National Center for State Courts


Improving Justice through Leadership
and Service to the Courts



CTC Session Articles

Education Session Article

This article was written in support of a presentation given at CTC7 in 2001.

Sentencing Support Technology in Oregon

By Michael Marcus


Recidivism statistics dramatically reveal the dysfunction of sentencing discretion; legislative and popular ballot initiatives have responded by limiting that discretion. Decision support technology has rarely been aimed at sentencing decisions, and when applied to sentencing generally compounds the error by presupposing that judges (or sentencing models) should emulate past judicial performance.

Sentencing support tools properly constructed and applied allow us to aim specific correctional responses at the offenders most likely to be diverted from criminal behavior by those responses. These tools promise to increase the public safety performance of sentencing judges and correctional providers (public and private); to justify the restoration of judicial sentencing discretion; to improve public policy decisions deploying correctional resources; and to alter the culture of criminal justice by allowing rationality a full role in the process.

Recidivism and the Role of Judges

With the exception of driving under the influence of intoxicants and prostitution cases which are generated to some extent by embarrassed first offenders who will not return to the criminal justice system the overwhelming majority of cases represent returning offenders who are in the midst of developing criminal careers.

Our local jail statistics provide one dramatic illustration of the problem. Those statistics reflect that of the 2,395 people jailed in Portland during July 2000,[1] 1,246 had been jailed in Portland on some other occasion within the previous 12 months. Twenty-two of the 32 jailed for Burglary in July, 2000, had been jailed in Portland on some other occasion within the previous 12 months B as had 22 of the 23 jailed for Robbery, 20 of the 26 jailed for Theft in the First Degree, 304 of the 372 jailed on drug charges, and 32 of the 39 jailed for vehicle theft.

More traditional measures agree that recidivism is by far the most common manifestation of criminal behavior. For example, Bureau of Justice Statistics for jails in the United States reflect that "More than 7 of every 10 jail inmates had prior sentences to probation or incarceration," and that "Of the 108,580 persons released from prisons in 11 States in 1983, an estimated 62.5% were rearrested for a felony or serious misdemeanor within 3 years, 46.8% were reconvicted, and 41.4% returned to prison or jail."[2]

And, although some have credited increases in prison terms for reducing serious crime rates in the United States (while others forcefully advance economic explanations), we’ve reached the point in this cycle where offenders are returning from prison at the same rate as we are sending other offenders to prison; serious crime rates have rebounded, and the most crime-ridden of communities are finding that whatever relief we provided by removing serious criminals from their midst has been overcome by returning offenders who have somehow not received "the message" we tried to send them by putting them in prison.

As far as I can tell, this experience is similar to that of other English-speaking jurisdictions.

Regardless of evidence that the most significant opportunities to reduce or prevent crime lie outside the criminal justice system, recidivism demonstrates that the courts have repeated opportunities to focus correctional resources on the offenders who are disproportionately responsible for criminal behavior; recidivism statistics demonstrate that our performance in reducing recidivism is shamefully inadequate.

That criminal justice persists in its dismal performance is not unrelated to the apparent disinterest with which it treats literature about what works to reduce crime. Overwhelmingly, the judicial participants in the criminal justice process avoid all accountability for recidivism: the causes of crime are beyond our control as judges, and we are to some extent content to yield the field to those political players who think they know better or cynically exploit the understandable fear and frustration of the public. We may wring our hands about encroachment on our role and our loss of discretion, but we seem content to leave it at that as long as no one measures our worth by our impact on recidivism. Treatment providers follow our lead and adopt performance measures other than reduced criminal behavior; their job is to succeed in having their clients "complete the program" and--even more importantly--to play well with paper work. No judges ask public correctional agencies or private programs feeding at the criminal justice trough to prove their worth in crime reduction, or to compete on that basis for cases.

o complete is our failure that we do not even attempt to improve our performance in reducing criminal behavior. We perpetuate instead an ancient prattle about mitigation and aggravation, about just deserts and "sending messages," accepting the legitimacy of what is in essence a state religion complete with myth and ritual. Our value is measured by our capacity to articulate sufficient official outrage while nodding occasionally toward compassion and making the "punishment fit the crime." The most advanced forms of this litany are like the federal sentencing guidelines, which have assigned points to every aspect of the prattle to bolster the illusion of a rational process. But something not worth doing is not worth doing well.

So it is that our occasional nod toward rehabilitation is formal at best. We sentence thieves to "theft talk," drunk drivers to alcohol evaluation and treatment, and drug users to substance abuse programs. That this is out of some sense of symmetry rather than any responsible attempt to reduce criminal behavior is evident from the fact that we make no attempt whatever to learn which of these destinations do anything to solve the problem their names imply, let alone whether the offenders show reduced criminal behavior as a result of our dispositions.

While we persist in this medieval dance, our constituents suffer. To the extent that some of those 22 jailed for robbery who were jailed before could have been diverted from crime by smarter sentencing the last time they were through our courtrooms and jails, we must share responsibility for their subsequent criminal behavior. By ignoring outcomes that matter, we have failed to prevent avoidable victimization. People are suffering theft, assault, and even homicide because we choose not to investigate responsibly which choices before us carry the best hope of diverting which offenders from criminal careers.

There was, perhaps, a time when doctors and judges were much more alike. We both wore robes, and we both maintained institutions apparently organized to maintain our status and our mystery. But medicine's shortcomings were harder to disavow. When patients die, doctors’ failure is apparent. Several hundred years ago, medicine recovered some of its ancient wisdom and began to rebound from the dark ages by pursuing science. But while science flourished, courts persisted in their old ways--producing testimony with torture while medicine learned to cure and eventually to prevent increasing numbers of illnesses by studying what works and what does not and when.

Today, medicine--for all its faults--has united academia and practice; doctors are trained to cure in teaching hospitals, tested on their expertise before they are allowed to have patents’ welfare in their hands, and expected to keep up with and apply literature. I like to think that the prospect of civil liability in court reinforces accountability and, in turn, performance measured by outcome.

Meanwhile, although we judges have matured beyond taking of evidence by torture, our sentencing acumen has advanced not one iota; instead of learning from our experience and applying what we’ve learned, we have instead merely constructed ritual sentencing matrices that have no connection whatever with any impact on criminal behavior. We’ve continued to diverge from the social sciences, so that in corrections academia and the courts almost never notice each other’s existence. Our abysmal public safety impact should therefore come as no surprise.

Technology in the courts

We have persisted in our adherence to ancient ritual in sentencing notwithstanding our transformation by computer technology. Courts have for years used word processing and maintained databases of case filings and activity. We now publish our dockets on the Internet, and exploit web technology internally in a host of ways--finding attorney and judge phone numbers; accessing appellate decisions, rules, statutes, local ordinances and constitutional material; conducting on-line legal research; and surfing the growing number of useful public and private web sites. We use e-mail and maintain discussion groups. We have an automated jury selection tool, and each of our jurors wears a barcode on a neck cord. We've just begun to roll out an automated criminal judgment which will read from and write to our case database, print hard copies for those in the courtroom, and share data efficiently with our criminal justice partners--such as the Department of Corrections and local jail authorities. Electronic filing in civil cases is just around the corner.

With this level of information technology, we have growing access to information about offenders, and increasing information about some of the potential providers of correctional services.

Courts have made precious little progress, however, in exploiting information technology to support sentencing decisions. Midtown Manhattan and Red Hook projects have gained international notice. They've facilitated public access to court processes, and provide a "sentencing workbench" to judges. Although the allegory is enticing, there has been no progress in the craft of sentencing itself.[3] The fatal flaw is the assumption that judges are trained and apparently inherently equipped to make good use of information in sentencing, so that merely providing more information improves the result.[4] Undoubtedly, knowing more about providers, and which programs are and are not available in which community or custodial settings must afford some improvement. But this approach fails to take the one step that holds the potential for bringing criminal justice into the modern age.

The most sophisticated sentencing support technology outside the United States at least adds analytical tools to the mere information provided by most judicial information systems. New South Wales and South Australia have systems that gather the law and analyze information about offenders in light of criminal history and other data, and even information about available sentencing tools--but to the wrong end: emulating past judicial behavior. The Scottish approaches are similar, but pursue the same misguided goal by a "sentencing information system" like the one in New South Wales, or by creating a model of judicial behavior that will produce predictable outcomes from standard input.

Proponents of these systems, like proponents of sentencing guideline schemes, tout their propensity toward uniformity in the name of "fairness." All such schemes are far too imprecise to acknowledge significant differences in circumstances; their claim to treating like alike is illusory. Just for one example, at least the guidelines I’ve seen fail to distinguish offenders in the same grid block on the issue of which are psychopathic--an issue that anyone should recognize as critical in sentencing and correctional decisions. And many differences that have substantial impact on the length of a presumptive sentence are far more arbitrary than rational. The example that comes immediately to mind is an arsonist I sentenced who qualified for a substantially lower presumptive sentence because the home he burned was occupied by a firefighter who managed to stop the spread of the fire before it caused the otherwise inevitable degree of damage--the crime seriousness turning in part on the dollar value of the damage. And by what sense is it "fair" to determine the presumptive sentence by reference to the value of a stolen car? Should the theft of a wealthy victim’s extra limo be punished more harshly than that of a working family’s only transportation? Even by just deserts notions, the claim to fairness is largely chimerical.

Much more importantly, uniformity in pursuit of persistent failure is fair to no one, least of all to citizens dependent to some extent upon our ability to protect them against repeated victimization by those we sentence "and to a significant extent to the defendants, whose misery we compound without diverting them from criminal behavior.

It is worth adding that the tools for which I advocate pursue the goal of doing as often as possible whatever is most likely to work under the circumstances. Once it is the circumstances that correlate with what is likely to work, and what options are available, that define which offenders are "alike," sentencing support tools supporting reduced recidivism are arguably even more productive of meaningful uniformity than any of the systems in place today in the name of uniform treatment. The difference is that present systems construct a bizarre calculus of just deserts and defend their performance with tautological[5] uniformity, while sentencing support tools properly aimed look for the differences that matter for public safety and encourage sentences sensitive to those differences.

The Promise of Sentencing Support Tools Designed to Focus on Public Safety

Decision support technology holds tremendous promise for many human enterprises. Public and private organizations are exploiting these tools in growing recognition that efficient analysis of accessible data can markedly accelerate performance. Businesses that need to profit to survive are finding this technology essential--none can survive by sending catalogs in the mail at random; insurance companies study their experience to set rates and adjust risks. Public health systems and HMOs increasingly use automated analysis to tailor responses and strategies to exploit their experience as captured in enormous databases. From the perspective of an HMO, a patient with a given profile and history should be treated in the way which most often works best on such patients with such profiles and history--so that the risk of occasions for subsequent and expensive interventions is reduced. From a treating doctor’s point of view, it is critical to know what response correlates best with cures for which patients.

One major hurdle between the criminal justice system and a similar path to efficiency is our profound confusion as to what it is we are supposed to accomplish. Even if, as I suspect, the relationship between our failures and ever-increasing demand for our services provides no operative motive for our adherence to just deserts, and even if I am wrong in attributing our persistence in an archaic tradition in part to a fear of accountability, we’ve been performing the sentencing-punishment liturgy so long that it is hard to see out of our cave. Once the light shines in, however, it is clear that at least in some part, we can and we must measure our performance by our impact on public safety. Once we realize that public safety is far more measurable than aggravation, mitigation, and just deserts, we can exploit decision support technology with the same beneficial impact on our work as that obtained by many other enterprises that have adopted these tools. And, because we are so far behind, our wise use of this technology will transform the criminal justice enterprise far more profoundly than it has enterprises whose outcome measures have long been understood.

Oregon has made bold steps in this direction. In 1996, a citizens’ ballot measure rewrote the relevant state constitutional provision to insert public safety into a list of criminal justice purposes: "protection of society, personal responsibility, accountability for one's actions and reformation."[6] In 1997, the Oregon Legislature passed a bill that made the requisite changes in a number of laws:

  • To make reduced criminal behavior the dominant measure of performance of criminal justice agencies (and their private partners) for juvenile and adult offenders;
  • To require that agencies share data (including juvenile and adult data) to facilitate assessment of the impact of correctional efforts on future criminal behavior; and
  • To require that criminal justice agencies use the data to see correlations between correctional responses and reduced criminal behavior.[7]

Also in 1997, the Oregon Judicial Conference resolved:

[T]hat in the course of considering the public safety component of criminal sentencing, juvenile delinquency dispositions, and adult and juvenile probation decisions, judges should consider and invite advocates to address the likely impact of the choices available to the judge in reducing future criminal conduct.[8]

In Multnomah County, Oregon, we have constructed a data warehouse and have developed the tools with which to give all involved in sentencing decisions rapid access to information about what works on which offenders.

Data warehouse technology pools data from multiple existing operational sources within the criminal justice system for the purpose of preparing and comparing analyses that help us all make better decisions. Our data warehouse now gathers data from courts, local law enforcement, prosecutors, and jails, and will soon include prisons, state-wide law enforcement and, eventually, private providers of programs designed for offenders. It transforms that data so it all speaks the same language, without requiring that we first unify and standardize the platforms and hardware on which the various criminal justice agencies support their operations. Then it assembles the data in a structure whose "architecture" is designed for efficient responses to the queries anticipated from the target users in our case, judges, attorneys, probation officers, pretrial release workers, probation and parole officers, and corrections counselors.

Finally, our project provides users with the tools that facilitate the analyses that users need. Our tools are in their beginning stages; we expect to be out of beta into phase I of useable tools by the time you read this.

Existing web-based tools afford users access to sentencing guidelines regulations, appellate decisions, and relevant statutes, as well as to information about local providers. Sentencing support tools will provide the analytical function.

During (or in preparation for) a sentencing hearing, a user would typically start using the sentencing support program by entering an OJIN (Oregon Judicial Information Network) case number,[9] but may also select any of a great number of identifiers, or use a name search.

When a user "submits" that information, the system presents a screen by which to select the charge for which the court will sentence the offender. It proposes count one, traditionally the most severe charge, but presents as well the other charges in the case, and allows a search for and selection of any criminal charge known to Oregon state or local law, including, for example, city littering ordinances.

When the user selects or accepts the default charge and submits this choice, the program exploits data stored in the warehouse extracted from operational systems for people like the offender, with a similar criminal history, age, gender, and ethnicity, and presents outcomes for each sentencing element used on such people when sentenced for such crimes.

The program displays the results for similar offenders sentenced in the past for similar offenses in a bar chart. From left to right, in order of frequency, the bars represent sentencing elements: Restitution, community service work, probation 1-18 months, jail 11-30 days, jail 4-10 days, forest work camp, and so on. Below the bar chart is a table that lists all of the raw numbers. The numbers below 30 are displayed in red, and are not represented by bars B a rough equivalent to statistical significance.

Three tabs B Charge, Recidivism, and Profile B afford the user access to the assumptions constructed by the program based on the offender and sentencing charge identified by the user.

For example, for a defendant being sentenced for Theft in the Second Degree, the program defaults to comparing only those people sentenced for property crimes. A user can change the control cohort by modifying that default to another category of crimes or to all crimes. The user can also change the definition of what group of offenders is "similar" to the defendant by changing the choices made by the software based on available information for the offender before the court. Part of that profile is based upon criminal history, which we have divided criminal into six categories of five levels of severity in each. We’ve established data rules for each of six categories, ranking people from "none" to "severe" in crimes of violence, sex crimes, drug crimes, property crimes, DUII/major traffic, and domestic violence crimes.[10]

The profile also includes the offender’s age, gender and ethnicity. The ability to filter cohorts based on ethnicity allows us to avoid compounding the disparate impact of criminal justice on minorities. It is neither fair nor accurate to assume that a minority defendant’s criminal history is necessarily comparable, in terms relevant to best sentencing practices, to the criminal history of an offender whose rate of apprehension, prosecution, and conviction is not exaggerated by racial profiling. In addition, many corrections providers target their programs to serve minorities; it is important to all of us that we are able to measure their impact on criminal behavior with accuracy and fairness--both of which are unattainable without the ability to adjust the cohort appropriately for analysis.

Should users wish to eliminate ethnicity from the analysis (or see the impact of doing so), these tools easily allow that option.

Our tools also include the ability to adjust outcome measures, although all are based on some flavor of recidivism B because the whole purpose of this effort is to do a better job of reducing criminal behavior. When sentencing for a property crime, the program accordingly defaults to measuring recidivism by a conviction for a property crime within three years of the sentence that it is comparing. But, for responsible pursuit of public safety, we need to be able to look at various recidivism measures--for example, when sentencing a sex offender on a property crime. Flexible recidivism measures also allow us to identify opportunities we would not notice were we focused only on the crime of conviction. We may notice that of the choices available for a particular property crime offender, none vary significantly in their correlation with reduced property crime recidivism, but one or two show marked improvement in drug or driving under the influence crimes.

Similar considerations support our need to vary the length of time during which we consider recidivism, and this program allows us to choose six months, three years, or five years as the period during which a new crime counts as a performance measure. We can also look for arrests rather than convictions during any of these periods. All of these choices have implications, and all can be debated. For example, the longer period we use, the more likely it is that circumstances other than our sentencing choice have intervened to produce the result; the shorter period we use, the less likely it is that we are capturing the long-run public safety implications of our choices.

And, particularly with crimes such as those representing domestic violence, it may be critically important to recognize arrests because the inability to prosecute many of these crimes due to victim unavailability may well mask our successes and our failures. In any event, our tools allow the parties and the court to consider the arguments, make the changes, and recalculate during (and in preparation for a hearing) so that the correlations can be observed and considered.

Even though we have not built our sentencing patterns on information about what works, sentencing support tools will offer tremendous improvement in our public safety performance from the outset. Because we will immediately see which sentencing choices have correlated (or not) with reduced behavior for relatively narrow categories of offenders who are like the one before the court, we will have a much higher likelihood than ever before of selecting a sentence that will reduce the offender’s criminal behavior. As our experience and our emphasis on what works grows, we can expect providers to compete on the basis of public safety success, and to improve their selection of "clients" they can benefit. Innovative programs will be able to exploit knowledge about what seems to work on which offenders to improve their success; new programs will capitalize on this knowledge to identify and pursue "market" niches in the offender treatment community.

The promises of this approach are many.

  • We should do a significantly better job of diverting offenders from criminal careers than we now do without access to the information these tools give us;
  • We should make significantly more efficient use of correctional resources, in and out of custody; we should stop funding programs that work on no one, and stop a program people it can't improve so it can have its successes with those it can; and we should do a better job of identifying offenders who need to be incapacitated;
  • We (and our criminal justice partners) should make more informed and more effective release and supervision decisions, and have a much better idea which probations to revoke and which to continue on probation than we now do;
  • We should earn back a good deal of public confidence for the courts and the criminal justice system;:
  • We should overcome the traditional gulf between academia and the courts, so that academia will benefit us (by refining our sentencing support tools, serving as a used and useful source of best practices, and responding to an entirely new demand for support for arguments about what works) and we will benefit academia (by providing real-world applications so critical to the reduction of projects that have no meaning or use outside academia, and by linking academia to public safety funding channels);
  • We should gradually but profoundly change the nature of the discussion in courtrooms and in public and policy-making debate by giving "what works" a full voice, including in public policy decisions about how to deploy and not to deploy criminal justice weapons to address a social problem. Perhaps we may even be wise enough to re-examine the drug war without fear of thinly veiled cries of blasphemy.

The future of sentencing support

We have a long way to go. With respect to the future of criminal justice, sentencing support tools are roughly where aviation was at the Wright Brothers’ first flights at Kitty Hawk. In the immediate future, we need to expand the automation and the quality of operational data collection at many levels of juvenile and adult criminal justice, law enforcement, and corrections. It turns out that actually using the data adds tremendously to quality control. We need to enhance sentencing support tools to allow us to analyze clusters of sentencing options commonly employed instead of merely individual sentencing components. We need to give users choices in including or excluding certain recorded behaviors other than arrests or convictions, such as police contacts generating "family beef" codes. We need to expand the range of included operational database sources for the data warehouse B such as motor vehicle records, alcohol and drug treatment records,[11] and juvenile records.[12] We need to add a host of attributes to offender profiles, such as mental health status, vocational background, substance abuse involvement, educational level, and perhaps even personality test scores from prison assessments.[13] We need to enhance the tools to allow more informed pre-trial release, probation supervision, and probation revocation decisions.

We also need to exploit the power of automated analysis. Eventually, the software should be "smart" enough to scour the variables so as to alert us to opportunities of which we should be aware. For example, as things now stand, it's up to users to decide whether to look at sex offense recidivism when dealing with a property crime offender; a smarter system would look without us and alert us if any notable peaks or valleys show up with respect to any of the dispositions we are considering with a property crime perspective.

Even all of this is just scratching the surface. As other related technology matures, we should have integrated computer-based resources to include such factors as the geographical location of possible treatment (or incarceration) providers so that the sentencing (or supervision) choice is aware of relevant information B for example, it may be that the offender's chances of completing a beneficial program are improved if we select a provider near his residence, even though one across town has a higher percentage of successful graduates; we may want to consider which prison facility has both the services appropriate for an inmate and an absence of fellow gang associates.

We have a long, long way to go, but we have taken the first steps out of centuries of darkness. Public safety will surely benefit, and we have every reason to expect that we can leave a legacy of brutally dysfunctional ignorance in the past where it surely belongs.

[1]. Portland Police Bureau Data Processing, August 25, 2000. The Portland Bureau of Police stopped producing these statistics in mid-2000; I am still waiting for their successor. An extract of the statistics is available at

[2]. Bureau of Justice Statistics Criminal Offenders Statistics,

[3]. The Midtown Manhattan project website for years carried on its "to do" list precisely the analytical application discussed in this paper. The application never materialized, and the web site is down.

[4]. To this day, the best training I’ve received in the area was in dog court, where the local animal control director had prepared a bench book with the relevant statutes and regulations, and staffed the courtroom with deputies happy to volunteer their experience in which dispositions seemed to solve problems best in which circumstances. Official training for judges and law students in sentencing is rare. Even when available, it generally serves merely to indoctrinate in the litany of the field, rather than to advance its effectiveness in terms of public safety. I’ve participated in sentencing seminars at the University of Minnesota School of Law and at Yale Law School, and can confirm that litany rather than science is the norm.

[5]. The tautology is this: if we define categories based on, say, criminal history and property value involved, then dictate a sentence dependent on categories, and if the scheme is perfectly implemented, we can claim uniformity. But if the categories aren’t "fair" (as when they fail to distinguish the theft for mere profit from theft to avoid a child's starvation) the resulting uniformity hardly serves fairness; if the resulting categories are ignorant of differences that matter for public safety, uniformity doesn’t serve safety. Treating like alike may well have some intrinsic value, particularly where issues of disparity based on ethnicity abound, but it is absurd to contend that we cannot achieve racial justice without abandoning public safety as a goal, or to defend a criminal justice system's dismal public safety performance on the ground that it achieves uniformity B particularly when its equal treatment performance is as dysfunctional as its public safety performance.

[6]. Article I, Section 15, of the Constitution of the State of Oregon.

[7]. 1997 Oregon Laws, Chapter 433 [1997 House Bill 2229].

[8]. 1997 Oregon Judicial Conference Resolution #1.

[9]. Our case numbers are unique to an individual; co-defendant cases have multiple case numbers, each corresponding with one defendant.

[10]. We’ve recently added domestic violence arrests to the data rules by which to assign offenders to a level (none, minor, moderate, major, or severe) for domestic violence crimes. Eventually, uses will have the option to toggle this feature on and off, but experience suggests that there are first offenders with no prior domestic violence problems and first offenders for whom this is the first conviction in spite of dozens of calls to the residence in the past on "family beef" complaints B and that this may be a distinction of importance for sentencing purposes if future behavior is an outcome measure.

[11]. Yes, we’ve considered the confidentiality issues so heavily laden with federal law and regulation. Our discussions with the Oregon Attorney General and the Director of the Oregon Department of Alcohol and Drug Abuse Programs have led to a consensus that aggregate data can be exploited for analytical purposes without legislative or regulatory change.

[12]. 1997 HB 2229, 1997 Or Laws ch 433, mandates sharing juvenile and adult data to help us see what happens to juveniles after we deal with them in delinquency (and dependency) hearings when they become eligible by age to show up in adult criminal justice data.

[13]. As a member of the Oregon Department of Corrections= "Inmate Incarceration and Transition Plan Design Team," I’ve offered input to such assessment design precisely to improve our ability to profile offenders so as to improve our dispositions in court, in prisons, and on probation.

Michael Marcus

Biographical Information

This biographical information may date from as far back as 2001. Please keep in mind that it may no longer be accurate.

Michael Marcus

No bio available.