RESEARCH STUDY
Domestic Violence Inventory Predictive Ability

BACKGROUND

Predicting the likelihood that domestic violence offenders will reoffend using traditional methods is often hampered by the lack of data upon which to base the predictions. This is especially true for first-time offenders who often have very little criminal history. Part of the problem is that traditional methods rely heavily on criminal history information. A sufficient amount of information is required in order to use traditional predictions. On the other hand, contemporary methods for predicting recidivism use techniques that employ many variables in prediction equations that do not rely on criminal history alone and can be accurate even with an absence of any criminal history information. Multiple regression analyses determine the best combination of possible predictor variables to produce prediction equations. A multiple regression approach to predicting domestic violence arrests using multiple offenders’ data can establish prediction equations for first-time offenders. Offenders who have committed multiple domestic violence offenses have criminal history data to develop prediction equations. These prediction equations can then be applied to first-time offenders who lack criminal history.

 CRIMINOGENIC NEEDS

An over-reliance on criminal history in prediction equations means that a first-time offender’s prediction of recidivism will be underestimated. However, incorporating other offender-related information, such as demographics, criminogenic needs and behavioral variables along with selected court history, offers accurate prediction. The Domestic Violence Inventory is specifically designed for domestic violence offender assessment and recidivism prediction. It includes predictor variables, such as offender demographics, behavioral variables (criminogenic needs) and criminal history, necessary to predict offender recidivism of domestic violence arrests.

 MULTIPLE REGRESSION

Behavioral variables are the criminogenic needs variables measured by Domestic Violence Inventory (DVI) scales. These variables include alcohol and drug problems, violence potential, controlling or dominating behaviors and stress coping abilities. Demographic predictor variables include age and education level. Other demographic variables, such as gender and race, are not needed as predictors because they are accounted for in DVI scale scores. Court history variables include age at first arrest, number of times arrested, number of times on probation or parole, probation or parole revocations, number of times sentenced to jail or prison, number of assault, alcohol and drug arrests. In a preliminary analysis of domestic violence recidivism that included 226 adult domestic violence offenders, the predictors found to be important (Multiple R = .938, p<.001) for predicting domestic violence arrests were, age, age at first arrest, total number of arrests, assault arrests, times on probation and all DVI scales (Alcohol, Drugs, Control, Violence and Stress Coping Abilities). Because demographic and criminogenic needs information are obtained from first offenders, prediction equations can accurately predict recidivism because there is little reliance on criminal history.

Multiple regression is a statistical procedure for combining several variables in an equation for predicting a target variable. The regression coefficient is the statistic for the closeness of fit for the relationship between predictors and the target. The regression coefficient can vary from zero for no relationship to 1 for perfect prediction. In this case, the number of domestic violence arrests is the target variable to be predicted.

 EQUATIONS

Another prediction equation study was conducted using DVI test results for 1,430 multiple offenders’ (2 or more domestic violence arrests). Because of the small range of data (98% had 6 or less) domestic violence arrests were combined with total number of arrests. These multiple offenders were divided into two groups. Group 1 (N=683) served to validate the prediction and Group 2 (N=747) was used to develop the prediction equation. The predictor variables included the demographics and court history variables listed above with DVI scales scores. The multiple regression coefficient was highly significant, F=68.75, p<.001, Multiple R=0.82.

 Predicted scores were calculated for Group 1 using the Group 2 prediction equation. Frequency distributions of the predicted scores were compared between the two groups. Group 1’s predicted scores were within 4.1 percent of Group 2 predicted scores. The predicted scores for both groups were very close to being identical. Group 1 predicted scores were highly statistically significantly correlated with actual number of arrests, r=.78, p<.001. The correlation coefficient between predicted scores and actual arrests demonstrates that the predicted scores are very accurate.

 Applying these multiple offenders’ results to first offenders can provide an estimate for the probability of a first offender reoffending. The higher the predicted score is the more likely it is that an offender will reoffend. But what constitutes a high probability for reoffending? Again multiple offenders’ results can be used to establish problem probabilities or, in other words, a threshold for high likelihood of reoffending. The predicted score gives a direct estimate of the likelihood of reoffending whether an offender is a multiple offender or first offender. Predicted scores above the threshold indicate a high likelihood of reoffending.

SUMMARY

This study developed a prediction equation for domestic violence arrests and applied this equation to first time offenders to estimate the likelihood of committing another domestic violence offense in the future. The prediction equation was developed and validated with multiple offenders. Threshold for problem probability for reoffending was also developed from multiple offenders’ results. The results of this study demonstrate that domestic violence arrests can be accurately predicted.

This “predictive ability” study represents the beginning of Risk & Needs recidivism prediction efforts in the domestic violence offender field. Since this research began we have attempted to identify significant contributing factors for our recidivism predictor equation. We have explored demographics, criminal history and criminogenic needs. Since these studies appear promising, Risk & Needs is interested in participating in focused longitudinal studies and will contribute DVI support in our, others and collaborative or joint research. Additional information can be provided upon request. Our goal is accurate domestic violence offender recidivism prediction.