A Creative Use of the "Further Research Needed" Concept

Another issue that requires more consideration is the reason for the discrepancies between our findings and the previous results of Harcourt and Ludwig (2006). We employ a similar modeling strategy, regressing crime changes on arrest changes, yet arrive at divergent conclusions. One obvious difference between the two studies is the dependent variable. Homicide and robbery serve as the dependent variables in the analyses reported above, whereas Harcourt and Ludwig model a composite measure of violent crime based on the sum of criminal homicides, rapes, felonious assaults, and robberies. However, when we substitute a composite violent crime measure in our regression models, the results for misdemeanor arrests and the cocaine measure replicate those for homicide and robbery. The differences across studies for the cocaine measure are perhaps not surprising, given that our proxy is measured at the precinct rather than at the borough level, but the varying results for misdemeanor arrests are perplexing. We have included a range of control variables similar to those of Harcourt and Ludwig, but some differences in model specification exist. An important task for future research is to isolate those design features of the respective studies that can account for the divergent findings.

Next: "What happens when you actually regress the outcome on the independend variables is an important task for further research."