Abner, K.S. (2013). Dimensions of structural disadvantage: A latent class analysis of a neighborhood measure in child welfare data. Journal of Social Service Research, 1-14
Recent studies have shown the importance of community factors—especially poverty levels—as predictors of child abuse and neglect. Previous work has largely been based on census tract-level indicators, which make it hard to identify specific community risk factors; this present study takes an entirely different approach. Data are individual caregivers’ responses to the Community Environment Scale (CES) from the first wave of the National Survey of Child and Adolescent Wellbeing-II (NSCAW-II); this scale provides measures of two concepts; social order (crime and safety) and social capital (neighbourliness). Latent class analysis (LCA) of these CES responses was used to divide the total NSCAW-II sample (N = 5,872) into three groups:
Class 1: High social order/medium social capital (n = 2055)
Class 2: High social order/low social capital (n = 2801)
Class 3: Low social order/low social capital (n = 1016).
Multinomial logistic regression was then used to examine the relationships between class membership and a set of “covariates” (family and caregiver characteristics from the NSCAW-II). Many of these covariates were found to significantly predict class membership; results were generally consistent with earlier studies: (a) poor and minority group caregivers were significantly overrepresented in Classes 2 and 3—that is, these caregivers rated their neighbourhoods as having low social capital, suggesting the possibility that social capital might be the link between community poverty level and child neglect; (b) the risk of physical abuse was significantly higher for Class 3 than for Class 2, suggesting a possible link between social order and child maltreatment.
The author points out that future work is needed to understand additional possible risk factors, and goes on to make a number of useful suggestions for developing neighbourhood-based interventions for possible preventive work in communities.
Data were based on the abridged 9-item version of the CES, all items calling for 3-category Likert-type responses: 5 social order items (how big a problem?) and 4 social capital items (how does your neighbourhood compare to others?). A program called Latent Gold was used for the LCA of the responses; 1-class through 10-class solutions were calculated, and the 3-class solution chosen, based on model interpretability and parsimony. A parallel analysis of the subsample of permanent caregivers was also done, but only the results for the full NSCAW-II sample are reported. Separate multinomial logit regressions were done for each set of covariates. Multiple imputation (Stata 12) was used to replace missing values for all variables.
In this study, the main emphasis was on the LCA; the logistic regression was presented mainly as a way to “validate” the resulting grouping of subjects. However, the most potentially useful result seems to have been the identification of social capital as a possible link between community poverty and child maltreatment. Though the study was generally well conducted, this was a rather elaborate way of getting there. An alternative might have been to omit the LCA completely, and simply correlate individual caregivers’ CES scores with the set of covariates.