NCJ Number
225313
Date Published
November 2008
Length
30 pages
Annotation
This study proposes a new approach for evaluating techniques for imputing missing data in the analysis of data for the Supplemental Homicide Reports (SHR), which are widely used in criminological research.
Abstract
Scholars have argued convincingly that the problem of missing data in the SHR poses a significant obstacle to research. Although the general focus has been the absence of information pertaining to the relationship between the victim and the offender, some scholars have noted that missing information precludes researchers’ ability to look at race-specific and motive-specific patterns of offending. A common solution to this problem--discarding cases with missing data--can introduce significant bias into quantitative models and should be avoided whenever possible. Although this study does not provide a direct comparison of the strategies for addressing missing data in homicide cases, given their different data requirements, the authors outline what they view as the strengths and weaknesses of the strategies. They generally conclude that imputation strategies that aim to create distributions for cases with unknown values can be useful. In contrast, attempts to impute values for individual cases were problematic; none of the attempts were much more accurate than coding all cases with a missing “victim/offender (v/o) relationship” as acquaintance homicides. Imputation methods have advantages over dropping cases. Still, more research on applications is needed. In the first part of the authors’ analysis, they compared distributions of available variables for the three types of cases: complete in SHR, missing offender and incident information in SHR but known in the police report, and missing offender and incident information in both SHR and the police report. The second part of the analysis applied the four imputation strategies to “incomplete” datasets from the SHR. 8 tables, 28 references, and 1 appendix