NCJ Number
206505
Journal
Homicide Studies: An Interdisciplinary & International Journal Volume: 8 Issue: 3 Dated: August 2004 Pages: 163-192
Date Published
August 2004
Length
30 pages
Annotation
This study examined missing data in homicide research.
Abstract
The purpose of this study was to integrate the contributions of this journal into the broader context of the literature on missing data. The difficulty is that there is very little consensus as to how the literature should be organized. There are a multitude of distinctions and classifications that serve particular purposes, but none seem to have widespread consensus. There are two types of incomplete data: item nonresponse and unit nonresponse. Item nonresponse occurs when an item is not present or if it is unusable. Unit nonresponse occurs if a unit is relevant to the sample, but is not included. Although the distinction between item and unit nonresponse is important, it is not very useful in the classification of missing data techniques because some techniques can be used for both item and unit nonresponses. This study was divided into two major sections. The first section provides a brief overview of missing data techniques using a classification offered by Little and Rubin (1987). The second section provides a brief description of the articles included in this issue dividing them into the following categories: procedures based on completely recorded units, imputation-based procedures, weighting procedures, and model-based procedures. Rubin and his colleagues have defined the different kinds of assumptions that can be made concerning patterns of missing data. The first of these assumptions is that data are Missing Completely at Random (MCAR). The data are said to be at MCAR if “the probability of missing data on Y is unrelated to the value of Y itself or to the values of any other variables in the dataset.” When this is true for all variables in the data set, the data are a random sample of the original subset of observations. A somewhat weaker assumption is that the data are Missing at Random (MAR). If the data are MAR the probability that an observation is missing can depend on the value of the observed items, but not the whole of the missing item itself. If the data are neither MCAR or MAR and the missingness cannot be predicted from the other variables present in the dataset, the missing mechanism is nonignorable. Future research suggestions include: improving understanding of how and why data are missing, discussion and research comparing imputation techniques should be improved, established methods of imputation will have to be transparent and accessible, and a system of accountability and sanctions will ultimately have to be used to reduce the problem of missing data in homicide to a minimum. Tables, references