Impact of Migration on Estimating the Incidence of Child Maltreatment Reports

Background & Purpose:

Recent studies have used the National Child Abuse and Neglect Data System (NCANDS) to construct synthetic cohorts to estimate cumulative incidence (CI) of child maltreatment. However, states independently submit NCANDS data, which does not contain information about contact with child welfare systems in other states. The lack of data regarding migration creates the potential for a child to be counted multiple times, leading to possible overestimates of the CI of maltreatment. For example, Kim et al. (2017) puts the CI of reported maltreatment at 37.4%, which is more than 1 in 3 US children. By contrast, in 2015 Children’s Bureau reported that the annual estimate of children reported for maltreatment was about 1 in 19 children. Thus, this study aims to assess the role of migration in CI analyses by comparing the synthetic cohort approach to CI estimation with a prospective cohort approach – a technique that would tend to underestimateCI due to migration. The results are then compared as plausible upper and lower bounds of child maltreatment incidence.

Methods:

Using synthetic cohort and prospective cohort life tables, this analysis estimates the incidence of children in a Western US state with a first-time report of child maltreatment from birth through age 4. For the synthetic cohort, we use NCANDS data from FFY 2010 and Kids Count population data by single age in 2010, treating it as an aging cohort. For the prospective cohort, we use a dataset of all children born in Washington in 2010 that has been deterministically linked with CPS administrative records. We then estimate life table values for both cohorts, using formulas provided by Gehan (1969). In both tables, we derive the conditional probability of having a report of maltreatment at each age, the cumulative probability in early childhood (0 through age 4), and the survival, probability density (PDF), and hazard functions.

Results:

The CI of first-time child maltreatment reports amongst children age 0 through 4 in the synthetic cohort is 15.1%, which is in line with the estimates of Kim, et al. The CI of the prospective birth cohort is 13.7%. For the birth cohort, the mode of the PDF distribution is at age one, while the NCANDS cohort appears bimodal, with a distributional peak at age zero and a second peak at age two. The pattern is similar for the hazard functions.

Conclusions & Implications:

Variability in functional form resulting from the two estimation approaches is consistent with the expected effects of migration on each approach; migration may lead to an overestimate of the maltreatment incidence using a synthetic cohort approach and an underestimate using a prospective cohort approach. These effects necessarily increase as a function of age, as the amount of time in which migration can occur increases. The significant difference observed in the estimates of this analysis, even by the age of four, suggests that further modeling and analysis is needed, and that future childhood maltreatment incidence studies should account for migration effects.