoods (ClampetLunduist, Dyncan, Edin, & King, 2006; Kroneman, Loeber, &
857
Hipwell, 2004). We are aware of one study that has examined the influence of neighborhood disadvantage on harsh parenting by gender. That study, conducted by Leventhal and Books-Gunn (2005), found that contrary to expectation, moving out of public housing in high-poverty neighborhoods into low poverty neigh borhoods was associated with greater parental harshness toward girls, but not boys. It was posited that the act of moving itself, and the disruption to social networks, may have been particularly detrimental to the mother-daughter relationship. Given the dearth of research concerning gender differences in the association be tween low-income, neighborhood context, and family functioning, no specific hypotheses regarding gender differences are made. We also take advantage of the wide age range of the children in our sample (5-16 years of age), and examine moderation by age group, testing whether the model holds for children who are in adolescence (ages 11-16; 44% of our sample) versus children who are younger (ages 5-10; 56% of our sample). To our knowledge, no studies of the FSM with a broad age group of children have tested for moderation by age group, perhaps because of a limited sample size. We address Conger & colleagues’ (2010) recommen dation to test whether the model is equivalent for families with younger, versus older, children in our sample. We posit that the model may be more salient for families with adolescents, as adolescence may bring additional stressors to family dynamics (e.g., greater parent-child conflict as child seeks independence from parents). Moreover, in comparison with younger children, adolescents may spend more time outside their home, which could cause parents to be more cognizant of neighborhood stressors.
Method Design and Sample This study uses two independent sources of data collected by the PHDCN, an interdisciplinary study of youth, families and neigh borhoods. Family level data are drawn from the first wave of the Longitudinal Cohort Study (PHDCN-LCS); neighborhood disor der data are drawn from the Community Survey (PHDCN-CS).
Longitudinal Cohort Survey Participants were drawn from a probability sample designed to capture the diversity of Chicago’s neighborhoods (see www.icpsr .umich.edu/icpsrweb/PHDCN/sitemap.jsp for a detailed descrip tion of PHDCNs design, sampling, and method). Data for 847 census tracts (from the 1990 U.S. Census) comprising the city of Chicago were combined to create 343 neighborhood clusters (NCs) that included two to three census tracts (—8,000 residents). A stratified probability sample of 80 neighborhood clusters (crossclassified by racial/ethnic composition and SES) was then drawn from the 343 neighborhood clusters. Households within these 80 NCs were randomly selected and screened for eligibility; —1,000 children/primary caregivers within each of seven age cohorts (birth, 3, 6, 9, 12, 15, and 18 years) were interviewed in their homes (N = 6,226), rendering a sample representative of children in the city of Chicago in 1995.
Community Survey The goal of the Community Survey (CS) was to gather in-depth information about the NCs from residents who were not partici-
BARAJAS-GONZALEZ AND BROOKS-GUNN
858
pants in the LCS, so as to obtain reports of neighborhood charac teristics independent of the study participants. These data were collected between the years 1994-1995 (at approximately the same time data was being collected for the longitudinal cohort survey) and involved a three-stage sampling strategy. First, city blocks were randomly selected from the 343 NCs. Second, house holds within blocks were randomly selected, and third, one adult from each household was randomly chosen to complete the survey. In total, 8,782 adults provided information on the characteristics and social processes of their neighborhood. Variables were aggre gated to neighborhood-level averages. Every NC in the LCS was represented in the CS (see http://www.icpsr.umich.edu/icpsrweb/ PHDCN/sampling.jsp). The current study used Wave 1 data on children from cohorts 6, 9,12, and 15 (i.e., those who were roughly 6, 9, 12, or 15 years of age at Wave 1; N = 3,324). Respondents who were not the child’s biological mother (n = 523 respondents who were either the child’s biological father, grandmother, or other relatives) or were missing information as to their relationship to the child (n = 29) were excluded from the sample. Mothers who were missing birth country information (n = 19) and mothers who did not identify as African American, European American, or Mexican were ex cluded from the sample (n = 621 identifying as Puerto Rican, Central and South American, Asian, African, European and Mid dle Eastern). Thus, the focus in this study was on the African American (AfAm), European American (EuroAm), and Mexican American families (United States bom and foreign bom mothers; MexUS, Mexlm, respectively) who were interviewed at Wave 1 (n = 2,132; see Table 1 for descriptive statistics). M easures The measure of neighborhood disorder was drawn from the CS, which occurred concurrent in timing to the Wave 1 LCS, and was based on reports from residents living in the same neighborhood but not participating in the study. All maternal and family char acteristics were drawn from the LCS and based on mother report. As part of the LCS, a demographic and behavioral information
interview was administered to the primary caregiver. Demographic information on education, race, income, birth country, and marital status was obtained. In addition, the primary caregiver completed structured interviews and self-report questionnaires assessing a range of health and behavioral indicators (see http://www.icpsr.umich.edu/ icpsrweb/PHDCN/instruments.jsp#longitudinal for greater measure detail). The items, means, and SDs for the independent variables of interest are presented in Table 1. Correlations among all variables of interest are presented in Table 2. Income-to-needs ratio. Annual family income and family size were used to create an income-to-needs ratio, by dividing the reported total annual family income by the official poverty thresh old for the respective household size in 1995. An income/needs ratio of one or less signifies poverty status. Lower family income has been used as a proxy for economic pressure (Lee, Anderson, Horowitz, & August, 2009). In the present sample, the mean income-to-needs ratio was 1.59 (SD = 1.26), with significant variation across subgroups (see Table 1). Maternal self-reported depressive symptoms. One item as sessing whether the mother suffered from symptoms of depression was drawn from the Family Mental Health and Legal History interview, based on the Family History Assessment Module (Janca, Bucholz, & Janca, 1992). Mothers were asked, “Has any one mentioned ever suffered from depression, that is, they have felt so low for a period of at least two weeks that they hardly ate or slept, or couldn’t work or do whatever they usually do?” There were 233 mothers who indicated that they had suffered from such a condition (0 = no; 1 = yes). Fear for safety. Three questions from the Exposure to Vio lence interview (Earls, Brooks-Gunn, Raudenbush, & Sampson, 1994) were used to construct a measure of fear for safety. At Wave 1, caregivers were asked about their fear in certain situations and their exposure to several different types of violent acts (e.g., “Are you afraid child may be hurt by violence in neighborhood?” “Are you afraid child may be hurt by violence in front of house?”). Response options for each question were dichotomous (0 = no; 1 = yes). The three items were summed so that higher scores
Table 1 Descriptives fo r Whole Sample and by Ethnic/lmmigrant Subgroup
Ml (SD)
Mexlm (n = 787) M (SD)
MexUS (n = 107) M (SD)
AfAm (n = 881) M (SD)
EuroAm (n = 357) M (SD)
1.59(1.26) 1.88 (0.34) 1.68(1.18) 1.20(1.41) 6.83 (6.40) 43.6% 48.7% 11.0%
1.18(0.86) 2.00 (.028) 1.91 (1.18) 0.86(1.22) 5.66 (5.71) 41.8% 48.9% 7.9%
1.71 (1.34) 1.90 (0.30) 1.86(1.07) 1.57(1.52) 7.40 (7.79) 41.1% 39.3% 9.5%
1.46(1.27) 1.91 (0.32) 1.61 (1.16) 1.44(1.46) 7.51 (6.38) 46.0% 48.7% 13.2%
2.72 (1.33) 1.56 (.031) 1.25(1.14) 1.24(1.50) 7.54(7.16) 42.6% 51.0% 13.0%
Total (.N = 2,132)
Income-to-needs ratiob,c,e'f,s Neighborhood disorderb,c,e-f Fear for safetyb'c,e,f Family conflictb'c,d Harsh parentingb-c,d Adolescent Male child Maternal depressive symptoms11
ANOVA/Pearson x2 F(3, F(3, F(3, F(3, F(3,
1825) = 135.67, p = .000 2128) = 170.76, p = .000 2098) = 28.76, p = .000 2121) = 27.06, p = .000 2116) = 14.01, p = .000 X2(3) = 4.58, p = .205 X2(3) = 3.46, p = .326 X2(3) =-13.50 , p = .004
Note. MAIm = Mexican immigrant; MAUS = Mexican American; AfAm = African American; EuroAm = European American. Mexican immigrant mothers had been in the United States an average of almost 15 years (M = 14.71, SD = 6.43). a Denotes statistically significant difference between groups at the .05 level. b Post hoc test indicates statistically significant difference between MAIm and MAUS at the .05 level. c Post hoc test indicates statistically significant difference between MAIm and African Americans at the .05 level. d Post hoc test indicates statistically significant difference between MAIm and European Americans at the .05 level. e Post hoc test indicates statistically significant difference between MAUS and African Americans at the .05 level. f Post hoc test indicates statistically significant difference between MAUS and European Americans at the .05 level. g Post hoc test indicates statistically significant difference between African Americans and European Americans at the .05 level.
NEIGHBORHOOD STRESSORS AND PARENTING
859
Table 2 Bivariate Correlations fo r Study Variables o f the Project o f Human Development in Chicago Neighborhoods (N = 2,132)
1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
*p < .05.
" p < .01.
1
2
3
4
5
6
1.00 -.46*** -.18*** -.15*** -.0 2 -.05*
1.00 .25*** .06** .01 .01
1.00 .16*** .12*** .10***
1.00 .33*** .15***
1.00 .15***
1.00
***p < .001.
indicated greater fear for safety. The internal consistency for these items was adequate (Cronbach’s a = .73; with alphas ranging from .65 (MexUS) to .76 (Mexlm)). Mean perceived lack of safety for this sample was 1.68 (SD = 1.18), with scores ranging from 0-3.00. Family conflict. The conflict subscale of the Family Environ ment Scale (FES; Moos & Moos, 1981) was used to assess family conflict. Caregivers reported whether each of five statements ac curately described their families (e.g., “family members often criticize each other,” “ members fight a lot”; 0 = false, 1 = true). Responses were summed across items, such that higher scores indicated greater conflict. Scores ranged from 0-5.00 (M = 1.20, SD = 1.41). Research supports the construct, concurrent and predictive validity of the FES (Moos, 1990). In the present study, the FES conflict scale demonstrated adequate internal reliability (Cronbach’s a = .70, with as ranging from .66 (Mexlm) to .75 (EuroAm)). Similar a levels have been reported by other studies of Latino families of varying acculturation levels using this measure (see Miranda, Estrada, & Firpo-Jimenez, 2000). Harsh parenting. Harsh parenting is characterized by paren tal inability to control emotions during interactions with their child, thereby yelling, threatening, or using physical aggression. Harsh parenting was assessed using seven items from the Conflict Tactics scale (CTS; Straus, 1979). Caregivers were asked how many times they had behaved in a certain manner when there was a problem with the child (e.g., “In the past year, when having a problem with child, how many times adult insulted/swore at child?” options: 0 = never, 1 = once, 2 = twice, 3 = 3-5 times, 4 — 6 -1 0 times, 5 = 11-20 times, 6 = more than 20 times). The items were summed so that higher scores indicated more aggres sion toward the child. The internal consistency for these items was adequate (Cronbach’s a = .73; with as ranging from .68 (AfAm) to ,76(EuroAm)). Mean mother-to-child aggression was 6.83 (SD = 6.42) with scores ranging from 0 to 39.00. The CTS has been used with ethnically diverse families (Lee, Brooks-Gunn, McLanahan, Notterman, & Garfinkel, 2013), including Mexican mothers in Mexico (Frias-Armenta & McCloskey, 1998). Neighborhood disorder. We used the neighborhood disorder scale from the Community Survey that consists of 12 items as sessing both physical disorder (e.g., “how much of a problem is graffiti in your neighborhood?” options: 3 = a big problem-, 2 = somewhat o f a problem', 1 = not a problem) and social disorder (e.g., “how much of a problem is groups hanging out causing trouble?” options: 3 = a big problem-, 2 = somewhat o f a problem', 1 = not a problem). The consistency of this scale is adequate
(Cronbach’s a = .70). See Sampson & Raudenbush (2004, p. 324) for greater detail about the creation and reliability of the neigh borhood disorder scale. A nalytic Strategy Analyses. Chi-square tests were used to test for significant group mean differences on categorical variables (e.g., child gen der); one-way analysis of variances (ANOVAs) with post hoc tests were used to test for group mean differences on continuous vari ables (e.g., income-to-needs ratio, family conflict, etc.) using SPSS 16. Variance and group sizes were considered to determine which post hoc procedure to use. We tested the hypothesized model using Mplus 6.0 (Muthen & Muthen, 2010). Maximum likelihood estimator with robust SEs (MLR estimation), TYPE = COMPLEX, and the CLUSTER command were used to account for the nested structure of our data (participants nested within neighborhoods). MLR estimates are robust to nonnormality and nonindependence of observations when used with the TYPE = COMPLEX and CLUSTER proce dure. This method of analysis yields standard errors of path coef ficients that are adjusted for neighborhood clustering (Muthen & Muthen, 2010) giving more conservative parameter estimates. Model fit was assessed using y2 fit statistics, the root mean square error of approximation (RMSEA; Browne & Cudeck, 1993), the comparative fit index (CFI; Bentler, 1990), and the Tucker Lewis Index (TLI). Good model fit is indicated by a nonsignificant x2 statistic, CFI and TLI greater than .90 (Bentler, 1990), and RM SEA less than .05 (less than .08 is adequate; Browne & Cudeck, 1993). Because the x2 statistic is affected by sample size, and larger sample sizes (more than 200 cases) may lead to a significant X2 even though the differences between the observed and predicted covariances are modest (Kline, 2005), the x2 ratio (x2/df), which adjusts for model complexity is reported. In general, a x2 ratio between 1 and 3 indicates good fit (Arbuckle & Wothke, 1999). We tested the model for subgroup moderation by constraining all paths to be equal across subgroups, and comparing the con strained versus the unconstrained models using the \ 2 difference test with the Satorra-Bentler scaled x2 (Satorra & Bentler, 1999), where a nonsignificant Ax2 indicates that the constrained model fits the data just as well and is more parsimonious than the unconstrained model. If the Ax2 was significant, however, we concluded that moderation had occurred, and proceeded to free one path at a time to identify paths that were significantly different across groups. We used the same procedure outlined above to
BARAJAS-GONZALEZ AND BROOKS-GUNN
860
check for moderation by gender and by age group. We assessed the significance of indirect paths using the MODEL INDIRECT com mand in Mplus, which computes delta method standard errors for the indirect effects (Muthen & Muthen, 2010). Treatment of missing data. Less than 5% of cases were missing data on fear for safety, conflict, depressive symptoms, and harsh parenting, but roughly 14% of cases were missing data on income-to-needs ratio. The full information maximum likelihood (FIML) method, which estimates model parameters and SEs using all available raw data, was used for all analyses. FIML has been shown to be more effective compared with other estimation ap proaches when dealing with missing data in structural equation models (Enders & Bandalos, 2001) and precludes the need for imputation or the deletion of incomplete cases.
Results Descriptive information on the sample is reported in Table 1 (by ethnic/immigrant group), Table 3 (by gender) and Table 4 (by age group). As shown in Table 1, nearly 42% of the sample was Mexican American (n = 894) and 41% of the sample was African American in = 881). Roughly half of the households in this study consisted of families in which both biological parents of the child were married to each other in = 1,030). Almost half the sample did not graduate from high school. There were no significant differences in child age or gender across the four ethnic subgroups. There were, however, significant group differences on all other variables of interest across the four ethnic subgroups. For greater detail, see Table 1. There were no significant differences on variables of interest between families of boys versus families of girls (see Table 3). There were some differences, however, when examined by age group. Specifically, families of adolescents had higher mean income-to-needs ratio (M = 1.67, SD = 1.27 vs. M = 1.52, SD = 1.25; t( 1,827) = —2.49, p = .013), reported greater fear for safety (M = 1.78, SD = 1.15 vs. M = 1.59, SD = 1.21; /(2,011) = —3.60, p = .000), and greater harsh parenting (M = 7.33, SD = 6.61 vs. M = 6.44, SD = 6.24; r( 1,923), -3 .1 5 = p = .002) in comparison with families of young children (see Table 4). Most variables of interest were significantly correlated (see Table 2). As expected, income-to-needs ratio was negatively associated with neighborhood disorder, fear for safety, and family conflict. Fear for safety and family conflict were both significantly associ
ated with each other, with maternal depressive symptoms, and with harsh parenting. See Tables 3 and 4 for bivariate correlations by gender and by age group. We will discuss the results of our tests for moderation by gender and by age first as those are most straightforward. To test for moderation by gender, we compared the fit of our unconstrained path model (X2/df = 3.62, CFI = .978, TLI = .918, RMSEA = .054, 90% Cl [.033, .075]), to one where all parameters were held to be equal for boys and girls (x2ldf = 2.18, CFI = .978, TLI = .963, RMSEA = .036, 90% Cl [.020, .051)]; constraining all paths of interest to be equal across the two groups did not result in worse model fit (Ax2 = 9.16(10), p = .516), suggesting paths were similar for families of male and families of female children. To test for moderation by age, we conducted the same procedure, and compared the fit of the unconstrained path model (x2/df = 4.38, CFI = .970, TLI = .889, RMSEA = .061, 90% Cl [.041, .082]), to one where all parameters were held to be equal for young children and adolescents (x2/d f = 2.73, CFI = .966, TLI = .943, RMSEA = .044, 90% Cl [.029, .058]); constraining all paths of interest to be equal across the two groups did not result in worse model fit (Ax2 = 13.76(10), p = .184), suggesting paths were similar for families of young children and adolescent children. All paths of interest in the model, with the exception of the path from income-to-needs ratio —> maternal depression, were signifi cant across gender and age group (see Appendix A, Table A l). Lower income-to-needs ratio was significantly associated with more family conflict, which was in turn, associated with greater harsh parenting. As expected, lower income-to-needs ratio was also significantly associated with greater neighborhood disorder, which was associated with greater fear for safety. Fear for safety was associated with both greater family conflict and greater en dorsement of maternal depressive symptoms. Maternal depressive symptoms and family conflict were both in turn, associated with greater harsh parenting. Test of indirect effects indicate that the indirect paths from income-to-needs ratio to harsh parenting via family conflict and via neighborhood disorder were significant across gender and age groups (see Appendix A, Table A2). The same procedure was followed to test for moderation by ethnic group. We compared the fit of our unconstrained path model (X2l df = 1.88, CFI = .977, TLI = .915, RMSEA = .044, 90% Cl [.018, .068]), to one where all parameters were held to be equal
Table 3
Means, SDs, and Correlations for Study Variables by Gender (Boys, n = 1,038; Girls, N = 1,094) Measure 1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
M
SD
1
2
3
4
5
6
1.61 (1.56) 1.88 (1.88) 1.71 (1.64) 1.18(1.22) 6.98 (6.68) 0.12(0.10)
1.26(1.26) 0.34 (0.34) 1.18(1.18) 1.38(1.45) 6.51 (6.32) 0.32 (0.30)
1.00 -.475*** -.228*** -.159*** —.061 + —.059f
-.452*** 1.00 .287*** .063* .018 .021
-.141*** .216*** 1.00 .183*** .137*** .117***
-.144*** .064* .139*** 1.00 .296*** .146***
.017 -.005 .100** .370*** 1.00 .137***
-.038 -.020 .074” .163' .154' 1.00
Note. Correlations for boys are reported below the diagonal and correlations for girls are reported above the diagonal. Girls’ means and SDs are reported in parentheses. Sample size for boys ranged from 902-1,031 and for girls ranged from 920-1,094 because of list-wise deletion. All other analyses utilized the full sample (N = 2,132) and accounted for missing data using full information maximum likelihood, t test indicated no significant difference in means between boys and girls. > < . 1 0 . *p < .05. * > < .0 0 1 .
NEIGHBORHOOD STRESSORS AND PARENTING
Table 4 Means, SDs, and Correlations fo r Study Variables by Age Group (Young Child, n Measure 1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
=
861
1,202; Adolescent, N
=
930)
M
SD
1
2
3
4
5
6
1.52(1.67) 1.89(1.87) 1.59(1.78) 1.17(1.24) 6.44 (7.33) 0.11 (0.12)
1.25 (1.27) 0.34 (0.34) 1.21 (1.15) 1.40(1.42) 6.24 (6.61) 0.31 (0.32)
1.00 -.473*** -.214*** -.149*** —.056+ -.047
-.448*** 1.00 .285*** .038 .023 -.016
-.153*** .212*** 1.00 .151*** .094** .105***
-.154*** TOO** .168*** 1.00 .300*** .151***
.011 -.009 .141*** .373*** 1.00 .150***
-.051 .024 .082* .157*** .140*** 1.00
Note. Correlations for young children are reported below the diagonal and correlations for adolescents are reported above the diagonal. Adolescents’ means and SDs are reported in parentheses. Sample size for young children ranged from 1,021-1,202 and for adolescents ranged from 808-928 because of list-wise deletion. All other analyses utilized the full sample (N = 2,132) and accounted for missing data using full information m a d m . i m likelihood. Compared with families with young children, families with adolescents had higher mean income, r(l,827) = -2.49, p = .013; also higher fear for safety r(2,011) = -3.60, p = .000; higher harsh parenting r(1,923) = -3.15, p = .002. *p < .05. **p < .001.
across the four ethnic groups (\ 2/d f = 1.59, CFI = .956, TLI = .943, RMSEA = .036, 90% Cl [.018, .051]); constraining all paths of interest to be equal across groups resulted in worse fit (Ax2 = 44.16(30), p = .046), suggesting subgroup moderation. Thus, we proceeded to test for subgroup moderation, freeing one path of interest at a time, to see where moderation was occurring. Freeing the path from disorder-* fear, the path from fear —» harsh parent ing, and the path from fear —*• family conflict resulted in nonsig nificant Satorra-Bentler \ 2 tests indicating subgroup moderation in those paths. Therefore, we tested a model where these three paths where unconstrained, whereas the rest of the paths were con strained to be equal across subgroups. Model fit indices indicated that this model fit the data well and was more parsimonious than our baseline model (x2 ratio = 1.36, CFI = .978, TLI = .965, RMSEA = .028, 90% Cl [.000, .046]; (Ax2 = 22.80 (21), p = .354). Upon examining the parameter estimates in the path from fear —* family conflict, we noted that they were relatively similar for Mexican immigrant, Mexican American, and African Ameri can families in comparison with European American families. Therefore, we constrained the path from fear —* family conflict to be equal for Mexican immigrant, Mexican American, and African American families, but left it free for European American families. Doing so resulted in a more parsimonious model with similar fit (X2 ratio = 1.36, CFI = .977, TLI = .965, RMSEA = .028, 90% Cl [.000, .046]).
As shown in Table 5, across the four groups, lower income-toneeds ratio was significantly associated with greater endorsement of maternal depressive symptoms, which was in turn, significantly associated with harsh parenting. Lower income-to-needs ratio was also significantly associated with more family conflict, which was in turn, significantly associated with harsh parenting. These indi rect paths from income-to-needs ratio to harsh parenting were significant across all ethnic groups (see Table 6). The additional paths from income-to-needs ratio to neighborhood disorder, from disorder to fear, and from fear to maternal depressive symptoms were all significant across ethnic groups. As expected, lower income-to-needs ratio was significantly associated with more neighborhood disorder. Neighborhood disorder, in turn was sig nificantly associated with fear for safety, though— consistent with moderation analyses—the association was stronger for the immi grant and United States born Mexican American mothers and European American mothers ((3MAim = -239; |3MAUS = .202; PEuroAm = -314) than for the African American mothers ((3 ^ ,^ = .094). The path from fear to family conflict was significant for every subgroup, but moderation tests indicated that the path was particularly strong for European American mothers (PEuroAm = .216) relative to the immigrant and United States bom Mexican American mothers and African American mothers (PMAim = 1-39; P m au s = °99; PEuroAm = 119). Fear was significantly associated with maternal depressive symptoms across every sub-
Table 5 Standardized Parameter Estimates fo r Final Path Model Mexlm (n = 787)
INC —* Dep INC -»• Con INC Dis Dis —» Fear Fear —* Con Fear —* Dep Con -> Harsh Dep -» Harsh Fear —* Harsh
MexUS (n = 107)
PCS©
P
VISE)
-.051 (.02) -.146 (.02) -.265 (.04) .239 (.05) .139 (.03) .121 (.02) .270 (.03) .080 (.03) .051 (.04)
.019 .000 .000 .000 .000 .000 .000 .002 .190
-.075 (.03) -.181 (.03) -.384 (.05) .202 (.09) .099 (.22) .101 (.02) .252 (.03) .064 (.03) .027 (.11)
AfAm (n = 881) P
(3 (SE)
.023 .000 .000 .024 .000 .000 .000 .014 .810
-.0 6 4 (.03) -.187 (.03) -.348 (.03) .094 (.04) .119 (.02) .101 (.02) .293 (.03) .089 (.03) .081 (.04)
EuroAm (n = 357) P
P(5£)
P
.015 .000 .000 .033 .000 .000 .000 .002 .027
-.063 (.03) -.185 (.03) -.390 (.04) .314 (.07) .216 (.07) .092 (.02) .267 (.03) .082 (.03) .187 (.07)
.015 .000 .000 .000 .002 .000 .000 .003 .005
Note. INC - income-to-needs ratio; Dep = maternal depressive symptoms; Dis = neighborhood disorder; Con = family conflict; Fear = fear for safety; Harsh = harsh parenting. Though not shown, the direct relation between income-to-needs ratio and harsh parenting was tested in all models, but was not observed to be significant.
BARAJAS-GONZALEZ AND BROOKS-GUNN
862
[5
ON^-sfmONOQcn < pNo X X ) — X ! XS X!
c M W) W) a> (u 5 ‘
© 2014 American Psychological Association 0893-3200/14/$ 12.00 http://dx.doi.org/10.1037/a0038242
Income, Neighborhood Stressors, and Harsh Parenting: Test of Moderation by Ethnicity, Age, and Gender R. Gabriela Barajas-Gonzalez
Jeanne Brooks-Gunn
New York University School of Medicine
Teachers College, Columbia University
Family and neighborhood influences related to low-income were examined to understand their associ ation with harsh parenting among an ethnically diverse sample of families. Specifically, a path model linking household income to harsh parenting via neighborhood disorder, fear for safety, maternal depressive symptoms, and family conflict was evaluated using cross-sectional data from 2,132 families with children ages 5-16 years from Chicago. The sample was 42% Mexican American, 41% African American, and 17% European American. Results provide support for a family process model where a lower income-to-needs ratio is associated with higher reports of neighborhood disorder, greater fear for safety, and more family conflict, which is in turn, associated with greater frequency of harsh parenting. Our tests for moderation by ethnicity/immigrant status, child gender, and child age (younger child vs. adolescent) indicate that although paths are similar for families of boys and girls, as well as for families of young children and adolescents, there are some differences by ethnic group. Specifically, we find the path from neighborhood disorder to fear for safety is stronger for Mexican American (United States born and immigrant) and European American families in comparison with African American families. We also find that the path from fear for safety to harsh parenting is significant for European American and African American families only. Possible reasons for such moderated effects are considered. Keywords: parenting, neighborhood disorder, fear, poverty, stress
Low income and poverty strain the well-being of families in many ways: the worry associated with financial hardship, the struggle to make ends meet, and the inability to provide material goods for one’s children are just some of the stressors parents face (Duncan & Brooks-Gunn, 1997; Gershoff, Aber, Raver, & Len non, 2007). Another way that income is tied to family well-being and parenting is through the restriction it places on where families can afford to live, as the neighborhood in which a family resides is determined—to some degree— by a family’s resources as well as segregation (Massey, Gross, & Eggers, 1991). Families with limited incomes are more likely to live in low-income neighbor hoods (Fiy & Taylor, 2012), which are more likely to offer lower-quality resources and have higher rates of crime and disor der (Sampson, Raudenbush, & Earls, 1997) in comparison with higher income neighborhoods. The purpose of this study is to contribute to the discussion of how income influences families, by examining neighborhood characteristics and perceptions of these characteristics in the link between income and parenting across a diverse group of families. This work is informed by two theoretical frameworks: (a) the family stress model (Conger & Elder, 1994; Elder, 1999), which
posits that the psychological stress associated with economic hard ship strains family relations and disrupts parenting; and (b) social disorganization theory, which suggests that neighborhoods with a high proportion of poor residents, high residential mobility, and/or high ethnic heterogeneity are disadvantaged in comparison with other neighborhoods, as these characteristics are hypothesized to weaken residents’ attachment to the neighborhood and limit re sources with which to support local institutions (Shaw & McKay, 1969). These conditions in turn are associated with the emergence of disorder (Massey & Denton, 1993; Sampson et al., 1997; Wilson, 1987) which is related to higher levels of mistrust and fear among residents (Ross & Mirowsky, 2001; Wallace, 2012). There fore, we integrate these two theories and test a model that builds on some of the links in the family stress model by including an indirect path from income-to-needs ratio to family conflict via neighborhood disorder and fear for safety (see Figure 1). We also test whether parental report of fear is directly associated with harsh parenting, as mothers under stress (e.g., living with fear for safety) may be more likely to use punitive parenting practices (WebsterStratton, 1990). We examine whether, in addition to maternal depressive symptoms and family conflict, low income is associated with living in more disordered neighborhoods, which in turn is associated with increased levels of fear, family conflict and ulti mately, harsh parenting. We test our model with a large (N = 2,132), diverse sample of Mexican American, African American, and European American families with children (ages 5-16 years) who participated in the Project of Human Development in Chicago Neighborhoods (PHDCN). Although research examining the pro cesses Unking low-income to harsh parenting has been conducted with different ethnic groups, as discussed below, most studies have
This article was published Online First November 10, 2014. R. Gabriela Barajas-Gonzalez, Department of Population Health, New York University School of Medicine; Jeanne Brooks-Gunn, Department of Human Development, Teachers College, Columbia University. Correspondence concerning this article should be addressed to R. Ga briela Barajas-Gonzalez, One Park Avenue, 7th floor, Center for Early Childhood & Development, Department of Population Health, New York, NY 10016. E-mail: [email protected]
855
856
BARAJAS-GONZALEZ AND BROOKS-GUNN
Figure 1. Model of indirect influence of income-to-needs ratio on harsh parenting via neighborhood disorder and fear for safety.
been based on smaller, nonrepresentative samples that do not include multiple ethnic/immigrant groups in one study, making it difficult to make comparisons about the processes operating for each group. The diversity of the PHDCN sample allows us to address this.
Low Income and Multiple Sources of Stress Considerable progress has been made in documenting the im pact of economic hardship on family functioning in European American and African American families (Conger, Wallace, Sun, Simons, McLoyd, & Brody, 2002; McLoyd, 1998), and more recently, in Mexican American families as well (Barrera et al., 2002; Loukas, Prelow, Suizzo, & Allua, 2008; Parke et al., 2004). Studies indicate that low-income and economic hardship strain parent mental health, increasing the likelihood of family conflict and reducing parents’ ability to interact with family members in a responsive and nurturing manner. Families with low incomes are more likely to live in disadvan taged neighborhoods (Fry & Taylor, 2012; Sampson & Sharkey, 2008) and face daily stressors associated with living in such neighborhoods— over and above having low family income— such as crime, harassment, and danger (Hill, Ross, & Angel, 2005; Wandersman & Nation, 1998). Perceptions of the neighborhood environment as violent, unsafe, or highly disordered can increase feelings of distress, both directly and indirectly, through increased feelings of fear, powerlessness, and a diminished sense of control (Brunton-Smith & Sturgis, 2011; Hill & Maimon, 2013; Ross & Jang, 2000). The stress imposed by such adverse neighborhood conditions has been found to generate angst above and beyond the effects of the individual’s own personal stressors, straining mental health (Cutrona, Wallace, & Wesner, 2006; Hill et al., 2005; Ross & Mirowsky, 2001; Wallace, 2012). Hill and colleagues (2005), for instance, used data from the Welfare, Children, and Families project (a household-based, stratified random sample of 2,402 poor mothers in low-income neighborhoods in Boston, Chicago, and San Antonio) to test whether neighborhood conditions influenced women’s health over and above individual characteristics. The authors distinguished individual disadvantage from neighborhood problems by including sociodemographic characteristics of indi viduals (e.g., education, economic well-being, and employment status) that correlated with the likelihood of living in a neighbor hood with high levels of disorder and with poor health. Women who lived in neighborhoods where they reported a lot of problems with drugs, crime, abandoned buildings, and unsafe streets had
significantly worse self-reported health than those who reported more order in their neighborhoods. Notably, the authors found no evidence of subjectivity bias— as people who were more distressed were not likely to report more disorder. In addition to being associated with mental health, neighbor hood conditions have also been linked to parenting. Studies, though varied with respect to the number and quality of family level characteristics accounted for, have found that parents who reside in dangerous neighborhoods report more controlling parent ing practices, more need for supervision, and greater use of harsh disciplinary techniques in comparison to similar parents in less dangerous neighborhoods (Gutman, McLoyd, & Tokoyawa, 2005; Kotchick, Dorsey, & Heller, 2005; Pinderhughes, Nix, Foster, Jones, & The Conduct Problems Prevention Research Group, 2001). Thus, it is possible that neighborhood disorder undermines the ability of families to function effectively, as parents who live in dangerous neighborhoods may be more likely to experience chronic tension and stress, increasing the likelihood of disciplining and parenting reactively (Hill & Herman-Stahl, 2002; McLoyd, 1990). Recently, several studies have tested the relations among lowincome, neighborhood stress, and family functioning with African American families (Gutman et al., 2005; Kotchick et al., 2005), Mexican American families (Gonzalez et al., 2011; White, Roosa, & Zeiders, 2012; White, Zeiders, Gonzales, Tein, & Roosa, 2013), and European American families (Deng et al., 2006). Kotchick and colleagues (2005), for example, interviewed 123 African Ameri can single mothers at several time points to understand whether exposure to neighborhood stress (e.g., presence of gangs, drug dealing, and shootings) was related to psychological distress (e.g., depression, anxiety) and, in turn, compromised parenting. Findings revealed that in fact, neighborhood stress was related to greater psychological distress, predicting less positive parenting practices 15 months later. Similarly, Deng et al. (2006) found that perceived neighborhood quality mediated the association between neighbor hood disadvantage and parent-child conflict among 189 lowincome European American and Mexican American mothers. Of interest to the authors, Gonzalez et al. (2011) and White et al. (2012) only found moderated associations between perceptions of neighborhood danger/risk and harsh parenting in their sample of 750 Mexican American families. This may have been, in part, because no measures of mother’s psychological well-being or family conflict were included in the models they tested. We build on the previous literature by including an indicator of maternal depressive symptoms and an assessment of family conflict—two key constructs that have been theorized to influence parenting, but that have not been consistently included in previous studies. More over, because we have a large sample of African American, European American, and Mexican American families, we are able to test for possible moderation by ethnicity and immigrant status, as some studies have suggested that that there may be there may be race-ethnic differences in how disorder is perceived (ChristieMizell et al., 2003; Sampson & Raudenbush, 2004). Sampson and Raudenbush (2004) for example, found that European Americans reported more disorder than African Americans on the same block group of a neighborhood. They posited that African Americans may have a higher threshold for perceiving disorder in comparison with European Americans (assuming African Americans had been exposed to more disorder in the past). Relatedly, immigrant moth-
NEIGHBORHOOD STRESSORS AND PARENTING
ers, especially those bom in countries where poverty is more common, may have been raised with expectations of neighbor hoods that differ from those of people born in the United States. Mothers raised in countries where poverty is more common may be less stressed by adverse neighborhood conditions in the United States, especially if the current conditions are better than the ones in which they were raised. Baccallao and Smokowski (2007) for example, conducted qualitative interviews with undocumented Mexican families and found that these immigrant families consid ered the most basic living conditions in the United States a marked improvement in comparison with the chronic poverty they endured in Mexico. We test for the possibility that neighborhood charac teristics influence mothers differentially by disaggregating our Mexican American sample by immigrant status and comparing first (foreign-bom mother and foreign-born child) and second (foreign-born mother, United States bom child) generation status families (“Mexican immigrant”) to third generation and beyond (United States bom mother, United States born child) families (“Mexican American”) and testing for moderation across the Mex ican immigrant, Mexican American, African American, and Euro pean American subgroups. Based on the literature reviewed above, we hypothesized that the association between neighborhood dis order and fear for safety may be stronger for European American and Mexican American families than for African American and Mexican immigrant families. In addition to moderation by ethnic/immigrant group, we test for moderation by gender and examine whether the model holds for mothers of boys versus mothers of girls. A handful of studies examining the association between economic hardship, maternal depression, and parenting have tested for moderation by gender (Conger et al., 1992, 1993; Dennis, Parke, Coltrane, Blacher, & Borthwick-Duffy, 2003; Gutman, McLoyd, & Tokoyawa, 2005; Mistry, Vandewater, Huston, & McLoyd, 2002) with some finding no evidence of moderation (Mistry et al., 2002) and others finding differences for mothers of boys versus girls (Conger et al., 1992; Conger et al., 1993; Dennis et al., 2003). For example, in a series of studies analyzing cross-sectional data from 7th Grade European American boys (Conger et al., 1992) and then girls (Conger et al., 1993), Conger and colleagues found a direct association between caregiver depressed mood and hostile parenting among adolescent boys, but not among girls. Conger et al. (1993) posited that one explanation for this may be that girls, being less oppositional and disruptive than boys, were perhaps less likely to elicit harsh parenting from a distressed caregiver. Similarly, in their study with 56 immigrant Latina mothers of children ages 4-13 years, Dennis and colleagues (2003) found measures of maternal economic hard ship and depression to be significantly correlated for mothers of boys but not for mothers of girls. Mistry et al. (2002) however, found no evidence of moderation by gender in their test of the family stress model with 419 African American, Hispanic, and European American families of children ages 5 to 12 years. With regards to the links between disorder, fear for safety, and parent ing, there is a scarcity of studies examining gender differences in the association between neighborhood context and parenting. The limited literature that does exist suggests that boys may be more vulnerable to the detriments of living in impoverished neighbor hoods than girls, in part because girls may be socialized to spend more time at home rather than in their neighborhoods (ClampetLunduist, Dyncan, Edin, & King, 2006; Kroneman, Loeber, &
857
Hipwell, 2004). We are aware of one study that has examined the influence of neighborhood disadvantage on harsh parenting by gender. That study, conducted by Leventhal and Books-Gunn (2005), found that contrary to expectation, moving out of public housing in high-poverty neighborhoods into low poverty neigh borhoods was associated with greater parental harshness toward girls, but not boys. It was posited that the act of moving itself, and the disruption to social networks, may have been particularly detrimental to the mother-daughter relationship. Given the dearth of research concerning gender differences in the association be tween low-income, neighborhood context, and family functioning, no specific hypotheses regarding gender differences are made. We also take advantage of the wide age range of the children in our sample (5-16 years of age), and examine moderation by age group, testing whether the model holds for children who are in adolescence (ages 11-16; 44% of our sample) versus children who are younger (ages 5-10; 56% of our sample). To our knowledge, no studies of the FSM with a broad age group of children have tested for moderation by age group, perhaps because of a limited sample size. We address Conger & colleagues’ (2010) recommen dation to test whether the model is equivalent for families with younger, versus older, children in our sample. We posit that the model may be more salient for families with adolescents, as adolescence may bring additional stressors to family dynamics (e.g., greater parent-child conflict as child seeks independence from parents). Moreover, in comparison with younger children, adolescents may spend more time outside their home, which could cause parents to be more cognizant of neighborhood stressors.
Method Design and Sample This study uses two independent sources of data collected by the PHDCN, an interdisciplinary study of youth, families and neigh borhoods. Family level data are drawn from the first wave of the Longitudinal Cohort Study (PHDCN-LCS); neighborhood disor der data are drawn from the Community Survey (PHDCN-CS).
Longitudinal Cohort Survey Participants were drawn from a probability sample designed to capture the diversity of Chicago’s neighborhoods (see www.icpsr .umich.edu/icpsrweb/PHDCN/sitemap.jsp for a detailed descrip tion of PHDCNs design, sampling, and method). Data for 847 census tracts (from the 1990 U.S. Census) comprising the city of Chicago were combined to create 343 neighborhood clusters (NCs) that included two to three census tracts (—8,000 residents). A stratified probability sample of 80 neighborhood clusters (crossclassified by racial/ethnic composition and SES) was then drawn from the 343 neighborhood clusters. Households within these 80 NCs were randomly selected and screened for eligibility; —1,000 children/primary caregivers within each of seven age cohorts (birth, 3, 6, 9, 12, 15, and 18 years) were interviewed in their homes (N = 6,226), rendering a sample representative of children in the city of Chicago in 1995.
Community Survey The goal of the Community Survey (CS) was to gather in-depth information about the NCs from residents who were not partici-
BARAJAS-GONZALEZ AND BROOKS-GUNN
858
pants in the LCS, so as to obtain reports of neighborhood charac teristics independent of the study participants. These data were collected between the years 1994-1995 (at approximately the same time data was being collected for the longitudinal cohort survey) and involved a three-stage sampling strategy. First, city blocks were randomly selected from the 343 NCs. Second, house holds within blocks were randomly selected, and third, one adult from each household was randomly chosen to complete the survey. In total, 8,782 adults provided information on the characteristics and social processes of their neighborhood. Variables were aggre gated to neighborhood-level averages. Every NC in the LCS was represented in the CS (see http://www.icpsr.umich.edu/icpsrweb/ PHDCN/sampling.jsp). The current study used Wave 1 data on children from cohorts 6, 9,12, and 15 (i.e., those who were roughly 6, 9, 12, or 15 years of age at Wave 1; N = 3,324). Respondents who were not the child’s biological mother (n = 523 respondents who were either the child’s biological father, grandmother, or other relatives) or were missing information as to their relationship to the child (n = 29) were excluded from the sample. Mothers who were missing birth country information (n = 19) and mothers who did not identify as African American, European American, or Mexican were ex cluded from the sample (n = 621 identifying as Puerto Rican, Central and South American, Asian, African, European and Mid dle Eastern). Thus, the focus in this study was on the African American (AfAm), European American (EuroAm), and Mexican American families (United States bom and foreign bom mothers; MexUS, Mexlm, respectively) who were interviewed at Wave 1 (n = 2,132; see Table 1 for descriptive statistics). M easures The measure of neighborhood disorder was drawn from the CS, which occurred concurrent in timing to the Wave 1 LCS, and was based on reports from residents living in the same neighborhood but not participating in the study. All maternal and family char acteristics were drawn from the LCS and based on mother report. As part of the LCS, a demographic and behavioral information
interview was administered to the primary caregiver. Demographic information on education, race, income, birth country, and marital status was obtained. In addition, the primary caregiver completed structured interviews and self-report questionnaires assessing a range of health and behavioral indicators (see http://www.icpsr.umich.edu/ icpsrweb/PHDCN/instruments.jsp#longitudinal for greater measure detail). The items, means, and SDs for the independent variables of interest are presented in Table 1. Correlations among all variables of interest are presented in Table 2. Income-to-needs ratio. Annual family income and family size were used to create an income-to-needs ratio, by dividing the reported total annual family income by the official poverty thresh old for the respective household size in 1995. An income/needs ratio of one or less signifies poverty status. Lower family income has been used as a proxy for economic pressure (Lee, Anderson, Horowitz, & August, 2009). In the present sample, the mean income-to-needs ratio was 1.59 (SD = 1.26), with significant variation across subgroups (see Table 1). Maternal self-reported depressive symptoms. One item as sessing whether the mother suffered from symptoms of depression was drawn from the Family Mental Health and Legal History interview, based on the Family History Assessment Module (Janca, Bucholz, & Janca, 1992). Mothers were asked, “Has any one mentioned ever suffered from depression, that is, they have felt so low for a period of at least two weeks that they hardly ate or slept, or couldn’t work or do whatever they usually do?” There were 233 mothers who indicated that they had suffered from such a condition (0 = no; 1 = yes). Fear for safety. Three questions from the Exposure to Vio lence interview (Earls, Brooks-Gunn, Raudenbush, & Sampson, 1994) were used to construct a measure of fear for safety. At Wave 1, caregivers were asked about their fear in certain situations and their exposure to several different types of violent acts (e.g., “Are you afraid child may be hurt by violence in neighborhood?” “Are you afraid child may be hurt by violence in front of house?”). Response options for each question were dichotomous (0 = no; 1 = yes). The three items were summed so that higher scores
Table 1 Descriptives fo r Whole Sample and by Ethnic/lmmigrant Subgroup
Ml (SD)
Mexlm (n = 787) M (SD)
MexUS (n = 107) M (SD)
AfAm (n = 881) M (SD)
EuroAm (n = 357) M (SD)
1.59(1.26) 1.88 (0.34) 1.68(1.18) 1.20(1.41) 6.83 (6.40) 43.6% 48.7% 11.0%
1.18(0.86) 2.00 (.028) 1.91 (1.18) 0.86(1.22) 5.66 (5.71) 41.8% 48.9% 7.9%
1.71 (1.34) 1.90 (0.30) 1.86(1.07) 1.57(1.52) 7.40 (7.79) 41.1% 39.3% 9.5%
1.46(1.27) 1.91 (0.32) 1.61 (1.16) 1.44(1.46) 7.51 (6.38) 46.0% 48.7% 13.2%
2.72 (1.33) 1.56 (.031) 1.25(1.14) 1.24(1.50) 7.54(7.16) 42.6% 51.0% 13.0%
Total (.N = 2,132)
Income-to-needs ratiob,c,e'f,s Neighborhood disorderb,c,e-f Fear for safetyb'c,e,f Family conflictb'c,d Harsh parentingb-c,d Adolescent Male child Maternal depressive symptoms11
ANOVA/Pearson x2 F(3, F(3, F(3, F(3, F(3,
1825) = 135.67, p = .000 2128) = 170.76, p = .000 2098) = 28.76, p = .000 2121) = 27.06, p = .000 2116) = 14.01, p = .000 X2(3) = 4.58, p = .205 X2(3) = 3.46, p = .326 X2(3) =-13.50 , p = .004
Note. MAIm = Mexican immigrant; MAUS = Mexican American; AfAm = African American; EuroAm = European American. Mexican immigrant mothers had been in the United States an average of almost 15 years (M = 14.71, SD = 6.43). a Denotes statistically significant difference between groups at the .05 level. b Post hoc test indicates statistically significant difference between MAIm and MAUS at the .05 level. c Post hoc test indicates statistically significant difference between MAIm and African Americans at the .05 level. d Post hoc test indicates statistically significant difference between MAIm and European Americans at the .05 level. e Post hoc test indicates statistically significant difference between MAUS and African Americans at the .05 level. f Post hoc test indicates statistically significant difference between MAUS and European Americans at the .05 level. g Post hoc test indicates statistically significant difference between African Americans and European Americans at the .05 level.
NEIGHBORHOOD STRESSORS AND PARENTING
859
Table 2 Bivariate Correlations fo r Study Variables o f the Project o f Human Development in Chicago Neighborhoods (N = 2,132)
1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
*p < .05.
" p < .01.
1
2
3
4
5
6
1.00 -.46*** -.18*** -.15*** -.0 2 -.05*
1.00 .25*** .06** .01 .01
1.00 .16*** .12*** .10***
1.00 .33*** .15***
1.00 .15***
1.00
***p < .001.
indicated greater fear for safety. The internal consistency for these items was adequate (Cronbach’s a = .73; with alphas ranging from .65 (MexUS) to .76 (Mexlm)). Mean perceived lack of safety for this sample was 1.68 (SD = 1.18), with scores ranging from 0-3.00. Family conflict. The conflict subscale of the Family Environ ment Scale (FES; Moos & Moos, 1981) was used to assess family conflict. Caregivers reported whether each of five statements ac curately described their families (e.g., “family members often criticize each other,” “ members fight a lot”; 0 = false, 1 = true). Responses were summed across items, such that higher scores indicated greater conflict. Scores ranged from 0-5.00 (M = 1.20, SD = 1.41). Research supports the construct, concurrent and predictive validity of the FES (Moos, 1990). In the present study, the FES conflict scale demonstrated adequate internal reliability (Cronbach’s a = .70, with as ranging from .66 (Mexlm) to .75 (EuroAm)). Similar a levels have been reported by other studies of Latino families of varying acculturation levels using this measure (see Miranda, Estrada, & Firpo-Jimenez, 2000). Harsh parenting. Harsh parenting is characterized by paren tal inability to control emotions during interactions with their child, thereby yelling, threatening, or using physical aggression. Harsh parenting was assessed using seven items from the Conflict Tactics scale (CTS; Straus, 1979). Caregivers were asked how many times they had behaved in a certain manner when there was a problem with the child (e.g., “In the past year, when having a problem with child, how many times adult insulted/swore at child?” options: 0 = never, 1 = once, 2 = twice, 3 = 3-5 times, 4 — 6 -1 0 times, 5 = 11-20 times, 6 = more than 20 times). The items were summed so that higher scores indicated more aggres sion toward the child. The internal consistency for these items was adequate (Cronbach’s a = .73; with as ranging from .68 (AfAm) to ,76(EuroAm)). Mean mother-to-child aggression was 6.83 (SD = 6.42) with scores ranging from 0 to 39.00. The CTS has been used with ethnically diverse families (Lee, Brooks-Gunn, McLanahan, Notterman, & Garfinkel, 2013), including Mexican mothers in Mexico (Frias-Armenta & McCloskey, 1998). Neighborhood disorder. We used the neighborhood disorder scale from the Community Survey that consists of 12 items as sessing both physical disorder (e.g., “how much of a problem is graffiti in your neighborhood?” options: 3 = a big problem-, 2 = somewhat o f a problem', 1 = not a problem) and social disorder (e.g., “how much of a problem is groups hanging out causing trouble?” options: 3 = a big problem-, 2 = somewhat o f a problem', 1 = not a problem). The consistency of this scale is adequate
(Cronbach’s a = .70). See Sampson & Raudenbush (2004, p. 324) for greater detail about the creation and reliability of the neigh borhood disorder scale. A nalytic Strategy Analyses. Chi-square tests were used to test for significant group mean differences on categorical variables (e.g., child gen der); one-way analysis of variances (ANOVAs) with post hoc tests were used to test for group mean differences on continuous vari ables (e.g., income-to-needs ratio, family conflict, etc.) using SPSS 16. Variance and group sizes were considered to determine which post hoc procedure to use. We tested the hypothesized model using Mplus 6.0 (Muthen & Muthen, 2010). Maximum likelihood estimator with robust SEs (MLR estimation), TYPE = COMPLEX, and the CLUSTER command were used to account for the nested structure of our data (participants nested within neighborhoods). MLR estimates are robust to nonnormality and nonindependence of observations when used with the TYPE = COMPLEX and CLUSTER proce dure. This method of analysis yields standard errors of path coef ficients that are adjusted for neighborhood clustering (Muthen & Muthen, 2010) giving more conservative parameter estimates. Model fit was assessed using y2 fit statistics, the root mean square error of approximation (RMSEA; Browne & Cudeck, 1993), the comparative fit index (CFI; Bentler, 1990), and the Tucker Lewis Index (TLI). Good model fit is indicated by a nonsignificant x2 statistic, CFI and TLI greater than .90 (Bentler, 1990), and RM SEA less than .05 (less than .08 is adequate; Browne & Cudeck, 1993). Because the x2 statistic is affected by sample size, and larger sample sizes (more than 200 cases) may lead to a significant X2 even though the differences between the observed and predicted covariances are modest (Kline, 2005), the x2 ratio (x2/df), which adjusts for model complexity is reported. In general, a x2 ratio between 1 and 3 indicates good fit (Arbuckle & Wothke, 1999). We tested the model for subgroup moderation by constraining all paths to be equal across subgroups, and comparing the con strained versus the unconstrained models using the \ 2 difference test with the Satorra-Bentler scaled x2 (Satorra & Bentler, 1999), where a nonsignificant Ax2 indicates that the constrained model fits the data just as well and is more parsimonious than the unconstrained model. If the Ax2 was significant, however, we concluded that moderation had occurred, and proceeded to free one path at a time to identify paths that were significantly different across groups. We used the same procedure outlined above to
BARAJAS-GONZALEZ AND BROOKS-GUNN
860
check for moderation by gender and by age group. We assessed the significance of indirect paths using the MODEL INDIRECT com mand in Mplus, which computes delta method standard errors for the indirect effects (Muthen & Muthen, 2010). Treatment of missing data. Less than 5% of cases were missing data on fear for safety, conflict, depressive symptoms, and harsh parenting, but roughly 14% of cases were missing data on income-to-needs ratio. The full information maximum likelihood (FIML) method, which estimates model parameters and SEs using all available raw data, was used for all analyses. FIML has been shown to be more effective compared with other estimation ap proaches when dealing with missing data in structural equation models (Enders & Bandalos, 2001) and precludes the need for imputation or the deletion of incomplete cases.
Results Descriptive information on the sample is reported in Table 1 (by ethnic/immigrant group), Table 3 (by gender) and Table 4 (by age group). As shown in Table 1, nearly 42% of the sample was Mexican American (n = 894) and 41% of the sample was African American in = 881). Roughly half of the households in this study consisted of families in which both biological parents of the child were married to each other in = 1,030). Almost half the sample did not graduate from high school. There were no significant differences in child age or gender across the four ethnic subgroups. There were, however, significant group differences on all other variables of interest across the four ethnic subgroups. For greater detail, see Table 1. There were no significant differences on variables of interest between families of boys versus families of girls (see Table 3). There were some differences, however, when examined by age group. Specifically, families of adolescents had higher mean income-to-needs ratio (M = 1.67, SD = 1.27 vs. M = 1.52, SD = 1.25; t( 1,827) = —2.49, p = .013), reported greater fear for safety (M = 1.78, SD = 1.15 vs. M = 1.59, SD = 1.21; /(2,011) = —3.60, p = .000), and greater harsh parenting (M = 7.33, SD = 6.61 vs. M = 6.44, SD = 6.24; r( 1,923), -3 .1 5 = p = .002) in comparison with families of young children (see Table 4). Most variables of interest were significantly correlated (see Table 2). As expected, income-to-needs ratio was negatively associated with neighborhood disorder, fear for safety, and family conflict. Fear for safety and family conflict were both significantly associ
ated with each other, with maternal depressive symptoms, and with harsh parenting. See Tables 3 and 4 for bivariate correlations by gender and by age group. We will discuss the results of our tests for moderation by gender and by age first as those are most straightforward. To test for moderation by gender, we compared the fit of our unconstrained path model (X2/df = 3.62, CFI = .978, TLI = .918, RMSEA = .054, 90% Cl [.033, .075]), to one where all parameters were held to be equal for boys and girls (x2ldf = 2.18, CFI = .978, TLI = .963, RMSEA = .036, 90% Cl [.020, .051)]; constraining all paths of interest to be equal across the two groups did not result in worse model fit (Ax2 = 9.16(10), p = .516), suggesting paths were similar for families of male and families of female children. To test for moderation by age, we conducted the same procedure, and compared the fit of the unconstrained path model (x2/df = 4.38, CFI = .970, TLI = .889, RMSEA = .061, 90% Cl [.041, .082]), to one where all parameters were held to be equal for young children and adolescents (x2/d f = 2.73, CFI = .966, TLI = .943, RMSEA = .044, 90% Cl [.029, .058]); constraining all paths of interest to be equal across the two groups did not result in worse model fit (Ax2 = 13.76(10), p = .184), suggesting paths were similar for families of young children and adolescent children. All paths of interest in the model, with the exception of the path from income-to-needs ratio —> maternal depression, were signifi cant across gender and age group (see Appendix A, Table A l). Lower income-to-needs ratio was significantly associated with more family conflict, which was in turn, associated with greater harsh parenting. As expected, lower income-to-needs ratio was also significantly associated with greater neighborhood disorder, which was associated with greater fear for safety. Fear for safety was associated with both greater family conflict and greater en dorsement of maternal depressive symptoms. Maternal depressive symptoms and family conflict were both in turn, associated with greater harsh parenting. Test of indirect effects indicate that the indirect paths from income-to-needs ratio to harsh parenting via family conflict and via neighborhood disorder were significant across gender and age groups (see Appendix A, Table A2). The same procedure was followed to test for moderation by ethnic group. We compared the fit of our unconstrained path model (X2l df = 1.88, CFI = .977, TLI = .915, RMSEA = .044, 90% Cl [.018, .068]), to one where all parameters were held to be equal
Table 3
Means, SDs, and Correlations for Study Variables by Gender (Boys, n = 1,038; Girls, N = 1,094) Measure 1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
M
SD
1
2
3
4
5
6
1.61 (1.56) 1.88 (1.88) 1.71 (1.64) 1.18(1.22) 6.98 (6.68) 0.12(0.10)
1.26(1.26) 0.34 (0.34) 1.18(1.18) 1.38(1.45) 6.51 (6.32) 0.32 (0.30)
1.00 -.475*** -.228*** -.159*** —.061 + —.059f
-.452*** 1.00 .287*** .063* .018 .021
-.141*** .216*** 1.00 .183*** .137*** .117***
-.144*** .064* .139*** 1.00 .296*** .146***
.017 -.005 .100** .370*** 1.00 .137***
-.038 -.020 .074” .163' .154' 1.00
Note. Correlations for boys are reported below the diagonal and correlations for girls are reported above the diagonal. Girls’ means and SDs are reported in parentheses. Sample size for boys ranged from 902-1,031 and for girls ranged from 920-1,094 because of list-wise deletion. All other analyses utilized the full sample (N = 2,132) and accounted for missing data using full information maximum likelihood, t test indicated no significant difference in means between boys and girls. > < . 1 0 . *p < .05. * > < .0 0 1 .
NEIGHBORHOOD STRESSORS AND PARENTING
Table 4 Means, SDs, and Correlations fo r Study Variables by Age Group (Young Child, n Measure 1. 2. 3. 4. 5. 6.
Income-to-needs ratio Neighborhood disorder Fear for safety Family conflict Harsh parenting Depressive symptoms
=
861
1,202; Adolescent, N
=
930)
M
SD
1
2
3
4
5
6
1.52(1.67) 1.89(1.87) 1.59(1.78) 1.17(1.24) 6.44 (7.33) 0.11 (0.12)
1.25 (1.27) 0.34 (0.34) 1.21 (1.15) 1.40(1.42) 6.24 (6.61) 0.31 (0.32)
1.00 -.473*** -.214*** -.149*** —.056+ -.047
-.448*** 1.00 .285*** .038 .023 -.016
-.153*** .212*** 1.00 .151*** .094** .105***
-.154*** TOO** .168*** 1.00 .300*** .151***
.011 -.009 .141*** .373*** 1.00 .150***
-.051 .024 .082* .157*** .140*** 1.00
Note. Correlations for young children are reported below the diagonal and correlations for adolescents are reported above the diagonal. Adolescents’ means and SDs are reported in parentheses. Sample size for young children ranged from 1,021-1,202 and for adolescents ranged from 808-928 because of list-wise deletion. All other analyses utilized the full sample (N = 2,132) and accounted for missing data using full information m a d m . i m likelihood. Compared with families with young children, families with adolescents had higher mean income, r(l,827) = -2.49, p = .013; also higher fear for safety r(2,011) = -3.60, p = .000; higher harsh parenting r(1,923) = -3.15, p = .002. *p < .05. **p < .001.
across the four ethnic groups (\ 2/d f = 1.59, CFI = .956, TLI = .943, RMSEA = .036, 90% Cl [.018, .051]); constraining all paths of interest to be equal across groups resulted in worse fit (Ax2 = 44.16(30), p = .046), suggesting subgroup moderation. Thus, we proceeded to test for subgroup moderation, freeing one path of interest at a time, to see where moderation was occurring. Freeing the path from disorder-* fear, the path from fear —» harsh parent ing, and the path from fear —*• family conflict resulted in nonsig nificant Satorra-Bentler \ 2 tests indicating subgroup moderation in those paths. Therefore, we tested a model where these three paths where unconstrained, whereas the rest of the paths were con strained to be equal across subgroups. Model fit indices indicated that this model fit the data well and was more parsimonious than our baseline model (x2 ratio = 1.36, CFI = .978, TLI = .965, RMSEA = .028, 90% Cl [.000, .046]; (Ax2 = 22.80 (21), p = .354). Upon examining the parameter estimates in the path from fear —* family conflict, we noted that they were relatively similar for Mexican immigrant, Mexican American, and African Ameri can families in comparison with European American families. Therefore, we constrained the path from fear —* family conflict to be equal for Mexican immigrant, Mexican American, and African American families, but left it free for European American families. Doing so resulted in a more parsimonious model with similar fit (X2 ratio = 1.36, CFI = .977, TLI = .965, RMSEA = .028, 90% Cl [.000, .046]).
As shown in Table 5, across the four groups, lower income-toneeds ratio was significantly associated with greater endorsement of maternal depressive symptoms, which was in turn, significantly associated with harsh parenting. Lower income-to-needs ratio was also significantly associated with more family conflict, which was in turn, significantly associated with harsh parenting. These indi rect paths from income-to-needs ratio to harsh parenting were significant across all ethnic groups (see Table 6). The additional paths from income-to-needs ratio to neighborhood disorder, from disorder to fear, and from fear to maternal depressive symptoms were all significant across ethnic groups. As expected, lower income-to-needs ratio was significantly associated with more neighborhood disorder. Neighborhood disorder, in turn was sig nificantly associated with fear for safety, though— consistent with moderation analyses—the association was stronger for the immi grant and United States born Mexican American mothers and European American mothers ((3MAim = -239; |3MAUS = .202; PEuroAm = -314) than for the African American mothers ((3 ^ ,^ = .094). The path from fear to family conflict was significant for every subgroup, but moderation tests indicated that the path was particularly strong for European American mothers (PEuroAm = .216) relative to the immigrant and United States bom Mexican American mothers and African American mothers (PMAim = 1-39; P m au s = °99; PEuroAm = 119). Fear was significantly associated with maternal depressive symptoms across every sub-
Table 5 Standardized Parameter Estimates fo r Final Path Model Mexlm (n = 787)
INC —* Dep INC -»• Con INC Dis Dis —» Fear Fear —* Con Fear —* Dep Con -> Harsh Dep -» Harsh Fear —* Harsh
MexUS (n = 107)
PCS©
P
VISE)
-.051 (.02) -.146 (.02) -.265 (.04) .239 (.05) .139 (.03) .121 (.02) .270 (.03) .080 (.03) .051 (.04)
.019 .000 .000 .000 .000 .000 .000 .002 .190
-.075 (.03) -.181 (.03) -.384 (.05) .202 (.09) .099 (.22) .101 (.02) .252 (.03) .064 (.03) .027 (.11)
AfAm (n = 881) P
(3 (SE)
.023 .000 .000 .024 .000 .000 .000 .014 .810
-.0 6 4 (.03) -.187 (.03) -.348 (.03) .094 (.04) .119 (.02) .101 (.02) .293 (.03) .089 (.03) .081 (.04)
EuroAm (n = 357) P
P(5£)
P
.015 .000 .000 .033 .000 .000 .000 .002 .027
-.063 (.03) -.185 (.03) -.390 (.04) .314 (.07) .216 (.07) .092 (.02) .267 (.03) .082 (.03) .187 (.07)
.015 .000 .000 .000 .002 .000 .000 .003 .005
Note. INC - income-to-needs ratio; Dep = maternal depressive symptoms; Dis = neighborhood disorder; Con = family conflict; Fear = fear for safety; Harsh = harsh parenting. Though not shown, the direct relation between income-to-needs ratio and harsh parenting was tested in all models, but was not observed to be significant.
BARAJAS-GONZALEZ AND BROOKS-GUNN
862
[5
ON^-sfmONOQcn < pNo X X ) — X ! XS X!
c M W) W) a> (u 5 ‘
