Neighborhood, Poverty Duration and Child Maltreatment Reports: A Multilevel Negative Binomial Study

Background/Purpose: Theory suggests that both neighborhood poverty and family poverty history over time can have unique effects on maltreatment risk. Unfortunately, few child maltreatment studies have taken a longitudinal view of poverty, using either cross-sectional samples or failing to track changes in poverty over time. This study uses a multilevel negative binomial analysis to examine how family poverty (i.e., family’s duration in poverty) and neighborhood poverty are associated with CAN reports while controlling for each other. Furthermore, a moderation effect of neighborhood poverty on the relationship between family poverty and CAN reports is tested.

Methods: Among St. Louis City and County children born between 1989-1994, all children with a first CAN report in 1993-1994 are selected and compose the "CAN" sample (n = 3,343 in 250 neighborhoods). The second sample, "AFDC" sample (n = 2,805 in 223 neighborhoods), is selected from the Aid to Families with Dependent Children (AFDC) cases through matching for child’s age, race, and residential location. In both samples, all children are followed until age 15.

During the follow-up period, this study counts CAN reports and years in poverty using public child welfare, income maintenance and Medicaid records. Neighborhood poverty is measured at the time of sampling (1993-1994) by the census tract child poverty rate. Child’s sex, mental health, and special education, and parent’s mental health, criminal history, and foster care history are controlled.

A multilevel negative binomial (NB) regression is used for analyses. A multilevel design is adopted to separate the neighborhood poverty effect from the family poverty effect while accounting for the nested data structure. A NB design is used as the residuals follow a NB distribution due to the counted nature and the severe positive skewness of the outcome variable.

Results: Family's number of years in poverty is significant and also has the largest effect size in both CAN and AFDC samples. For the CAN sample, families staying for 5 years in poverty show 2 times more CAN reports than families with no poverty experience. For AFDC sample, this increases CAN reports by 3 times. The analyses further reveal that family poverty has the largest unique contribution to the improvement of model fit in both samples. While controlling for family poverty, however, both the main and moderation effects of neighborhood poverty are not statistically significant. Given the large sample size, it is unlikely that this failure to detect a relationship was due to the lack of sufficient power.

Conclusions and Implications: While the association of family poverty with CAN reports is strongly confirmed, this study fails to find any unique neighborhood poverty effect on CAN reports after family poverty is controlled for. This result suggests that while broad efforts to alleviate poverty at the community level may have some benefit, policies designed to lift families out of poverty may have more immediate impacts on child maltreatment.