Doctoral Dissertations

Date of Award

12-1995

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Social Work

Major Professor

William R. Nugent

Committee Members

Priscilla Blanton, John Orme, David Patterson

Abstract

While most studies of risk assessment in the child maltreatment literature focus on characteristics of children and families who have been reported to authorities, or on the risk of recurrence of maltreatment, the present study focused on injury severity as the outcome. The purpose of the study was to develop and test, via a cross-validation sample, a model for predicting severity of injury in cases of physical child abuse.

Conducted on a subset of observations from child abuse reports in the U.S. Air Force Family Advocacy Program's central registry, data included selected variables from 6246 substantiated cases of physical child abuse from Fiscal Year 1990 through Fiscal Year 1994. The study included a total of fifteen predictor variables: four child characteristics (gender, age, race, handicapping condition), two family characteristics (number of children in the home, income). two report characteristics (source, season), and seven perpetrator characteristics (gender, age, race, relationship to the child, marital status, history of substance abuse, history of violence). Injury severity was categorized into four levels: mild, moderate, severe, or death.

It was hypothesized that each of the fifteen predictors would be related to injury severity when controlling for all other variables in the model, and that two interactions (child age-by-child gender, perpetrator gender-by-child gender) would be confirmed. Using Ordinal logistic regression procedures to analyze data, only eight of the fifteen predictors, and neither of the interactions, were found to be statistically significant predictors of injury severity. The eight significant predictors included income, child age, child race, perpetrator gender, perpetrator history of violence, perpetrator alcohol and/or drug involvement, perpetrator race, and source of referral. When this model was applied to the cross-validation sample, however, it fared poorly, predicting no cases in the severe injury or death categories. While statistically significant, the overall model was able to predict injury severity only six percent better than chance.

A test of the distributional assumption of ordinal logistic regression for the estimation sample revealed that the model was indeed misspecified, and that nonlinearities, interactions, and additional predictors should be included in future models.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS