The effect of the federal sentencing guidelines on inter-judge sentencing disparity.

• Author: Hofer, Paul J.
• Position: Statistical Data Included

1. INTRODUCTION

   1. THE GOALS OF SENTENCING REFORM

      The sentencing reforms of the past twenty-five years have had several goals, including "truth in sentencing," control of prison populations, and reduction of unwarranted disparity. The first goal was easily achieved when parole was abolished and the sentence imposed became the sentence served. Control of prison populations proved more difficult, and careful evaluation is needed to determine whether sentencing reform helped to check, or may have accelerated, the steady rise in prison inmates and crowding of prisons.(1) Among the most important goals motivating reform at the federal level was reduction of unwarranted sentencing disparity.(2) Yet more than ten years after enactment of the Sentencing Reform Act of 1984 and implementation of the U.S. Sentencing Commission's guidelines, there is no consensus on whether they have reduced unwarranted disparity in federal sentences.

      As reviewed below, the evidence is persuasive that, in the pre-guideline era, differences among judges in sentencing philosophies were the primary sources of unwarranted disparity. There are reasons to hope that the federal guidelines have been effective at controlling these differences: they are the most detailed guidelines ever developed, and they are mandatory. Judges can depart from the guidelines only for limited reasons that must be stated on the record, and that are then subject to appellate review. But these departures from the guidelines, as well as plea agreements that ignore the guidelines and inconsistencies in application of the guidelines, might reintroduce disparity into the system.

      In this paper, after reviewing problems that have plagued earlier research, we conclude that the "natural experiment" created by the random assignment of cases to judges in many courthouses around the country provides the best opportunity to evaluate the success of the guidelines at reducing disparity. Although the natural experiment approach does not permit evaluation of disparity in individual cases, and does not permit direct examination of prosecutor-created disparity, it does allow us to draw some conclusions about the effect of the guidelines on inter-judge disparity. In particular, it permits precise measurement of changes in the "primary judge effect"--the overall tendency of some judges to be more lenient or severe than others, as measured by differences in average sentences among judges with comparable caseloads. We believe this is an important barometer of the success of the guidelines because it represents the "tip of the iceberg" of disparity that would be observed--and is observed in simulation studies--when identical cases are sentenced by different judges.

      What we find in the results of this natural experiment is that the guidelines have significantly reduced overall inter-judge disparity in sentences imposed, much as the parole guidelines reduced disparity in time actually served prior to implementation of the sentencing guidelines. Together with the other research reviewed below, these findings suggest that the sentencing guidelines have had modest but meaningful success at reducing unwarranted disparity among judges in the sentences imposed on similar crimes and offenders.

      The success, however, is uneven. Some types of cases show no improvement, or show improvement in some cities but not in others. Further, there is evidence that some regional sentencing differences have increased under the guidelines, particularly in drug trafficking cases. These results demonstrate the need for further exploration of the sources of these remaining disparities, and of what additional changes are needed to make fair and uniform sentencing attainable for all types of cases in all places.

   2. DEFINING AND MEASURING UNWARRANTED DISPARITY

      To evaluate the guideline's success at reducing disparity we first must define the problem. A truism of sentencing research is that sentences should vary according to the seriousness of the crime and the dangerousness of the offender, but that "unwarranted disparity" is undesirable and unfair. Deciding what types of variations count as unwarranted disparity has been controversial, however. Most people readily agree that dissimilar treatment of similar offenders would be unwarranted, just as similar treatment of dissimilar offenders would improperly fail to reflect relevant distinctions among them. But this definition is purely formal, and thus incomplete, because it does not tell us the proper criteria for classifying offenders as similar or dissimilar.(3)

      Treating offenders differently based on a legally impermissible ground, such as race, results in the clearest case of unwarranted disparity--discrimination.(4) Discrimination may reflect intentional or conscious bias toward a group, or be a result of the distorting effects on rational judgment of unconscious stereotypes or fears. Either way, it is the most onerous type of unwarranted disparity and sentencing reform was clearly designed to eliminate it.(5) But discrimination is not the only type of unwarranted disparity.

      To most people, different treatment based on any factor that is irrelevant to the rules and purposes of sentencing--so-called "extralegal" factors, such as an offender's physical appearance--is unwarranted. The most common statistical approaches used to study disparity, multiple regression and related analyses, are compatible with this definition. In this research, as many of the legally relevant factors as is practically possible are measured for a large sample of cases, along with additional factors that are not legally relevant but that may help explain the sentences imposed.(6) These explanatory factors are then statistically correlated with one or more outcome measures, typically whether the offender received probation or imprisonment and the length of any prison term imposed. The amount of variation in sentences that is correlated with the legally relevant factors is statistically removed, or "controlled for," and the remaining variation is examined.

      For example, it may be that, on average, offenders receive six months more prison time for each prior conviction. The variation in sentences correlated with prior convictions is subtracted from the outcome measure, and the remaining variation in prison lengths is examined to determine what it correlates with. This process is repeated for all of the legally relevant factors. If the remaining variation correlates with an impermissible factor, such as race, the data are consistent with discrimination in sentencing. Other types of unwarranted disparity are suggested if factors such as the region in which the case was prosecuted are correlated with sentences. Variation that is not accounted for by any of the factors in the model may also be taken as a measure of unwarranted disparity, although other problems affect this measure.(7)

      A long line of regression studies of this kind, even before implementation of sentencing guidelines, has found that the most important influences on sentences are not impermissible factors such as race, or extra-legal factors such as region, but are the legally relevant factors that should determine sentences, particularly offense seriousness and an offender's criminal history.(8) Statistical analyses conducted at the U.S. Sentencing Commission subsequent to implementation of the guidelines reaffirmed that legally relevant factors remain the most important determinants of sentences today, explaining as much as 70% of the variation in imprisonment lengths. While such findings are reassuring, they do not obviate the need for further research and vigilance against unwarranted disparity, because most people think that any significant influence of legally irrelevant factors would be undesirable, even if these factors are not the primary determinants of sentences.(9)

      In fact, sentencing reform seems to have been a boon to research on disparity using regression approaches, with over ten such studies appearing in recent years on just the federal guidelines alone.(10) This type of research is popular because the guidelines explicitly identify many, if not most, of the factors that are legally relevant to the sentencing decision, and because sentencing commissions have collected huge amounts of data on these factors. But for several reasons, we believe that these studies can add relatively little, if anything, to our understanding of disparity under the guidelines, and are even less helpful in evaluating whether the amount of disparity has increased or decreased.

      First, there are significant methodological obstacles to using regression studies for measuring unwarranted sentencing disparity under the guidelines, and especially for comparing the amount of disparity before and after guideline implementation. Many of these obstacles are discussed in a later section. But perhaps the most significant for present purposes is the difficulty of obtaining comparable data for both time periods.

      Second, theoretical disagreements may arise over whose views about a factor's legal relevance should prevail.(11) For example, whether different treatment is warranted based on offenders' potential for rehabilitation depends on whether one believes rehabilitation is a legitimate purpose of sentencing. Views of rehabilitation have clearly changed over time, and disagreement over its role in sentencing still exists. We believe that because the Sentencing Commission is now legally charged with deciding what criteria should be legally relevant, it is appropriate to evaluate current and past practice in light of the Commission's policy decisions. At the same time, we recognize that what is considered disparity today has not always been considered disparity, and that some ways of evaluating past practice in light of current policies may "stack the deck" in favor of finding a reduction of disparity.(12)

      In addition, even if we agree that the...