The Effect of Missing Item Data on the Relative Predictive Accuracy of Correctional Risk Assessment Tools

Area of Science:

• Criminology
• Psychometrics
• Statistics

Background:

• Missing data are common in risk assessment tools.
• The impact of missing data on predictive accuracy is understudied.
• Simpler methods like summing or proration are often used, but multiple imputation is theoretically superior.

Purpose of the Study:

• To compare the validity of multiple imputation, summing, and proration for handling missing data in risk assessment.
• To investigate the impact of varying percentages of missing data on predictive accuracy.

Main Methods:

• Utilized STABLE-2007 (N = 4,286) and SARA-V2 (N = 455) datasets from Canadian men on community supervision.
• Introduced missing data across six conditions, ranging from 1% to 50% deletion.
• Compared the relative predictive accuracy of three imputation techniques: summing, proration, and multiple imputation.

Main Results:

• Relative predictive accuracy was not significantly affected by the amount of missing data.
• Proration and multiple imputation performed comparably in terms of predictive accuracy.
• Summing available items led to an underestimation of absolute risk.

Conclusions:

• Proration is an empirically justified and effective method for handling missing data in risk assessment samples.
• Simpler techniques like proration can be as valid as multiple imputation for maintaining predictive accuracy.
• Risk assessment practitioners can confidently use proration when faced with missing data.