Revisiting the Use of Total Incarceration Variables in Sentencing Models

• In Event: Beyond a Simple Regression: The Less Visible Disparities in Sentencing Outcomes

Abstract

A very common practice in quantitative sentencing research is the use of dependent variables based on incarceration, either the dichotomous in/out status or the continuous length of incarceration. In either case, the researcher would focus on the difference between incarceration and non-incarceration outcomes, rather than the difference between prison and jail. However, Holleran and Spohn (2004) argued that doing so can lead to the underestimation of extralegal disparities in the court system, as prison and jail reflects different sentence conditions. We revisited this issue using a more recent dataset from Miami-Dade County, Florida. We estimated a set of multinomial logistic regression models and quantile regression models that differentiates prison and jail outcomes, and compared the findings with those from models using the total incarceration variables that mix prison and jail together. In both cases, accounting for the difference between prison and jail revealed additional extralegal disparities that would have been invisible under the total incarceration variable approach. The findings suggest that the sentencing outcome is more complicated than the simple incarceration and no incarceration dichotomy.

Authors

• Shi Yan, Arizona State University

• Andrea N. Montes, Arizona State University

• Marisa Omori, University of Missouri - St. Louis

• Nick Petersen, University of Miami

• Cassia Spohn, Arizona State University