U.S. flag

An official website of the United States government, Department of Justice.

Nonparametric Imputation Approach for Dealing with Missing Variables in SHR Data

NCJ Number
Homicide Studies Volume: 8 Issue: 3 Dated: August 2004 Pages: 255-266
Date Published
August 2004
12 pages

This study examined an imputation strategy that appears to offer a viable alternative approach to the Supplementary Homicide Reports (SHR).


This study examined the procedures developed by Williams and Flewelling (1987), the distributional assumptions concerning variables for which there are missing data were inferred from known cases in the same jurisdiction but done separately within several categories of circumstance. The rational for this strategy was based on the fact that certain types of homicide, as defined by their circumstance code, were disproportionately represented among incidents with missing offender information. These homicide types exhibited significantly different distributions on offender characteristics among the cases where offender information was present. The accuracy of imputation was improved by allowing different adjustments to be made for missing data based on circumstance codes of the cases with the missing data. In reviewing those adjustment procedures, and in applying them to a study of trends in youth-perpetrated homicide rates, several enhancements were developed. First, it was observed that among the records with no missing data, victim variables were moderately related with offender characteristics and the victim-offender relationship. The second modification to the original procedure involved expanding the number of incidents used to extrapolate the distributions of the missing variables. The new procedure specifies that the distributions used as the basis for imputation be derived from all homicide events within the United States Census-designated geographic division of the incident in question, aggregated over a 5-year period. A limitation to the described procedure is that the optimal set of covariates is unknown. Another important limitation of the adjustment procedure is that the characteristic of having missing data does not influence the imputed values. To further access the validity of current missing data adjustment approaches and improve the accuracy of these techniques further analysis of large-scale studies is needed to determine the actual characteristics of events that are initially unknown and therefore not recorded in the SHR. 1 table, 12 references

Date Published: August 1, 2004