| Abstract: | Studies of empirical relationships between indicators of prison crowding and inmate violence have uncovered null, negative, and positive relationships. These mixed findings may be due, in part, to cross‐study differences in definitions of crowding, levels of analysis, and sample designs. We compared findings across some of the more popular approaches to study the relationship between facility crowding and the prevalence of inmate assaults in order to determine the implications of different methods for variation in estimates. Multi‐level data from a national sample of 10,022 men confined in 203 state correctional facilities during 1997 were examined. Findings revealed differences across methods in the direction and significance of the crowding/assault relationship. These differences were then considered in order to derive a strategy for more uniform research on the topic. This strategy consists of including both total inmate population and design capacity as separate predictors in the same model, examination of tri‐level data (inmates, facilities, and states) in order to control compositional differences in inmate populations across facilities and to remove confounding state‐level differences in crowding levels and assault rates, and more careful consideration of secondary analyses of complex samples with sample weights. |
