Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
6 Research Agenda for the Study of Sentencing GENERAL RESEARCH STRATEGY The rapid pace of changes in sentencing policies and practices has stim- ulated research designed to assess those changes and to aid future re- forms. If it is to be useful to policy makers, future research on sentencing must balance short-term and long-term perspectives and research goals, capitalize on the natural experiments associated with changes in sen- tencing, improve the quality and availability of data, and use a variety of methodological approaches. A BALANCED PROGRAM A balance between quick-response, highly targeted research projects and longer-term, more basic efforts is needed. Changes in sentencing policy must be viewed from a longer-term and broader perspective so that results of policy shifts can be assessed in the context of larger social processes and changes. For example, crime rates rose sharply in the 1960s and early 1970s and leveled off in the late 1970s. Meanwhile, during the 1970s deterrence became an increasingly important goal of the criminal justice system. Some have argued that the leveling off of crime rates in the late 1970s was attributable to the deterrent effects of sanctions. However, the changes in the crime rate could also have been related to demographic shifts associated with the postwar baby boom and a variety of other social changes. Understanding the changes in 259
260 RESEARCH ON SENTENCING THE SEARCH FOR REFORM crime rates requires partitioning these, and other, possible causes, which in turn necessitates long-range as well as short-range studies. Historical and theoretical analyses of long-range changes, in addition to their in- trinsic value, are necessary to provide a context for interpreting the results of specific, short-range policy evaluations. CAPITALIZING ON NATURAL EXPERIMENTS THE IMPORTANCE OF TIMING Because of the limited opportunities for planned or designed interven- tions, evaluations of the impact of policy changes must rely heavily on natural experiments. Natural experiments arise, for example, when a prosecutor decides to test a new case-screening policy or a legislature enacts a new sentencing statute. In supporting studies and evaluations of natural experiments, funding agencies are caught in a dilemma. To provide useful and timely information to the criminal justice community, programs and policy changes must be evaluated promptly. Because of the variety of potential changes that might be evaluated, the need to collect baseline data on operations before a change is put into effect, and the amount of time required to develop and publish requests for proposals, select a contractor, and implement a research project, funding agencies must anticipate changes and support short-term evaluations. But in so doing they risk jumping the gun, selecting the wrong program to evaluate, and supporting evaluations and studies that are completed before the changes being studied have been fully implemented and op- erations have become normal. One promising strategy for capitalizing on natural experiments, eval- uating them promptly, and minimizing the costs of false starts would be the creation of an ongoing center with operational and technical ex- pertise to identify opportunities for experiments and to advise on the formulation and execution of study designs. The National Institute of Justice should create such a center. The center could have discretionary funds to award selected applicants up to $10,000 to conduct early fea- sibility studies to determine the existence and viability of research op- portunities arising from policy changes. Applicants for feasibility grants could submit informal proposals providing adequate evidence that (1) relevant data are available and accessible, (2) qualified research staff are available to pursue the opportunity, and (3) the opportunity is worth pursuing as a longer-term study. Feasibility studies could establish the key variables, examine the availability of baseline data, and develop a detailed design of the research project that would then be submitted as a proposal to the National Institute of Justice for longer-term funding.
Research Agenda for the Study of Sentencing 261 A small center staff, augmented by consultants, could screen opportu- nities as they arise by reviewing submissions on a continuous basis (pro- viding approval or rejection to applicants within a few weeks) and pro- viding technical assistance to grantees in the development and execution of their research designs. Such quick reaction and initial assessments of feasibility would permit timely screening of opportunities to prevent premature funding of extensive and costly evaluations prior to adequate exploration of their potential and problems. DATA NEEDS Modest improvements in the data available in the existing data series and statistical systems maintained by various jurisdictions and the Bu- reau of Justice Statistics (BJS) could significantly enhance the oppor- tunity for answering questions related to sentencing policy. An impor- tant limitation of the FBI's Uniform Crime Reports (UCR), for example, is the absence of a full characterization of the demographic attributes of arresters, particularly by age, race, and sex, simultaneously parti- tioned for each offense type. Such a full partition would permit com- parison of the consistency of the attributes of arresters with those of prisoners. The development of such characterization should be consid- ered in the impending reassessment of the UCR being undertaken by the Bureau of Justice Statistics and the FBI. Easily accessible data on prosecutorial and court processing are rarely compiled, thereby making it difficult for researchers to follow cases systematically through the criminal justice process. In view of the varying quality and sophistication of prosecutorial and court data systems, the most fruitful strategy for developing indicators of court-processing char- acteristics would be to support further development of more uniform and consistent data and management information systems in those ju- risdictions already collecting such data rather than by attempting a uni- form nationwide collection system. On corrections, the Prisoners in State and Federal Institutions series provides annual data on gross prison populations by state but does not include a detailed breakdown of the composition of the prison popu- lation in each state. The Profiles of State Prison Inmates provides val- uable detailed individual data on a national sample of prisoners in cor- rectional institutions, but these data are available only for 1974 and 1979, and they cannot be disaggregated by state; hence, they cannot be used by individual states for their planning purposes. State-level dis- aggregation, at least for the larger states, should be included in subse- quent national surveys. Furthermore, these two sources (the annual
262 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM counts and the detailed profiles of inmates) cannot be combined to study offenders in the various state prison systems by crime type, sentence length, and demographic characteristics. For example, one cannot de- termine from existing national data the changes over time or variations across jurisdictions in the demographic characteristics, prior conviction records, offense types, and sentence characteristics of prisoners. In order for legislatures to consider the impact of sentencing policy on prisons, it is necessary to estimate the consequences of a particular policy, for example: imposing a particular determinate sentence for specified groups of offenses or imposing a particular mandatory minimum sentence for offenses involving weapons or for second-time felons. While each state can be expected to formulate its own estimation models and to collect its own data for this purpose, a national project should be organized to foster such developments, including surveying the provisions being considered in var- ious sentencing legislation and designing for common use a standard sam- pling and data collection protocol that could be easily adapted by any state considering changes in sentencing policy. Census data on jail inmates are far cruder than those for prison in- mates, both because of the rapid turnover of the jailed population and because of the large number and variety of jail facilities across the nation. Attention should be given to a periodic sample of jail populations to learn more about their composition and how it changes over time. This kind of data would be especially important when significant changes in sentencing legislation are being considered. Such changes are likely to affect the plea-bargaining process and, as a result, the jail populations in pretrial detention. For example, stiffer sentences such as those called for by a mandatory minimum sentencing law may encourage more of the people vulnerable to the terms of the law to demand a trial, and this might slow processing through the courts and so increase the pretrial detention population. Detailed surveys of jail populations should be taken before a major sentencing change and at several points following its implementation. Aside from assessments of the direct impact of leg- islative changes on jail populations, analyses of such survey data could shed some light on the~plea-bargaining process. In most jurisdictions, only highly aggregated data are collected on the number of cases disposed of by the courts, on flows into probation or prison, and on releases on parole. Even a partition by offense type is often not available. These crude data cannot provide useful information on the effects of alternate sentences or sentencing policies to the key participants in the sentencing process. These objectives are best pursued with individually based data that record attributes of defendants, their
Research Agenda for the Study of Sentencing 263 offenses, and their prior records, and follow the movement of their cases through the criminal justice system. Offender-based transaction statistics (OBTS) systems found in a number of jurisdictions are intended to collect such data, but very few jurisdictions maintain such data com- pletely and reliably. Recognizing the political, logistical, and fiscal con- straints on widespread development and improvement of such data sys- tems, it is important, at a minimum, to focus attention on the most serious offenders e.g., those indicted for serious offenses- and to de^- velop an individually based statistics system for them. At a time of severe budget cuts, the identification of a minimum uniform core of standard data items on case processing through the court and corrections systems for collection across jurisdictions becomes particularly important. Standard items would include the basic demo- graphic attributes of an offender; a characterization of the key elements in prior record; the current arrest charges, including certain key attri- butes of the offense (e.g., weapon use); whether there was a charge reduction associated with a guilty plea; and final disposition data, in- cluding conviction charges, sentences, and data on the execution of the sentence, such as date received and released by supervising agencies. Improved and expanded common data bases containing two important classes of information at the state level are needed to advance knowledge about sentencing: cross-sectional aggregate information on numbers of crimes and on the processing of suspects, defendants, and offenders by the criminal justice system; and disaggregated longitudinal information on the processing of individual cases from arrest through the courts and into the corrections system. Most jurisdictions collect the former but each in its own idiosyncratic way; only a few collect the latter. Such disaggregated data are essential for understanding the outcomes of the criminal justice process and for developing projections of the impact on prison populations of various policy options. A number of states have pursued this objective through OBTS systems. Development and sup- port for improved uniform systems should be continued and other states encouraged to set up such systems. In addition, it would be useful to invest some research effort in studies of the quality of these sorts of administrative data. If official data are to figure significantly in future research, their strengths and weaknesses should be better documented. Much has been learned, for example, from comparisons between the National Crime Survey (an annual survey of criminal victimization) and the Uniform Crime Reports data, and these kinds of projects should be continued. In addition, similar com- parisons are needed for court and prison statistics.
264 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM DIVERSIFYING RESEARCH APPROACHES Research based on observation and interview techniques is an essential aspect of a broad research strategy designed to illuminate the sentencing process. Quantitative methods alone cannot adequately capture or con- trol for many features of social interaction, the importance of the sub- jective dimension of behavior, the need to tap the meaning of a situation in shaping decisions made by actors in the criminal justice system, and the effects of subtle behavioral cues. Qualitative techniques are partic- ularly valuable in generating hypotheses about how and why actors behave as they do, in interpreting the meanings individuals give to their decisions, in clarifying relationships and patterns of interaction among actors in various criminal justice agencies, and in describing the informal decision rules by which agencies and their officials create a gap between the law on the books and the law in action. Qualitative approaches to learning about prosecutorial decisions which are fundamental to sen- tencing and very poorly understood are especially important. Since a choice of research method must be related to the research question, no single approach can be given highest priority in all situa- tions. However, we believe there has been too much reliance on simple statistical analyses of cross-sectional data, and we urge vigorous efforts to broaden the range of methods used in the study of sentencing. These should include qualitative studies that provide new insights and hy- potheses for further testing and, when appropriate, greater use of ex- perimental and quasi-experimental designs that will permit causal in- ferences associated with specific operational changes. DETERMINANTS OF SENTENCES DISCRIMINATION It is not likely that research will provide a definitive estimate of the influence of racial discrimination on sentence outcomes in general. A more useful alternative to a global approach is to focus on the effect of race in particular jurisdictions, time periods, and sets of circumstances. Furthermore, the methodological problems that impede knowledge about the effect of race also characterize efforts to assess the impact of sex, age, and socioeconomic status on sentence as well as the effect of case- processing variables such as attorney type and type of plea. We have defined sentencing broadly to include a series of decisions affecting cases as they pass through the criminal justice system. In ad- dressing the issue of discrimination in sentencing, several problems must
Research Agenda for the Study of Sentencing 265 be addressed. Detecting the presence of discrimination based on race, socioeconomic status, sex, or some other case attribute is difficult be- cause the effect of discrimination is often small in comparison with the effects of current offense and prior record. Hence, sensitive measure- ments of these latter variables are required to discover small effects. This problem is further complicated by aggregation effects, which may mask individual instances of discrimination for certain crime types, in certain jurisdictions or courts, or on the part of individual decision makers. These different settings and circumstances should be explicitly ex- amined in future research through use of sufficiently disaggregated data. This research should also explore the role of situational variables in different jurisdictions in making inappropriate factors more or less sa- lient in the sentencing decision. To reduce the risk of selection bias, which can be a problem when one examines only sentence outcomes, research on discrimination should examine the handling of cases as early as possible in the criminal justice system. Studies of discrimination should emphasize the treatment of less serious offenses, which offer greater room for discretion and greater opportunity for discrimination, and should examine in detail the various stages between arrest and imprisonment to discern the degree to which discrimination exists at any of these intermediate stages. Research on discrimination requires a variety of complementary meth- odological approachesincluding structural modeling, longitudinal studies, statistical analyses of aggregate court-processing data, obser- vation and interview techniques, and experimental and quasi-experi- mental designs to supplement the cross-sectional studies of discrimi- nation that currently predominate. The social importance of discrimination suggests the need for contin- ued research on this topic, but this research should not simply continue existing approaches to this topic. New studies should be designed to answer questions like the following: In what ways are particular groups of offenders benefited or disadvantaged by differential treatment? Is discrimination related to crime type or offense seriousness? Is discrim- ination related to victim-offender relationships or case-processing var- iables? In which jurisdictions or types of jurisdictions is discrimination found? CASE-PROCESSING VARIABLES The role of case-processing variables in sentencing is also a fertile area for further research. The belief that a guilty-plea discount exists and is
266 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM necessary tor court functioning is an important source of support for plea bargaining. However, the extent of such a discount remains un- known. One useful approach to the discount question is an examination of the defendant's decision to plead guilty or go to trial, a topic on which only limited research is currently available. It is assumed that the dif- ferential between sentences in pled and tried cases accounts for the decision to plead guilty, but many other factors (e.g., the expense of a trial attorney, loss of time from work while in court, and the cost of time spent in jail prior to a trial) may also contribute to the decision. Further research examining why defendants plead guilty is desirable. This research should look across jurisdictions and courtroom cultures and should separate case-processing from other variables. Heumann's (1978) suggestion that there are subtleties in sentence discount policies needs further investigation. In particular, research should explore the extent to which court personnel in various settings distinguish between "dead bang" cases, in which conviction is a virtual certainty and for which the defendant may be "punished" for going to trial, and cases in which there is a real factual dispute and the legal ambiguities are felt to justify the additional expense of a trial. An additional question for research is whether the elimination of plea bargaining leads to an increase in trials. Rubinstein et al. (1980) report that the ban on plea bargaining in Alaska resulted in only a slight increase in the number of trials with no backlog of such cases. However, a differential between sentences for those convicted at trial and those who pled guilty may have discouraged trials and may actually have resulted in a shift from overt to covert bargaining. Another approach that might be considered for studying the plea/trial sentence differential involves gathering data on the final sentence offers of the prosecution in a set of cases. Among those that go to trial, the sentence imposed after trial can be compared with the sentence that would have been imposed had the defendant agreed to plead guilty. Such a research strategy has a set of natural controls for case attributes since it involves the same cases. Two difficulties arise in this type of research. First, since most cases involve guilty pleas, such a research strategy requires prospectively gath- ering data on a large number of cases in order to obtain enough trial cases for analysis. (A retrospective approach is unlikely to work because offers are not recorded and participants' recall is not sufficiently good.) Second, interpretation of a sentence differential if it emerges is not simple. A harsher sentence after trial may indicate punishing defendants for failing to plead guilty, but other explanations may also account for
Research Agenda for the Study of Sentencing 267 the finding: the trial may bring out details about the crime (particularly aspects indicating a heinous nature) or more extensive prior-record in- formation; when defendants take the stand and deny their guilt, judges or juries may be punishing them for two crimes, the one for which they have been convicted and the crime of perjury; or defendants' failure to take the first step toward rehabilitation that admission of guilt is some- times said to imply may be the basis for harsher sentences after trial. Interviews with judges after they sentence trial cases might illuminate the extent to which a differential is produced by the mode of disposition itself as opposed to the other factors. It would also be useful in sorting out Heumann's assertion that frivolous trial cases are punished while "real" triable cases are not. DISPARITY There are two principal unanswered questions in studies of disparity: How much unexplained variation is due to systematic differences among decision makers rather than to planned or to apparent disparity? What are the nature, magnitude, and sources of the differences that are found? Providing answers to these questions will help to clarify the sources of disparity, thus focusing debate on whether the identified differences are warranted or not, which is a value question. Research on disparity, however, faces problems of measurement error arising from inadequately measured variables, omitted variables, and sample selection biases. In addition, there is the problem of classifying "like cases" and the identifying criteria to be used in grouping cases as similar or different. Cases that appear alike initially may, on closer scrutiny, differ in subtle ways (e.g., one defendant may be emotionally disturbed) or in not-so-subtle ways (e.g., two cases in which the con- viction offenses are the same as a result of plea negotiations may differ substantially in the actual underlying offense behavior). Conversely, two cases that differ with respect to the conviction offenses may involve essentially similar offense behavior. Consequently, research to improve the estimates of the determinants of sentences will also contribute to the identification of disparity. Often what appears to be disparity in sentences may actually result from inadequate models of the sentencing decision. The extent of this seeming disparity can only be reduced with improved models of sen- tencing, but existing models will not be improved simply by adding more variables. Instead, observation, interview, and experimental studies are needed to create models that better reflect the processes by which in- teractions among court personnel affect decision making and improve
268 RESEARCH ON SENTENCING THE SEARCH FOR REFORM the measurement of key variables such as offender culpability and of- r tense seriousness. To shed further light on interjurisdictional variation in sentencing and the influence of community attitudes as environmental constraints on judicial behavior, further studies might follow up on Gibson's (1978a) study of judges who "ride circuit." Controlling for defendant and case attributes and for judge attributes and role conceptions, the decisions of judges who serve in several diverse communities can be examined to assess their responsiveness to local norms and to explore the sources of public influence on judicial decisions. An additional source of disparity associated with environmental con- straints on judicial decisions, which may limit judges' willingness to sentence convicted offenders to confinement, is the physical conditions in local jail facilities and the availability of alternative sanctions (as well as judges' knowledge of each). Many judges may be reluctant to send minor offenders to crowded and dangerous jails for even brief periods; new or uncrowded facilities, conversely, may encourage greater use of incarceration. Tests of the impact of conditions of incarceration on sen- tencing outcomes might be carried out through an interrupted time- series analysis of the sentences of individual judges prior to and following the opening of new jail facilities. PUBLIC OPINION Better understanding of public perceptions of crime seriousness and the severity of penalties and how these affect judicial behavior is also de- sirable. Blumstein and Cohen (1980) found that, although there is strong consistency in the relative ranking of sentence severity across crime types, there are important differences among social and demographic groups about the absolute magnitude of sentences to be imposed. Sim- ilarly, the public's chosen sentences correspond to the actual sentences and time served in relative terms across crime types, but they differ considerably in absolute magnitudes. In opinion polls, respondents' recommended sentences generally re- flect their responses to only brief abstract offense descriptions. A re- spondent is told little about the circumstances of the offense and less about the offender. One might come closer to actual sentences with studies that measure opinions in response to more fully elaborated de- scriptions that reflect the typical offense-offender scenario. Overall, there is a need for further study of the public's judgment about appro- priate sentences, the information bases on which people make those
Research Agenda for the Study of Sentencing 269 judgments, and further exploration of the role of public opinion in shaping sentencing policy. RESEARCH APPROACHES A mixed research strategy is needed to increase understanding of the determinants of sentences. Progress in modeling sentencing decisions requires fuller knowledge of the decision-making process. Methodolog- ical advances are needed to address problems of sample selection and measurement error and to develop better measures of the key variables. These, in turn, require qualitative and experimental research. Qualitative studies can be especially fruitful in identifying variables that motivate decisions but do not appear in official data and in illu- minating how the interactions among criminal justice system actors affect decision making. Research on the flow of information and influence among criminal justice system personnel and the impact of this infor- mation on decision making, for example, can contribute to an under- standing of the ways in which these decision makers view case serious- ness. Studies detailing the interactions between judges and prosecutors, between prosecutors and defense attorneys, and between judges and probation officers should illuminate the kinds of issues these actors raise in discussions of cases, the questions they ask, and the manner in which they talk about offender culpability, offense seriousness, and prior re- cord. In considering prior record, for example, it may be that some individuals or work groups emphasize the length of an offender's arrest record while others focus on previous imprisonment. Efforts to explore the determinants of sentence outcomes often are hampered by reliance on the available data, which tend to be retro- spective and reflect only official sources. Instead, researchers might attempt prospective data collection, interviewing and observing the ac- tors whose behavior will subsequently be modeled and collecting de- tailed data on the variables that appear to be important but are often neglected. This preliminary exploration of the decision-making process could be especially fruitful in identifying variables that motivate deci- sions but do not appear in official records. In particular, data related to variables that affect the assessment of offender culpability are often not recorded in court archives or, when available, are often ignored by researchers because they are not easily quantified or coded. For ex- ample, a presentence report may describe how circumstances like the loss of a job, a death in the family, or an older brother's influence contributed to a defendant's behavior. These factors, along with vari-
270 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM ables like motive, level of planning, vulnerability of the victim, and foreknowledge of the likely level of harm, may strongly affect judicial and prosecutorial assessments of culpability, predictions about future criminal activity and dangerousness, and, ultimately, the sentence. One way to illuminate the subtle factors influencing sentencing de- cisions is to trace a set of sentences as they are formulated. In conducting such prospective research on sentence development, a researcher can independently measure variables like level of planning and can tap de- cision makers' assessments of these variables by interviewing them prior to and following the decision. For example, judges might be interviewed immediately after a sentencing hearing or after reviewing case records in chambers; probation officers might be interviewed after completing a presentence report that includes sentencing recommendations; and prosecutors and defense lawyers might be interviewed immediately after pretrial negotiation conferences. Such data could be used to generate hypotheses about the nature of the decision-making process and models of sentencing decisions. These models can also be used to explore whether variations among decision makers are associated with consideration of different case characteristics, different assessments of culpability and prognosis, or different weights given to the variables in the formulation of sentences. The results of such process explorations can then be directly tested in experimental studies and structural modeling efforts. One such ex- perimental technique for manipulating variables like level of planning, type of motive, and victim provocation that are difficult to measure is "the vignette procedure" (e.g., Rossi and Nock, 1982~. Briefly, one designs a set of "stories" or "cases" in which the parts can be randomly interchanged. For example, in one story or case the offender is black and the victim is white, in another the offender is white and the victim black. One might also use such a set of vignettes to explore assessments of the strength of evidence with a sample of attorneys and judges who would be asked to rate the stories on such dimensions as likelihood of indictment, of a substantial charge reduction through negotiation, and of a verdict of guilty if the case were tried. By regressing the ratings of such vignettes on the vignette characteristics, one might estimate the relative importance of factors determining the strength of evidence in a case. Experimental simulations of sentencing by judges or others in a labo- ratory or other experimental setting also provide useful data, particularly when used in conjunction with nonexperimental data to validate results obtained by the latter or to provide alternative estimates that are subject to different sorts of biases. A number of studies have used simulations
Research Agenda for the Study of Sentencing 271 and experiments to analyze the details of judges' decision processes, including the variables they used in making decisions and the order in which those variables were considered. Such studies of decision making should be performed with participants throughout the criminal justice system, including probation officers and prosecutors. The results might provide information on the relative importance of otherwise unobserved factors in the selection process, thereby helping researchers to identify additional variables that are important to measure. If conducted in a variety of jurisdictions, they could indicate the consistency or variability of the effect of these factors. Experimental simulations can also be used to address some questions that cannot be answered by observational data. For example, judges might be asked to choose for hypothetical cases both a determinate sentence and a minimum and maximum sentence. Their answers could be used to evaluate the implications of laws on determinate sentencing. Experimental work with judges and prosecutors, guided by the results of the kind of field studies outlined above, would permit a direct test of the impact of individual case characteristics on sentence outcomes. Such experimental research would also make it possible to evaluate the impact of case characteristics that occur infrequently but exert a major influence when they do appear. Modeling is the most frequently used approach to the study of the determinants of sentences. Among the modeling problems that need to be addressed by future research on sentencing are those of omitted and inadequately measured variables and sample selection biases. Many po- tentially relevant variables are omitted from models of sentencing out- comes, particularly those investigating discrimination in sentencing. These omissions can lead to biases in the estimated effects of the included variables when the included variables are correlated with the omitted or inadequately measured variables. Similar problems can occur when such variables are poorly measured. An obvious remedy, simple to prescribe but often difficult to accomplish when research is conducted with inadequate data sets, is the inclusion of a richer set of adequately measured independent variables. New research projects should make every effort to include information on a wide variety of individual case and offender attributes and to explore the relative usefulness of alter- native measures of included variables. One approach to addressing the problem of inadequate measures of various determinants of sentences is to use a model that explicitly links multiple observed indicator variables with the unobserved determinants of case disposition (Gerber et al., Volume II). Such models, often called structural models involving latent variables, estimate the effects of unob-
272 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM served variables from observations of the outcomes on other variables whose values are postulated to be determined as functions of the unob- served variables. In studying sentencing using such structural models, the primary determinants of case disposition (seriousness, quality of evidence, and prior record) are treated as unobserved or latent variables. The model consists of a series of structural equations representing prin- cipal indicators for which data are available (e.g., charge, pretrial re- lease, bail amount, type of legal representation, conviction at trial, and severity of punishment). Using data on available observed determinants, it is then possible to estimate the effects of unobserved variables on case outcomes without directly measuring these unobserved variables. For the study of racial discrimination, this approach may make it possible to disentangle the various sources of race-outcome correlations that are likely to reflect both discriminatory and nondiscriminatory factors at distinct stages of criminal justice processing without requiring improved measures of the unobserved variables. Structural equation models must be based on a comprehensive theory of the operation of the criminal justice system and the motivations of its principal actors and on a theory of measurement error. Such theories do not yet exist, so that any model will rest on a variety of questionable assumptions. Nevertheless, structural modeling presents an alternative approach to cross-sectional studies that rely on largely inadequate mea- sures of the primary determinants of sentencing to understand sentenc- ing outcomes; structural models of the case disposition process sharpen researchers' focus on areas of ignorance and suggest new hypotheses for further testing. Wide agreement on any particular formulation of an identified structural model is unlikely, but consistent findings obtained under a variety of different model formulations can increase confidence in the validity of the findings. The handling of sample selection biases depends fundamentally on the source of the correlation between the unmeasured and measured determinants of sentences in the selected sample. If the correlation arises from unmeasured factors common to selection and sentencing that are independent of included variables in the full population, there are a number of available adjustments that rely on explicit estimates of the selection process to generate unbiased estimates of the determinants of sentences (e.g., Berk and Ray, 1982~. The key is obtaining data for the full population before selection occurs. For example, the data set might include cases as they enter the criminal justice system, and it must include measures on the kinds of factors that determine how deeply a case is likely to penetrate into the system in addition to measures of the determinants of sentences.
Research Agenda for the Study of Sentencing 273 These estimation techniques are not appropriate when there is also a correlation between the unmeasured and measured factors in the full population. Use of these techniques requires having adequate measures of any correlated but unmeasured determinants of selection and sen- tences. An alternative to improved measurement is to develop models of both the selection and sentencing processes, including in the models, when appropriate, the unmeasured determinants of these processes. Having specified a system of structural equations for the various proc- esses that includes common latent (i.e., unmeasured) variables in several of the equations, the effects of the latent variables can be estimated from common movements observed in multiple outcome variables. Once again, estimating this combined system of selection and sentencing re- quires data for a sample of cases before selection occurs (see, e.g., Garber et al., Volume II; Klepper et al., Volume II). Further research addressing the problems of measurement error and sample selection biases should include basic methodological work that formally compares the alternative correction techniques and documents the relative effectiveness of each and the trade-offs among them. Re- search is also needed to assess the robustness of proposed alternative methods to correct for sample selection biases. In particular, future research should explore the sensitivity of the estimated effects of various determinants of sentences to measurement error and sample selection biases. This analysis would include use of a variety of alternative mea- sures of offense seriousness and prior record, both in replications with the same data sets and with independent data sets, to assess the sensi- tivity of the results to the particular measures used. Alternative models of the selection and sentencing processes should also be explored within any single data set to test the sensitivity of the results to the particular sets of assumptions in any model. To the extent that consistent results are found under a variety of alternative measures and model formula- tions, it will increase confidence in the available estimates for the de- terminants of sentences. On the other hand, substantial variations in the results would signal that biases arising from measurement error and sample selection are likely to be serious problems in any estimates of the determinants of sentences. Important progress in modeling sentencing outcomes is likely to result from more adequate treatments of the complexity of the dependent variable, sentence outcome. Instead of using a single scale to represent sentence severity, with all the arbitrariness such a scale represents, there should first be a qualitative dependent variable representing the choice among the various kinds of sentence options being considered. Then
274 RESEARCH ON SENTENCING THE SEARCH FOR REFORM the second part of the model would address the magnitude of each selected sentence type. An adequate model of the determinants of such a multivariate sen- tence outcome variable is likely to be a complicated function of many input variables, some related to the crime, some to the criminal, some to the decision maker, and some to the context in which the decision is made. These variables may interact in important ways that need to be explored in future research. Once such a model has been specified and validated, e.g., by comparing its predictions with new sentence out- comes, it is possible that simplifying approximations can be found that would isolate those essential variables relating to the crime, those re- lating to the criminal, etc. If an approximation adequately reflecting sentence outcomes can be found, one might then consider using this approximation as a basis for developing single-scale variables that com- bine the many individual measures of independent variables into a single variable, reflecting offense seriousness, for example. Given the generally limited state of knowledge in modeling the determinants of sentences at this time, however, resources should be devoted primarily to model development and data collection, with secondary emphasis on devel- oping scales of the determinants of sentences. Efforts to scale certain key variables like offense seriousness and prior record, however, may usefully contribute to the model development effort proposed by pro- viding useful insights into cognitive issues involved in individual decision making. STRUCTURING SENTENCING DECISIONS PREDICTION RESEARCH AND SELECTIVE INCAPACITATION The sentencing guidelines developed to date have emphasized current offense seriousness and prior criminal record. This is in sharp contrast to the original guidelines of the U.S. Parole Commission, which ex- plicitly considered factors related to predictions of future recidivism in making release decisions. It has been suggested that sentencing guide- lines might be similarly designed to emphasize selective incapacitation. This would require research to identify those factorts) that best predict subsequent rates of offending and weighting them in the guidelines to ensure long sentences for the relatively small number of high-rate violent offenders. Greenwood and Abrahamse (1982) suggest that it may be fruitful in terms of reductions in both crime and prison population to identify high-rate violent offenders (measured by the number of serious
Research Agenda for the Study of Sentencing 275 crimes per year of liberty) *om the larger number of low-rate offenders on the basis of background and other characteristics. Such an effort may be viewed from several perspectives. Supporters suggest that it can be viewed as a way of reallocating scarce prison cells so that they will confine that group of offenders likely to commit the most crimes and as a means of decreasing the sentence lengths of the large number of offenders with low offense rates without substantially increasing crime. Critics point to the injustice of basing any individual's punishment in any respect on behavior that has not happened and may not happen. The principle of selective incapacitation and research on it thus in- volves value and legal questions about which there is disagreement. For some, the value choice would be influenced by the question of how good a prediction of individual criminality can be made. If the prediction is good and if it would not result in imprisonment of individuals who would not otherwise be in prison but would result in a reallocation of time in prison, they would consider such an approach to be permissible. For others, any use of selective incapacitation raises insurmountable ethical and legal problems. Their view is that specific individuals ought not be punished on the basis of some prediction about their future criminality and that the criminal justice system must scrupulously avoid taking such actions. There is also fundamental concern about the variables that would be used in any such prediction. The use of an individual's social or economic characteristics for such a purpose presents the greatest hazard, and there are important legal questions that challenge the use of juvenile record information and records of official contacts that do not result in convictions. Furthermore, since many variables that predict recidivism may be correlated with race or minority status, their use could have disturbing discriminatory consequences. An additional concern arises from the problem of errors inherent in all predictions of future behavior. If an explicit selective incapacitation policy were implemented, the false negatives (i.e., those released who commit new crimes) would be easily identified, but the false positives (i.e., those imprisoned because they were predicted to commit crimes but who would not have done so) could not be identified they could not demonstrate that they would not have committed new crimes if released. Thus only one of the two types of error that might occur can be observed, the erroneous release, and there is some danger that the pressure to avoid such errors would lead to increasingly tight standards for release. While the increased use of such prediction methods for parole decision making in recent years does not appear to have resulted
276 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM in tightened release standards, the relatively hidden nature of the parole process, compared with the much more visible sentencing decisions, may make the experience with parole not applicable to sentencing. Even those who are prepared to consider such approaches remain skeptical that good prediction models invoking only legally valid vari- ables can be devised and validated and result in decisions that are ap- preciably better than those of good practitioners. Most prediction re- search has been based on retrospective data and is always subject to "shrinkage" in predictive validity when applied in new settings. Fur- thermore, even when the internal predictions are good, the predictive quality diminishes appreciably when the variables that may be used in the prediction are restricted to official records of convictions (see, for example, Chaiken and Chaiken, 19824. Interviews with judges reveal that most do take into account their own assessment of an individual's subsequent criminality and that they do so using whatever variables they have at hand, including many of questionable quality or predictive validity in presentence investigation reports. Thus careful and validated prediction research may identify particular patterns of variables that offer valuable new insights to judges and prosecutors and thereby enable them to improve their sentencing practice. The problems of predicting offender criminality involve all the issues previously discussed regarding efforts to model sentencing outcomes: measurement, scaling, model misspecification, and selection bias. Any selective incapacitation scheme should have valid and reliable answers to the following questions: What are the magnitudes of the anticipated prediction errors? How many and which groups of offenders are likely to suffer from such a policy, and which ones are likely to benefit? How much do different variables contribute to predictions? How many of what types of crimes would be averted under the scheme? What crimes might increase as a result? How would the crime and imprisonment consequences differ from current practice? Addressing these questions can facilitate a more informed assessment of the value choices. Ulti- mately, policy makers considering a sentencing policy based on individ- ual prediction must weigh the benefits of crimes averted against the costs and dangers of incorporating in policy the biases and errors that are inherent in any model. Even with an adequate model, it is extremely important that the agencies that support such research subject any find- ings to validity testing and also consider carefully the possibility of misuse and the potentially inappropriate consequences of the introduction of selective incapacitation policies.
Research Agenda for the Study of Sentencing APPELLATE REVIEW 277 Appellate review of sentences has had little influence in the United States, but it is institutionalized and widely believed to be quite influ- ential in England. Since there are no systematic data on its effects, further study of the operation and impact of the English system might be useful to guide American reformers, who have urged its widespread adoption. One might examine the impact of major appellate decisions in England by analyzing the impact of several widely cited cases on subsequent lower court decisions to determine the process and pattern by which lower courts adapt. It would also be desirable to examine how appellate sentence review is working in those few jurisdictions in the United States that have established presumptive sentencing standards with appellate review (e.g., Minnesota and Pennsylvania). Such studies should look at the numbers of appeals, the issues raised, and the impact of the decisions on departures from standards over time. Interviews with judges could indicate knowledge of and adherence to appellate deci- sions. IMPLEMENTATION OF NEW POLICIES The adoption of sentencing reform whether through criminal code revision, statutory determinate sentencing laws, or parole or sentencing guidelines is a complex political process. In their efforts to understand the relation of legal change and social reform, political scientists and legal scholars have frequently studied the politics of passage and im- plementation of substantive legal changes, but they have less often fo- cused on the politics attendant on procedural changes in the law and in the judicial process. Comparative case studies of sentencing reform ef- forts in various jurisdictions would be a first step toward the construction of a theory of change in sentencing institutions. Any such theory would have to incorporate the role of the latent and symbolic functions of policy innovation and should also reflect the sources and limits of change. The process of implementing a new sentencing policy is too often overlooked by reformers, and research should illuminate that process. A policy change is not self-executing, and preliminary data suggest that the effects of some mandatory minimum sentencing statutes, bans on plea bargaining, and sentencing guidelines vary considerably, depending on the manner in which they were implemented. Qualitative examina- tions of the implementation process should be a part of impact evalu- ations, since they may provide the key to understanding the nature and
278 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM scope of any effects that are observed. Such examinations might include study of the knowledge of a new policy and attitudes toward it both before and after training or distribution of materials among judges, lawyers, and other court personnel. Sentencing guidelines shift discretion to prosecutors but limit their ability to threaten an extreme penalty. Without such a threat, defendants may be less easily induced to plead guilty and that result might in turn require a larger guilty-plea discount to keep the trial rate at a manageable level. Thus guidelines could bring about a shift in the dynamics of the plea-negotiation process and thereby provide a natural experiment on which research should capitalize. Such studies could examine courtroom work group norms and plea-negotiation practices before and after im- plementation of guidelines. EFFECTS OF SENTENCING REFORMS Evaluations of the impact of sentencing reforms thus far have been preliminary; there is a need for better research designs in any future evaluations. The development of such designs could be fostered by a center to support feasibility studies for impact evaluations (discussed above); more generally, there should be a systematic effort to design a broad evaluation research strategy). Such a strategy should be geared to establishing evaluations that meet both the short-term need for timely management feedback on an innovation and the long-term need for more rigorous evaluations that provide more definitive findings. A mix of methodological approaches should be used in impact studies. Statistical analyses of case-processing data available from centralized data systems can tell only part of the story of efforts to change complex court and corrections processes. Systematic interview and participant observation need to accompany quantitative evaluations as an essential part of the effort to understand sentencing behavior in a social and institutional context. Future evaluation studies should have more extended observation periods; time-series analyses should have several observation points rather than simple two-point, preinnovation and postinnovation, research de- signs. Extended time-series analyses are needed to distinguish effects associated with reforms from continuation of trends. Multiple post- reform observations are desirable to ensure that a reform has actually been implemented and that its effects have reached a stable point. Such an approach also avoids the selection bias associated with cases that are resolved early, which may differ in important ways from those that take
Research Agenda for the Study of Sentencing 279 a longer time to move through the system. Multiple observations over a longer follow-up period ensure a more representative case mix. Furthermore, outcome measures must address all stages of case pro- cessing, not only those directly affected by a reform. The variety of opportunities for the shift of discretion and the associated potential for nullifying the effects of a reform require observations at both earlier. and later stages. Evaluations of mandatory minimum laws and plea- bargaining bans, for example, should include data on pretrial dismissals and charging patterns as well as on sanctions imposed on convicted offenders. Impact studies should examine not only changes in the se- verity of sentences for those convicted but also changes in the mix of cases that appear for sentencing to identify any associated changes in the pattern of case screening. It is particularly desirable to examine adaptations by actors in the criminal justice system, such as alterations in charging and plea-bargaining practices, to determine how they affect implementation of a reform and may undermine its desired effect. Sim- ilarly, exogenous changes in case attributes, such as changes in patterns of offending, perhaps resulting from demographic shifts in the popu- lation, may occur; these changes will affect sanction outcomes inde- pendently of any reform and must be accounted for. Statewide statutory sentencing guidelines, which represent an impor- tant reform, should be evaluated. Evaluation efforts should examine the extent to which the changes in offender populations projected under the guidelines are realized. They should also test the correspondence between anticipated and actual changes in intrastate disparity. Studies of individual judicial compliance with guidelines are needed. Compliance is not merely behaving in ways consistent with the guideline rules; rather, it means conscious prescription of a sentence in accordance with the guidelines. In order to determine compliance, one must have at a minimum data on sentences by individual judges before and after introduction of the guidelines so that changes in sentencing patterns can be identified. Cases outside the guidelines may not indicate noncom- pliance if there are aggravating or mitigating circumstances that justify departures; sentences within the guidelines may ignore such circum- stances and thus also represent noncompliance, although those cases would be harder to identify. Pursuing this issue further may require learning how judges actually make decisions and determining why they sentence as they do. Research on compliance should analyze a sample of cases, including those that depart from the guidelines and some of those that ostensibly comply with them. Those analyses must be aug- mented by interview and observation studies to provide information on
280 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM the role of plea bargaining in generating outcomes and on the justifi- cations considered in choosing a sentence within or outside the guide- lines. In selecting which policy innovations to evaluate, it is necessary~to consider whether a change is really intended and is likely to be effective in changing practices. When the goal of a change appears to be primarily symbolic, an evaluation is not warranted. While there is no simple way to distinguish a real from a symbolic change, certain situational factors may strongly point toward a gesture that is largely symbolic. For ex- ample, if a mandatory minimum sentencing law is adopted when prisons are already filled to capacity, and if no provision is made to deal with the expected increase in the number of prisoners if there is compliance with the law, there is likely to be little actual change. While studies documenting a null effect may be valuable, their value lies primarily in the understanding they can provide of the adaptation process itself. Absent such a focus, such studies should be given lower priority for funding than a change that is likely to have real effects and to become a candidate for replication elsewhere. SENTENCING POLICY AND PRISONS Sentencing policy should include consideration of the impact of changes on prison populations; hence, there should be support for improving techniques for estimating prison populations. Because such estimates are likely to be used by future sentencing commissions in many juris- dictions, an investment by the National Institute of Justice in devel- opment of existing models and technology transfer is likely to be cost- effective. Projection models such as the flow model developed by the Minnesota Sentencing Guidelines Commission could be generalized so that they could be made available to other jurisdictions. There is a widely held belief that prison crowding has harmful effects on inmate health and behavior. However, few studies adequately doc- ument the effects of various conditions of confinement on the prison population. Given the willingness of the Supreme Court (in Rhodes v. Chapman 452 U.S. 337 t19814) to consider such studies in setting stan- dards and the paucity of reliable data on which to base such standards, research is needed to sort out the complex and overlapping effects on inmate morale, health, and behavior of a variety of factors: physical and social density; institutional size; control and disciplinary style; in- mate composition (especially as it affects inmates' victimization risk); amount of time mandatorily spent in living quarters; and prison term.
Research Agenda for the Study of Sentencing 231 Research on the effects of crowding and other prison conditions should look across institutions, controlling for institutional populations and physical attributes. But because those individuals who are viewed by corrections authorities as the most troublesome are often assigned to the institutions with the worst physical conditions, separating the effects of the environment from selection effects arising from the assignment process represents a difficult research problem. Ideally, such research would use an experimental design that randomly assigns prisoners to various conditions. To the extent that it proves impossible to design experimental studies, research could involve quasi-experiments that con- trol for selection processes or natural experiments that take advantage, for example, of the reassignment of prisoners that accompanies the opening or closing of a prison facility or of a unit within an institution. Only rudimentary data are available on the effects of changes in the goals of sentencing and the shift to determinacy on prisonerstheir behavior and program participationor on managementprogram availability and disciplinary practices. Further studies of the impact of determinacy on prison management practices and on prisoners would be desirable. The problem of growing prison populations and their relationship to changes in prison capacity has been the subject of recent debate that is important because of its wide publicity and the difficult policy choices regarding prison expansion now being confronted by many states. Our review of existing evidence indicates that neither capacity nor population alone can account for much variation in the other. There is a need to reconsider the question of the growth of prison populations; to develop models of prison capacity change that include exogenous demographic, social, political, legal, and economic factors that appear to determine variations across states and time periods; and to test the models in different states to provide more complex explanations for the variation found in the population-capacity relationship. For example, one might have expected that the current pressure of crowded prisons would be reflected in a mixture of building to increase capacity and reducing the severity of sanctions imposed on convicted offenders. Generally, how- ever, only a limited amount of building has occurred, and the severity of sanctions imposed appears to have increased. Development of such models should be associated with efforts to explore adaptive responses by various jurisdictions as sentencing policy changes or as new prison capacity becomes available. It has been hypothesized that there will be an excess of prison capacity after 1990 in certain states as the baby boom generation passes through the ages of highest likelihood of imprison-
282 RESEARCH ON SENTENCING: THE SEARCH FOR REFORM meet. This possibility offers an additional opportunity for testing the degree to which sanctions increase in response to the increased avail- ability of prison capacity. Evaluations of programs intended as alternatives to incarceration have generally failed to provide reliable answers to two related questions: how frequently alternative programs are actually used as alternatives to incarceration and the programs' impact on prison population. To obtain those answers it is necessary to measure the displacement effects of the programs, i.e., the extent to which offenders sent to a particular alter- native program would, in its absence, have gone to prison and, con- versely, how often the programs are used instead as supplementary sanctions for offenders who would not have been incarcerated. Mea- surement of the displacement effects, however, requires an adequate model of the in/out decision. Jurisdictions with explicit sentencing pol- icies, such as Minnesota, provide the opportunity to distinguish among offenders on the basis of clearly articulated sentencing policies, thereby allowing a more adequate test of the effects of alternative programs. In summary, there are a wide variety of important research questions regarding sentencing principles, policies, and practices. Studies are needed to overcome the methodological difficulties and address the substantive issues related to the determinants of sentences; the practical and the- oretical implications of various approaches to structuring sentencing decisions; and the effects of changes in sentencing policies on criminal justice system practices, sentence outcomes, and prison populations. These issues pose a difficult but important challenge for the next gen- eration of research on sentencing.
