Abstract
Using national data from felony cases processed in state courts (n = 48,006), the current study investigates the nature and magnitude of contextual variability associated with sentencing outcomes. Multivariate models are first estimated to identify the main effects of various offender and offense variables on sentencing decisions. Conjunctive analysis is then used to evaluate the contextual variability of each of these main effects across all observed combinations of offender and offense attributes. Separate analyses are also conducted among states with and without mandatory sentencing guidelines to explore whether these guidelines reduce this variability across different contexts. Findings from this study and its comparative methods are discussed in terms of implications for future research on criminal sentencing and assessing the contextual variability of the main effects of particular legal and extralegal factors.
Notes
1. For example, under a multivariate model of main effects, the finding of an estimated 20% gender difference in the risks of imprisonment represents the adjusted mean differences between men and women that derive from partitioning out the influence of the covariates of gender and then calculating the average effect across all the other variables in the analysis. It is in this way that the analytic methods underlying a main-effects model necessitate the assumption that the predicted effect of any particular variable is represented by a constant value that is invariant across all possible contexts. For a recent application of this type of multivariate model of main effects, see Brennan and Spohn (Citation2008).
2. To adjust for skewed distributions, statistical models with log transformations of sentence length were also estimated in this study. There were no differences in the substantive conclusions across these different specifications of the functional form of the models. We present the untransformed results because the interpretation of group differences in sentencing length across contexts is more intuitive in this form (i.e. differences in months of imprisonment vs. the log of months). The use of median (rather than means) as measures of central tendency for box plots also decreases the impact of skewed distribution on the conjunctive analysis of the length of imprisonment because the median is less influenced by extreme scores.
3. The only exceptions to the conclusion of similar results across methods involve the net impact of mandatory guidelines and Hispanic offenders on the risks of imprisonment. In particular, the net effect of each of these variables was statistically significant (p < .05) and associated with lower risks of imprisonment when listwise deletion was applied to missing data. However, the lower net risks for Hispanic offenders had not achieved this conventional level of significant (p = .06 rather than p < .05) when the dummy variables for missing data were included. When multiple imputation was conducted, the net negative impact of mandatory sentencing was more pronounced. No other differences were found across the different methods for the treatment of missing data.