Criminal Careers and "Career Criminals,": Volume I (1986)

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.

Crime Control Strategies Using Criminal Career Knowledge The criminal career paradigm suggests three general orientations for crime con- tro! strategies: prevention, career modifi- cation, and incapacitation. This chapter first presents the evidence and conclu- sions on programs intended to prevent young children from becoming offenders. It then reviews efforts to modify ongoing criminal careers through interventions with the family and efforts to change sub- stance abuse behavior and employment status, both of which are associates! with a high frequency of serious offending. This chapter also updates the work of the Na- tional Research Council Pane! on Deter- rent and Incapacitative Effects (Blum- stein, Cohen, and Nagin, 1978) by examining the crime control effects of different sentencing policies from an in- carceration perspective. Preventive strategies are intencled to reduce the number of nonoffenders who become offenders. Two types of knowI- edge are needled to design effective pre- ventive strategies: prospective indicators of subpopulations having high participa- tion rates ant! iclentified interventions that recluce participation in targeted sub 109 populations. Although the panel's review of experimental findings on the effective- ness of various interventions in reducing participation fount! little evidence sup- porting specific interventions as preven- tive mechanisms, it die] fire! a few cancli- ciates sufficiently promising to warrant a call for experimental replications. Career mollification strategies are fo- cusecl on persons aIrea(ly known to be criminals and seek to reduce the fre- quency or seriousness of their crimes. Also, modification strategies usually at- tempt to encourage the termination of ongoing criminal careers by creating fear of punishment, by modifying offenders' values, or by expanding opportunities for legitimate alternatives to crime. Career modification strategies, such as counsel- ing or behavioral therapies, intervene cli- rectly in the life of the offender to change criminal behavior. The pane] was partic- ularly interested in the substantial holly of research relating recidivism rates to various interventions, inclucling commu- nity-basecl supervision as an alternative to incarceration, verbal and behavioral therapies intenclec! to inhibit aggressive

110 and antisocial behavior, substance abuse treatment programs, and programs to im- prove employment prospects. Some of these interventions might also effectively inhibit future participation among nonof- fenders. Methodological problems limit the knowledge that most of these studies provide about the effectiveness of those interventions in modifying the criminal career, but several studies in which large samples were randomly assigned to alter- native treatments provide evidence about the relative effectiveness of some inter- vention programs. Incapacitation strategies focus on the crime reduced as a result of removing offenders from society during their crim- inal careers. While the use of imprison- ment as a means of incapacitation may also have career modification effects by altering offender behavior through in- rehabilitation programs, the "incapacitative effect" refers only to crimes prevented in the community by interrupting criminal careers during peri- ods of incarceration. This incapacitative effect depends fundamentally on the magnitude of individual crime rates, A. Two types of incapacitation are often dis- tinguished: collective incapacitation and selective incapacitation. Collective inca- pacitation refers to the effects of aggre- gate policies involving changes in the average sentences applied to average of- fenders. Collective strategies, such as a percentage increase in all prison sen- tences, were the subject of the previous Pane! on Research on Deterrent and Incapacitative Effects (Blumstein, Co- hen, and Nagin, 19781. Selective incapac- itation refers to policies that target in- creased incapacitation more narrowly on selected offenders. One such policy pre- scribes longer prison terms for the few offenders who commit serious offenses at especially high rates. As part of its work, this panel commissioned a reanalysis of recent Rand Corporation data on varia mate rehabilitation CRIMINAL CAREERS AND CAREER CRIMINALS lions in reporter] A among inmates and on the selective incapacitative effects of us- ing longer sentences for the high-rate offenders among those inmates. STRATEGIES TO PREVENT PARTICIPATION Research has identified demographic and nondemographic indicators of a high risk of eventual participation in crime. On the basis of the widely measured demo- graphic variables of age, sex, race, and place of residence, young black mates in  large urban locations appear to be at par- ticularly high risk of arrest: approximately 30 percent have been arrested by age 18 for an FBI inclex offense (see Blumstein and Gradcly, 1982:Table 11. And while the peak age for the first arrest is 15, there is significant risk at even earlier ages: about one-third of the black males ar- rested by age 18 had been arrested by age 13. As noted in Chapter 2, family charac- teristics associated with the onset of de- linquency at any age include earlier pa- rental delinquency and criminality, poor child supervision, and marital conflict in two-parent households. Early, intense, and diverse antisocial behavior by a child, poor school performance, and low measured IQ are other indications of a high risk for participation in crime. Full-time employment at a young age when school attendance is expected may also be associated with participation in crime. Several problems limit the potential of these factors in focusing preventive inter- ventions. Most important, considerable ambiguity surrounds the interpretation of many indicators, especially the demo- graphic variables, which often function as stand-ins for other unmeasured social and economic factors. In addition, the causal relationship between measured variables and criminal participation is often impre- cise and may vary with circumstances.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE Many children present some or all of the high-risk characteristics ant! antecedent behaviors but clo not become offenders, and any classification rule invoking the indicators wit! produce many false posi- tives (Loeber and Dishion, 19831.i Al- though the ability to predict future partic- ipation in crime increases cluring the late teenage and adult years, effective options for preventive intervention may well de- crease as the high-risk youth becomes older and less amenable to change. How- ever, some of the relationships observer! at ages blithe earliest at which any of the chilc~hoocI characteristics have been studied in relation to adult offending might also be observable at younger ages. Thus, an important research issue is de- termining the optimal age for interven- tion: if the intervention is early, the iden- tification of high-risk children is likely to produce many errors; if the intervention is late, there will be less chance of suc cess. Experimental Evaluations of Preventive Interventions Early identification of children or teen- agers at relatively high risk of becoming offenders raises the possibility of inter- vention strategies to prevent members of high-risk groups from becoming offencI- ers. During the 1960s and 1970s, federal funding was provicled for many juvenile iThere have been several attempts to develop formal classification rules that could be used to identify predelinquent children. One of the oldest, the Glueck Social Prediction Table, included five family-related factors that differentiated 500 delin- quents from a matched set of 500 nondelinquents (Glueck and Glueck, 1950~. A number of method- ological problems in this work have been identified (see Zeisel, 1968; Hirschi and Selvin, 1973), and in a validation of the scale- which targeted first-grade males in high-delinquency areas in New York City and followed them for 7 years-75 percent of the predictions of delinquency were inaccurate (see Lundman, 1984~. programs, some of which had crime pre- vention as one of their objectives. There is a large body of research evaluating these programs, which has been exten- sively reviewed (e.g., Lunclman, McFar- lane, and Scarpitti, 1976; Wright and Dixon, 1977; Farrington, Ohlin, and Wil- son, 19861. In general, this research has been characterized by small sample sizes and designs that were either nonex- perimental or cTic3 not have adequate con- trols. Also, the follow-up periods have usually been 2 years or less: although these periods are probably adequate to assess the short-term effectiveness of the  interventions in correcting antisocial be- haviors, they are not Tong enough to show effectiveness in preventing later criminal activity. However, the panel clid identify some promising experimental studies of interventions oriented specifi- cally toward! the precursor problems of bad family management, Tow school per- formance, and early antisocial behavior. In these few studies, the findings were sufficiently strong to suggest that further replications might establish their effec- tiveness in inhibiting participation in offencling. Interventions with the Family There are a few evaluations of efforts to train parents in identifying problem be- haviors as they emerge and using such modes of discipline as rewards and tirne- out periods to change them. Loeber (1984) reviewed three studies of experi- mental interventions with families of problem but not criminal-children aged 3-14, based on this orientation (Walters and Gilmore, 1973; Karoly and Rosenthal, 1977; Bernal, Klinnert, and Schultz, 19801. Two of the studies clem- onstratecl temporary effectiveness in cor- recting the overt antisocial behavior: re- ductions of about 60 percent in the frequency of the targeted behaviors, ob

2 served up to 1 month after completion of the program. However, the only study with a follow-up period long enough to discover long-term behavioral changes- 2 years-failed to detect a decrease (Bernal, Klinnert, and Schultz, 1980~. Longer follow-ups and searches of juve- nile court records would be needed to learn whether the interventions suc- ceeded' in inhibiting participation in criminal behavior. One particular intervention with fami- lies family social learning is based on the theory that early antisocial behavior is first learned through imitation of other family members and interaction with them, that ineffective parenting fails to inhibit minor antisocial behaviors from escalating in seriousness, and that these behaviors may lead children into trouble with school authorities or with the juve- nile justice system (see Patterson, Cham- berlain, and Reid, 1982~. The goal of this type of intervention is to eliminate paren- tal behaviors that serve as a model for aggressive antisocial behavior by the child and to suppress the child's antiso- cial behavior when it occurs. Theoreti- cally, parental training could begin at any time before the child reaches high school age. While these preventive approaches have been advocated (Pulkkinen, 1983; Loeber, 1984) and tried, the follow-up periods in the few experimental studies of such interventions have been too short to permit evaluations of their effective- ness in inhibiting participation in o£end- ~ng. Preschool Programs The Perry Preschool Project, carried out in Michigan, was primarily directed at improving the school performance of disadvantaged black children through a Head Start program lectual stimulation CRIMINAL CAREERS AND CAREER CRIMINALS (Schweinhart and Weikart, 1980~. This Head Start program was unique because it also examined the impact of a preschool program in reducing later criminal activ- ity. Results from a long-term evaluation suggest that the program had generally' beneficial effects. The experimental group of 58 children receiving the enrich- ment showed gains in measured IQ at age 5, although this gain, in comparison with a control group, dissipated entirely by age 14. But at the latter age, the experimental subjects displayed higher school achieve- ment, better self-reported classroom be- havior, and less self-reported participa- tion in offending than the control subjects  (Berrueta-Clement et al., 1984~. The find- ings with respect to school achievement were replicated in 10 similar experiments conducted at about the same time, but these experiments did not collect official records of participation in crime or delin- quency (Consortium for Longitudinal Studies, 1983~. Comparisons of officially recorded criminal participation also suggest lower rates among the 58 preschool participants than among the control group of 65: 40 percent of the controls but only 25 per- cent of the participants were charged or arrested as adults.2 While these differ- ences are statistically significant and en- couraging, an intervention based on a sample of this size needs further testing. Replication of the analysis of criminal participation for the other 10 Head Start experiments would be an important next step in furler assessment of the pro- gram's potential for reducing criminal participation. providing intel- and enrichment 2A somewhat smaller difference between pre- school participants and controls was reported with respect to the percent arrested as juveniles (Ber- rueta-Clement et al., 1984~. However, because some arrested juveniles were not included in the analysis, Mat difference is not discussed here.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE Other Interventions Various school- and community-based preventive interventions have been ex- perimentally evaluated with respect to their electiveness in correcting undesir- able behavior in children identifier! as at · 1 CC · ~ ~ · . ~ ~ ~ risk e.g., aggressive, antisocial, vu nerable to delinquency") by parents, teachers, or professional observers. The interventions that have been studied in- clude special school programs to improve self-concept or cognitive skills (e.g., Reckless and Dinitz, 1972; Garrison ant! Stolberg, 1983; Kettlewell and Kausch, 1983), behavior management techniques such as contracting ant! token economies (Stuart, layaratne, and Tripodi, 1976), in- dividual counseling (McCord, 1978; Shore and Massimo, 1979), work-exper- ience programs (AhIstrom and Havighurst, 1971; Hackler and Hagan, 1975; Wodarski et al., 1979), buddy pro- grams (O'Donnell, LyUgate, and Fo, 1979), and- social learning programs (Wodarski and Pedi, 1978; Johnson and Breckenridge, 1982~. The sample sizes in these studies ranged from 32 to 1,000, ant! follow-up periods ranged from none be- yoncI the life of the program up to 30 years. Of these experimental studies, sev- eral interventions seem encouraging, but furler study is needed. One 2-year intervention (Bry and George, 1979, 1980; Bry, 1982) involved points awarded weekly by teachers for good behavior, discussions in group meetings, and extra school trips, and re- ports on behavior were made to parents. A group of 66 seventh-grade students se- lected on the basis of school or family problems was assigned ranclomly to re- ceive the intervention or not. The exper- imental subjects did not experience the subsequent deterioration in school atten- dance and academic achievement that oc- curred in the 33 controls who did not 113 receive the treatment. In a follow-up in- terview 1 year after the program, mem- bers ofthe experimental group reported a participation rate of 33 percent for vandal- ism, auto theft, grand theft, or robbery, compared with a rate of 55 percept among the control group. In another follow-up 5 years after the intervention ended, a search of county juvenile probation rec- ords showed that the participation rate in serious or chronic offending was 9 per- cent ant! 27 percent for experimental and control subjects, respectively. As with the Perry preschool programs, these results are encouraging, but replications in other  sites with larger samples are necessary. Summary: Preventive Interventions The major reviews of experimental ev- idence on the effectiveness of preventive interventions suggest some encouraging possibilities, but they do not establish that any approach tested to date effec- tively reduces subsequent criminal par- ticipation. Heac! Start programs appear to improve school performance, and there is evidence suggesting that one such pro- gram reduced later participation in of- fending, but analysis of data on criminal participation from other Heart Start sites is needed to establish the finding more definitively. Social learning programs for families of antisocial preadolescents have sometimes shown short-term effective- ness in ameliorating the early antisocial behavior that frequently precedes delin- quency, but these programs have not yet been evaluates! as effective in pre- venting clelinquency itself Future ex- periments of this type of intervention need to use longer follow-up periods, inclucle searches of juvenile justice sys- tem records, and compare results across alternative family arrangements, such as single- versus two-parent households. The theoretical underpinnings and pro

114 gram operations of the interventions also need to be made explicit in this research so that results can be aclequately evaTu- atecI and promising interventions tested in other settings. Some of the findings of no difference between treatment and control or com- parison groups may be attributable to reactive effects of the experiment. In pre- ventive interventions clesigned to im- prove conduct at school, for example, children may resent being stigmatized or feel angry at being separated from their classmates. Moreover, since teachers usu- ally know the identities .of the treatment subjects but not of the control group sub- jects, inflated expectations or hostility re- lated to the program could lead to differ- ential subjective performance standards. If either a child's or a teacher's behavior is influenced by the structural character- istics of the experiment, then the program components may not achieve the desired effect. Future experiments of this type of intervention should be aware of these reactive effects. One alternative may be school-based programs that are designed to supplement basic education and are clirected at all enrolled children rather than some selected "high-risk" group (see Greenwood and Zimring, 19851. The design of some interventions may also reflect an incomplete or inaccurate theoretical orientation, which may ac- count for findings of negative effects. Pre- ventive interventions (or career moclifica- tion strategies) groun(led in social learning theory and behavior moclifica- tion may ignore the possibility that cor- rection of the targeted behavioral prob- lems in one setting (say, the home) may not modify them in another setting (such as the school). And, as cliscussed by Loeber (1982), children with conduct dis- orders often display different problem be- haviors in different settings. The typical behavioral intervention, however, fo- cuses on modifying behavior in one set CRIMINAL CAREERS AND CAREER CRIMINALS ting, with little or no transfer of program components such as behavioral con- tracts, reinforcements, or punishments- into other settings. Correction of a prob- lem behavior in the home would be inef- fective in inhibiting participation in crime if it were replacer] by disruptive behavior in the classroom. In this case, postintervention evaluations, especially Tong-term follow-up studies that examine police or court records, would be unlikely to show an effect of the narrowly targeted behavioral intervention. In short, most of the evaluative re-  search on preventive interventions is nondefinitive because of design prob- lems, small sample sizes, inadequate follow-up periocls, and inacloquate ran- domization. Moreover, because ofthe dif- ficulty and cost of large-scare, fully ran- domizecl experiments, research ant! evaluation in this area might advance more quickly under a two-stage process. In the first stage, quasi experiments and experiments with small samples could explore the effectiveness of many types of programs; in the second stage, the pro- grams that had emerged as most promis- in~ could be replicated on a larger scale with a more rigorous research design. Head Start and social learning programs have shown promise in small-scale dem- onstrations, and they appear to be ready for larger-scare studies. For other types of interventions, more attention needs to be placed on design quality and on full reporting of the initial experimental re- sults. CAREER MODIFICATION STRATEGIES In contrast to strategies intended! to prevent nonoffenders from becoming of- fenders, career modification strategies are directed at identified offenders and are intended to reduce the frequency, seri- ousness, or duration of their criminal ca

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE reers.3 Interventions intended to modify criminal careers are most frequently eval- uated in terms of the fraction of offenders rearrested or reconvicted within a speci- fied follow-up period- the recidivism rate. While providing a gross measure of reductions in criminal activity, the recid- ivism rate does not adequately distin- guish which aspects of careers are af- fectecI by an intervention. If follow-up periods are short relative to the average interval between arrests, for example, it is often difficult to distinguish arrest-free intervals during a still-active career from career termination. Moreover, unless re- ciclivism is defined by crime-specific cri- teria, it would fait to detect reductions in offense seriousness a desirable career modification. Few studies, however, have used measures that aclequately tap the different dimensions of criminal careers. A broad array of modification strategies was examined by Lipton, Martinson, ant! Wilks (1975) ant] by the earlier Panel on Research on Rehabilitative Techniques (Sechrest, White, and Brown, 19791. While that panel found no treatment strat- egies that were widely effective in reduc- ing recidivism, they conclucled that some treatments may be effective for certain groups of offenders and recommended more rigorous and imaginative research on offender rehabilitation. Among the in- terventions that the panel consiclered es- pecially worthy of further considerations were alternative sentencing ant! confine- ment arrangements, extensive family in- terventions, employment and vocational programs, ant] increaser! support after an inmate's release from prison. In assessing the crime control potential of such career modification strategies, this pane! did not 30ther strategies can affect behavior through broader environmental or social influences, such as changes in sanction or economic policy. These pol- icies, while not directed at individual offenders, can modify criminal career dimensions. ~5 replicate the previous comprehensive re- views. However, adopting a career per- spective, it die! consider the implications of large-scare experiments involving al- tematives to standard criminal ant! juve- nile sanctions as well as evaluations of family interventions, employment pro- grams, and postrelease support programs not explicitly incorporated in the earlier reviews. In adclition, in view of the dem- onstrated relationship between the inten- sity of substance abuse ant] the frequency of offending, the pane! also reviewed sev- eral recent large-scare studies ofthe effec-  tiveness of treatment of substance abuse in modifying criminal careers. Among the selected studies, some show sufficient general promise to warrant furler explo- ration; they are discussed in the remain- cler of this section. Career Modification: Community-Based and Family Treatment Programs juveniles Inclividually oriented treatment pro- grams for juvenile offenders are typically introducer] as altematives to regular juve- nile court processing (either incarceration or probation). The programs most often inclucle community-based treatment such as counseling, group therapy, or voca- tional programs; treatment of the entire family; or in some cases, diversion from the juvenile court altogether (see reviews by Clarke, 1974a, b; Farrington, 1982; Farrington, OhTin, ant! Wilson, 19861. Studies of the programs permit compari- sons of subsequent recidivism rates for those processes] in the regular juvenile court system (inclucling residential treat- ment and probation) with those assigned to community-based treatment, diversion programs (including family treatment and no treatment), ant! intensive supervised probation. In nearly all the stu(lie(1 pro

116 grams, samples of persons arrested or re- ferred at about the same time and age were randomly assigned to an experimen- tal or control group. Such randomization minimizes problems involving selection bias and lends confidence that postinter- vention comparisons between experi- mental and control groups are a valid measurement of the differential effect of the experimental intervention on offend- ing frequency, or at least on recidivism rates. The majority of follow-up studies on alternative treatments concluded that ju- venile delinquents treated with group therapy, usually supplemented by home visits or intensive parole with additional supervision, had recidivism rates that were no higher than delinquents released from training schools.4 Other studies that reported Tower recidivism rates for alter- native treatments had methodological problems that affected the validity of the results. For example, in the Provo exper- iment (Empey and Erickson, 1972), treat- ment subjects had lower recidivism rates than the training school controls, but the randomization features of the experiment broke down because the judge was reluc- tant to institutionalize some offenders. During a 2-year follow-up period of the California Community Treatment Pro- gram (Palmer, 1974), youths randomly as 4Examples of the studies include Pond (1970), Empey and Lubeck (1971), Palmer (1971, 1974), Jesness et al. (1972), and Lerman (1975~. Institu- tional programs for juveniles and adults were exten- sively reviewed by Sechrest, White, and Brown (1979), and a recent update (Farrington, Ohlin, and Wilson, 1986) reports that group counseling in juve- nile institutions has little effect on recidivism. See also Coates, Miller, and Ohlin (1978) for a discus- sion of the experience in Massachusetts, which closed virtually all its juvenile correctional institu- tions in the 1970s. In their place, the state devel- oped a network of community-based facilities (resi- dential and nonresidential); this change has had no substantial effect on juvenile recidivism rates (see also Lundman, 1984~. CRIMINAL CAREERS AND CAREER CRIMINALS signed to the alternative parole program had lower recidivism rates, based on pa- role revocations, than those sent to an institution. But a later review of these results (Lerman, 1975) pointed out that rearrest rates of the two groups did not differ. Apparently, parole revocations fol- lowing arrest were used more frequently for those who had been institutionalized. Recent experimental studies of alterna- tive diversion programs also report recid- ivism rates that are no different from  those associated with standard juvenile justice system processing.5 In one such study at the Oregon Social Learning Cen- ter (see Reid, 1983), 70 chronic juvenile offenders (average age, 13) were ran- domly assigned either to a privately oper- ated, family-oriented behavioral treat- ment program or a regular counseling program run by the juvenile court, and a third group of 38 similar juveniles was selected as an additional control group. The family intervention techniques in- cludec! training parents to identify and record antisocial behaviors, emphasizing family negotiation strategies for better parent-child interactions, and behavioral contracts for both Me child and parents. Follow-up data from juvenile court rec . 5See Venezia (1972), Baron, Feeney, Thornton (1973), Quay and Love (1977, 1979), Berg et al. (1978), Berg, Hullin, McGuire (1979), Byles and Maurice (1979), Mrad (1979), and Severy and Whitaker (1982~. In addition, one large evaluation of diversion programs as alternatives to either release without treatment or further penetration of the juve- nile justice system reported that the recidivism rates associated with the three groups-diversion, re- lease, and regular processing-were virtually iden- tical in follow-ups conducted after 6 months and 1 year (Dunford, 1981~. This study involved over 2,500 juveniles in four states who were referred for offenses other than personal crimes or status of- fenses, and who were randomly assigned to one of the three comparison groups. Slightly under half had a previous record of arrest and most were male. Diverted juveniles received individual and family counseling along with some employment, educa- tional, and recreational services.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE ores for the three groups revealed that, by 1 year after intake, the juveniles in the experimental treatment group hac! expe- rienced a more rapist decrease in the average number of officially recorclec3 of- fenses than juveniles in the other two groups. However, after 3 years, the differ- ences between the groups hac! disap- pearect. Inconsistent results from other studies that compare personal counseling, family crisis counseling, and specific forms of behavioral family therapy with regular processing suggest the neec! for furler attention to the nature ofthe intervention. Individual and group counseling, whether in institutions or in the commu- nity, appear to have little effect in modi- fying the criminal behavior of serious juvenile offenders. The more recent be- haviorally oriented programs, however, appear more promising, at least in the short term. In one diversion experiment that was based on behavioral reinforce- ment principles, juveniles referred by the police engaged in specialized family pro- grams designed to affect family commu- nication, coping skills, and negative envi- ronmental situations. The experimental group had Tower rearrest rates in the fol- low-up year than those observed in a no-treatment control group in one of two sites (Binder, Monahan, and Newkirk, 1976; Binder, no... The total study in- volved 1,104 juveniles, but the control groups had fewer than 100 youth at each site. Similar experiments carrier! out by Davidson et al. (1977) and Alexander and Parsons (1973) also reported some differ- ences between experimental and control groups, but the sample sizes in those studies were small. These encouraging short-te~-~ results suggest that behavioral approaches to treatment may warrant ex- perimental replication with more serious juvenile offenders, evaluation in other sites, longer follow-up periods, and larger control groups. , ~. . . 117 Adults Experimental research on adults has generally found no benefit from various community-based interventions (e.g., Lamb and Goertzel, 1974; Folkard, Smith, and Smith, 1976; Lich~anan and Smock, 1981), which were not considered by the earlier Pane! on Research on Re- habilitative Techniques. A new type of sentencing alternative, intensive commu- nity surveillance, is presently being con- sidered in several states as an option for aclult felons currently sentenced to prison  or regular probation (see Petersilia et al., 19851. Such programs would usually com- bine multiple weekly contacts with a probation officer with mandatory employ- ment, community service, possible re- strictions on movement (e.g., evening curfews), ant] restitution. This type of intermediate sentence could be imposer! in place of incarceration on offenders who are identified as needing more supervi- sion and control than is usually provided by regular probation, but less than repre- sented by incarceration. Several evaTua- tions of these programs are currently in progress. Summary: Community-Base and Family Treatment Programs Experimental research suggests that for juvenile offenders, community-based treatment does no worse than routine juvenile justice system processing" in- cluding use of probation and training schools in affecting recidivism. These programs may be preferable because of their lower cost and reduced restrictive- ness, but they may not be acceptable in all communities. If incarceration is re- placed by these programs, juveniles gen- erally may view the risk of serious pun- ishment as small, and victims may fee! that justice is not being served. Perhaps most important, few experiments are

118 available that might provide information about the effects of community-based in- terventions on the offense frequency and seriousness of adult offenders, although some recent intensive surveillance pro- grams (for a review, see Petersilia et al., 1985) warrant careful evaluation. Some therapeutic and behavioral inter- ventions appear successful in reducing antisocial and criminal behavior for juve- niles when evaluated in small-scare ex- periments, but no specific intervention emerges as being consistently effective. Many other interventions exist (for exam- ple, outdoor education programs, physi- cally oriented approaches like Outward Bound, and restitution programs for juve- niles), but few of these alternative treat- ments have yet been experimentally tested (see Greenwood and Zimring, 19851. As discussed in the above section on preventive interventions, the general ab- sence of definitive results within a vast evaluation literature suggests a need to focus evaluation resources on careful, well-designed replications of small-scale programs that are especially encouraging, either because of some positive empirical ~_ a, results or promising theoretical perspec- of criminal activity. In some studies, in tive. At present, the results of studies on creases or decreases in daily narcotics use career modification strategies may appear were associated with parallel changes in discouraging pertly because ofthelackof individual offending frequencies (Wish continuity and cumulation that results and Johnson, Volume II; McGIothlin, from an approach to evaluation that views Anglin, and Wilson, 1978; Ball et al., each study as a separate entity rather than 1981; Ball, Shaffer, and Nurco, 19831. as part of an ongoing investigation. Fur- There are also limited data that suggest thermore, many studies that show no ef- that decreased use of alcohol may be fects suffer from poor research designs associated with reductions in violent, as (i.e., small samples, lack of randomiza tion, short follow-up periods); the lack of effect is not necessarily because of inef fective interventions. Careful inspection of these no-effect findings can be useful in designing subsequent follow-up exper iments in a sustained series of cumulative research. However, even promising experimen CRIMINAL CAREERS AND CAREER CRIMINALS tat findings must be tested in practice because interventions that appear suc- cessful in a small short-term experiment may fait to have an effect once the pro- gram is implemented on a larger scale or in the context of the criminal justice sys- tem. Most career modification strategies (as well as preventive interventions) are pilot studies with limited samples, run by a small committed staff. It is possible that any positive results are the result of the commitment of the innovators, therapists, counselors, and other staffto their specific intervention. Several researchers have cautioned that because the level of staff  commitment may decrease when a pilot effort is expanded to a large program or transferred to another jurisdiction, the success of the pilot program may not be renlinat~d in the new setting (see Reid, 1983; Greenwood and Zimring, 19851. Career Modification: Substance Abuse Treatment Recent empirical research on drug use and criminal involvement indicates that, at the individual level, intensity of drug use is strongly correlated with frequency saultive criminal events (see Collins, Vol- ume II; Chaiken and Chaiken, 1982a, 1984~. In light of these findings, the rate of commission of some types of crime by drug-using offenders might be reduced through effective drug abuse treatment programs. A considerable literature exists on the effectiveness of drug abuse treatment

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE programs in reducing individual criminal behavior in drug-using criminals.6 Most research is directecl at the treatment of narcotic affliction, especially heroin, al- though research on treatment for other drugs, such as PCP and cocaine, is in- creasing. Several studies, such as those of the California Civil Addict Program, have shown that court-orclered drug-free out- patient treatment accompanied by super- vision, including urine testing, and weekly visits to a parole officer is associ- ated with recluced criminal activity and higher rates of employment (e.g., McGIothTin, Anglin, and Wilson, 1977; Collins and Allison, 1983; Anglin and McGIothlin, 1984~. Other studies have also found that legal pressure is positively relater! to retention in treatment pro- grams. This research is important be- cause time in treatment is a significant factor in the reduction of criminal behav- ior after treatment (see Collins et al., 19~34; Hubbard et al., 1984~. The evidence regarding the effects on illegal behavior of methadone treatment for heroin afflicts (which is rarely court orderecI) is less consistent. Some studies show modest reductions in nondrug crim- inal behavior, but only methadone clients in treatment for 2 to 3 years have substan- tially fewer arrests (e.g., Lukoff, 1974, 1976; Sechrest, 1979; McGIothTin and Anglin, 1981~. In the Drug Abuse Report- ing Program (DARP), several types of treatment have been examined, and methadone programs also appeared to lower arrests, but the largest reductions in criminal behavior occurred during treatment (see Sells et al., 1976; Simpson et al., 19781. The Treatment Outcomes Prospective Study (TOPS), however, reports more fa 6Some of this literature is reviewed in Wish and Johnson (Volume II); see also Sells (1979), Gandossy et al. (1980), Collins et al. (1982a, by, Cooper et al. (1983), and Kaplan (1983:208-235~. ~9 vorable results. TOPS is following clients voluntarily admitted into a drug treat- ment program of their choice to assess the effectiveness of different treatment serv- ices, although it does not include a nontreatment comparison group (see Hubbard et al., 19831. A recent report indicates that voluntary clients in metha- done maintenance treatment programs self-reported significantly Tower levels of "predatory" criminal behavior 12 months after treatment ended, especially if they remained! in treatment for at least 3 months (Hubbard et al., 1984~. Similar  effects were found for clients in voluntary outpatient drug-free programs. Because the TOPS clients chose to seek treatment and the outcome measure was self-report- ed criminal activity, their motivation to succeed may be an important factor in the results such that they cannot be attributed solely to the effects of the treatment pro- grams. In addition to the criminal activity and drug results in the TOPS study, full- time employment after treatment in- creased for outpatient cirug-free clients, but it declined somewhat for methadone clients (Hubbard et al., 19841. Less is known about the effect of resi- dential treatment programs on criminal behavior for heroin or other drug users, but most of the available research con- cludes that residential treatment appears to reduce criminal behavior (see Coombs, 1981; DeLeon, 1984; Hubbard et al., 19841. These types of programs are most likely to accept clients with the worst recent criminal histories and the most extensive patterns of drug use, but they are also more likely to be resisted by drug users, and some ofthe methods used (e.g., strict punishments, group pressure) have been controversial. Many of these pro- grams are privately operated and usually do not publish statistics on the criminal behavior of clients after treatment. A federal criminal justice initiative known as TASC (Treatment Alternatives

120 to Street Crime) is directed at diverting drug users including heroin users and addicts, as well as those abusing alcohol, barbiturates, PCP, and other drugs into treatment instead of imprisoning them (U.S. Department of Justice, 19791. The programs often divert those for whom prison is seen as too severe, such as occa- sional marijuana users charged with pos- session. Although TASC is no longer fed- erally funded, it continues in areas where it has local support. One recent evalua- tion of TASC programs for 1979-1980 compared treatment outcomes for TASC referrals, other types of criminal justice referrals, and clients not involved in the criminal justice system (Collins et al., 19841. The comparison of pre- and post- treatrnent behavior showed criminal par- ticipation was reduced in the year after treatment (either outpatient drug-free or residential programs) for clients from all referral sources. This reduction was stronger when the indicator of illegal be- havior was arrests rather than self-report- ed behavior and when the clients were under some pressure from the criminal justice system. Moreover, legal pressure (e.g., periodic urine testing, threat of pro- bation or parole revocation) significantly increased retention in treatment, which, in turn, led to lower participation in crim- inal behavior in the year after treatment ended as measured by both arrests ant! self-reports. However, no treatment-free control group was included in the study, and the tote follow-up period was only 1 year. In assessing the effectiveness of all types of drug abuse treatment programs, methodological difficulties often con- found the evaluations of program effects. The most serious problems concern the lack of random assignment to different programs, inadequate follow-up periods, failure to follow up all clients (including program dropouts), limited use of control groups, and lack of proper controls for CRIMINAL CAREERS AND CAREER CRIMINALS other relatecl factors (Lukoff, 1974; Nash, 1976; Gerstein, 19811. Any comparison across types of treatment programs (i.e., methadone maintenance, residential treatment, drug-free outpatient programs) is especially problematic because of sub- stantial differences in research designs and client populations (Gandossy et al., 19801. More cared! experimental studies are thus desirable, especially to evaluate whether drug abuse treatment has pri- marily an incapacitative effect (through surveillance or resiclential programs) or a rehabilitative impact on drug-using ot-  fenders. Career Modification: Interventions to Improve Employment Status The empirical association observed between chronic unemployment and fre- quent criminal activity suggests that in , ~ terventions that improve offenclers' em- ployment status might also modify their criminal careers. Of the approaches that have been tested as methods of reducing offenders' unemployment, the panel con- sidered two that have been evaluated in terms of their effect on offending: occupa- tional skills training and postrelease em- ployment assistance to offenders. Occupational Skills Training: Job Corps lob Corps is a comprehensive program that provi(les occupational skills training, basic and remedial education, and other services to disadvantaged youths in a res- idential program that averages about 6 months. Enrollees in lob Corps are pri- mariTy minority mates between the ages of 16 and 21, and they generally possess characteristics associated with a high risk of participation in crime, such as low school achievement and erratic employ- ment histories. Of youths in Job Corps cluring the summer of 1977, at least 38

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE percent had been previously arrested and 19 percent had been convicted (Maliar et al., 1982:14~. Although lob Corps enroll- ment is officially voluntary, it is wiclely believed that a defendant's promise to enroll in lob Corps is frequently obtained by a judge in lieu of stanciarc! criminal or juvenile justice proceedings. In its comprehensive assessment of evaluations offederally funcled youth em- ployment programs, the National Re- search CounciT's Committee on Youth Employment Programs (Betsey, Hollis- ter, and Papageorgiou, 1985) cited lob Corps "as a program for which there is strong evidence concerning program ef- fectiveness [in achieving gains in em- ployment and earnings.] The quality of the evaluation reviewer! [(MalIar et al., 19821] . . . lends confidence to these con- clusions." That evaluation covered crim- inal activity and found different program effects on the criminal career during the resiclential stay and the postresidential period. The evaluation reports an esti- mated reduction in the annual arrest rate per participant during the period of pro- gram participation, but the estimation methodology, which includes both of- fenders and nonoffenclers in computing rates, does not permit separation of par- ticipation in criminal activity from fre- quency effects. The authors attribute the reduction to the incapacitative effect of fences. restricted activity and monitored behav- ior in the residential setting and to the reduced incentive to commit crimes be- cause participants' material needs are met in the program. In a 4-year postprogram follow-up, crime-reduction effects were more ambiguous and crime-specific: ar- rest rates for the crimes of murder, rob- bery, and larceny or motor vehicle theft were lower for participants than for a matched sample of nonparticipants in lob Corps; rates were higher for burglary, drug violations, and some categories of personal crimes. ]2] Postprison Assistance The MacArthur Foundation (Far- rington, Ohlin, and Wilson, 1986) re- cently reviewed four experiments involv- ing postrelease assistance to expedite the adjustment of prisoners in the commu- nity. In one Danish experiment, 250 pris- oners were randomly assigned just before release either to a treatment group that was given special help (finding work and accommodations, financial assistance, help with family relations, and negotia- tions with creditors) or to a control group that was not offerer! the help. Signifi- cantly fewer of those receiving the help  were reconvicted during a follow-up pe- riod of at least 6 years after release from prison, and the difference was greatest among those win the highest expected rates of recidivism (Berntsen and Christiansen, 19651. The results were es- sentially replicated in a British study with a 2-year follow-up period (Straw, 1974), but the intervention failed in another rep- lication with a 1-year follow-up period (Fowles, 19781. In addition to the shorter follow-up periocI, the experiment that failed was targeted at short-term prison- ers, and they may not have neecled the postrelease assistance as much as the sub- jects in the successful experiment, who had served relatively Tong prison sen Two major American experiments have tested the theory that unemployment counseling and benefits, by reducing the financial and employment difficulties of released prisoners, would also reduce re- ciclivism rates. In a Baltimore pilot exper- iment (Living Insurance for Ex-Prisoners, or LIFE), over 400 released prisoners who were considered high risk were ran- domly assigned to a control group Mat received no benefits or to a treatment group that received unemployment ben- efits (of about $60 a week for up to 13 weeks) and job placement assistance

122 (Ross), Berk, and Lenihan, 1980~. The benefits were reduced by 50 percent of all earnings above $40 per week and were interrupted during any period of incarcer- ation. Within a 1-year follow-up period, the releasees who received the financial assistance experienced a 22 percent rear- rest rate for theft compared with a 31 percent rate for the control group while rates for violent crimes were virtually identical in the two groups. The nonfinan- cial placement assistance showed virtu- ally no effect on the rearrest rate. These findings were encouraging and so led to a more ambitious follow-up pro- gram, the Transitional Aid Research Project for Ex-Offenders (TARP). In TARP, representative samples of about 975 releasees each from Texas and Geor- gia prisons were randomly assigned to experimental and control groups and monitored through interviews and ad- ministrative records. The groups receiver! no assistance, placement assistance only, or unemployment benefits according to one of three schedules that varied in terms of duration and the rate at which benefits were reclucec] to offset earnings. The overall finding showed no difference in rearrest rates between the groups that received assistance and those that did not during a 1-year follow-up after release (Berk, Lenihan, and Rossi, 1980:777; Zeisel, 1982a:3781. Berk, Lenihan, and Rossi interpreted this result in terms of two offsetting ef- fects of the financial assistance: (1) a cli- rect reduction in the incentive to engage in criminal activity, which would reduce rearrest rates; and (2) a reduction in time worked, which wouIc3 increase spare time, leading indirectly to an increase in rearrest rates. They then argued that with a payment plan stripper! of the work dis- incentive effects, "modest amounts of fi- nancial aid can reduce recidivism among ax-felons" (1980:7841. Their result is con- sistent with the hypothesis of a null effect CRIMINAL CAREERS AND CAREER CRIMINALS of financial assistance on offending be- havior, as later argued by Zeise] (1982a). Zeisel also pointer] out that even under the countervailing-effect interpretation of Berk ant! his colleagues, it was not clear that such a payment plan without work disincentives could be clevised; both Rossi, Berk, and Lenihan (1982:393) and Zeisel (1982b:396) states] that additional controlled experiments were the only re- liable means of determining whether such a plan could be clevised. INCAPACITATION STRATEGIES  Knowledge of criminal careers is cen- tral in trying to estimate the number of crimes avoided by removing an offender from society, that is, the incapacitative effects of criminal justice policies. Inca- pacitation theory is based on the recogni- tion that an offender cannot commit crimes in the general community while he is incarcerated. However, he may commit crimes in prison or his street crimes may be carried out by someone else (drug sales, for example), and these effects diminish the incapacitative effect. Incapacitative effects are achieved pri- mariTy by incarceration; criminal justice system policies on bait, sentencing, and parole release affect the probability or duration of incarceration. As described at the beginning of this chapter, there are two kinds of incapacitative effects: selec- tive and collective. Incapacitative effects can be increased through collective incapacitation policies only by increasing the total level of incar- ceration; with selective incapacitation policies, the incapacitative effects might be increased by reallocating current ca- pacity primarily to offenders who repre- sent the greatest risk of offending. Collec- tive incapacitation policies are consistent with the equal treatment concerns of a just-deserts retributive sentencing policy, while selective incapacitation policies fo

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE cus as much on the offender as the of- fense. Of course, the degree to which selective incapacitation policies couIcT be effective clepencis critically on the ability to distinguish Tow-risk from high-risk of- fenclers, i.e., offenders who are preclictec! to represent the greatest risk of future crime in terms of frequency, seriousness, or career duration. Furthermore, the op- portunity to pursue selective incapacita- tion policies will always be circum- scribec! by the legal anc] ethical concerns that punishments for comparable offenses not be grossly clissimilar from one another nor grossly disproportionate with the se- riousness of the offense. Research on incapacitative effects, con- siclering only collective incapacitation policies, was reviewer] by a previous panel (Blumstein, Cohen, anc! Nagin, 19781. A brief update of some subsequent results on collective incapacitative effects is proviclec! here, but the major focus is recent explorations of selective incapaci- tation. The technical cletails of various aspects of estimating incapacitative ef- fects are in the technical note at the enc! of this chapter. Collective Incapacitation Estimates of collective incapacitative effects focus on average consequences for crime reduction associated with incarcer- ation. Reviewing various analyses of col- lective incapacitative effects, Cohen (1978:201 - 210; 1983: 12 - 21) estimates that, uncler incarceration policies prevail- ing in the early 1970s, crime reduction achiever! through collective incapacita- tion was less than 20 percent. Achieving furler reductions in crime from 1970 levels through increases in collective in- capacitation were likely to be similarly moclest anc! wouIc! have requirec! major increases in prison populations. Specifically, Cohen (1978:226, 240- 241) estimated that, with an average A of 5 |23 inclex crimes per year per offender anc] expected time server! per inclex crime of .02 year (a number exceeding the value founc! in many states), a 10 percent recluc- tion from 1970 levels of inclex crimes wouIc! have requirec! that the expecter! time server! per inclex crime, anc! thus the prison populations, be more than clou- blecI. For higher values of A, the requirec! prison population increases from 1970 levels are proportionately reclucecI: if A were cloublec! to 10, for example, a prison population increase of about 50 percent wouIc! have yielclec! the same 10 percent reduction of crimes; alternatively, with A  of 10, a cloubling of prison populations wouIc! have resultec! in about a 20 percent reduction of crime. Even these estimates of incapacitative effects are optimistic. Crimes committee! by groups of offenders might not be clis- ruptec! by incarcerating a single offender: for example, an auto theft ring may re- place the contributions of the incarcer- atec! member. Moreover, the estimated incapacitative effect that results from an Increase In average hme server . per crime is overstated because the estimates as- sume that if free in the community, of- fenclers wouIc] continue committing crimes at rate A, while in fact the criminal careers of some incarcerated offenders wouIc! terminate naturally cluring their incarceration. All of these factors Tower the crime reduction achiever! through collective incapacitation. (The potential deterrent, rehabilitative, or criminogenic effects of aciclitional incarceration are not consiclerec! here.) The estimates of collective incapacita- tive effects just cited were clevelopec! on the basis of U.S. incarceration experi- ences cluring the 1960s anc! early 1970s. Since that time, prison populations have increased substantially in the Unitec! States; between 1973 anc! 1982, for exam- ple, the number of inmates in state anc! fecleral institutions almost cloublecI, from

24 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE 5-1 Estimates of Crime Reduction from Collective Incapacitation Associates] with Almost Doubling State and Federal Prison Populations Between 1973 and 1982 Robbery 1973 1982 Burglary Increase 1973 1982 Increase Observed crime rate (UCR reported crimes per 100,000 population) 183.1235.9 +28.8% 1,222.51,484.5 +21.4% A = 14.3 A = 36.7 Upper-Bound: Prevented crime rate (UCR reported crimes by inmates per 100,000 popu ration) 92.0174.2 +89.3% 422.5790.5 +87.1% Incapacitative effects 33.4%42.5% +27.2% 25.7%34.7% +35.0% A = 5.0 A= 15.0 Lower-Bound: Prevented crime rate (UCR reported crimes by inmates per 100,000 popu lation) 32.260.9 +89.1% 172.7323.1 +87.1% Incapacitative effects 14.9%20.5% +37.6% 12.4%17.9% +44.4% aIncapacitative effect = Prevented crimes/prevented and observed crimes. SOURCE: Data from Cohen (1985b). 204,211 to 396,072. This 94 percent in- crease in imprisonment provides a natu- ral setting for assessing the gains in crime reduction achieved from increased col- lective incapacitation. Using a model described below, Cohen (1985a,b) examines the collective incapac- itative effects of increases in incarceration nationally during the 1970s. The esti- mates of incapacitative effects in Table 5-1 compare the crime rates actually re- ported by the FBI in the annual Uniform Crime Reports (UCR) to comparable esti- mates of the prevented crime rates attrib- utable to inmates, i.e., the increment to the reported crime rate that would have occurred had the incarcerated inmates been free in the community.7 7To develop a "prevented crime rate" for inmates that is comparable with the reported crime rate, the "prevented crime rate" includes only those "pre- vented crimes" for inmates that would have been reported in the UCR. In particular, multiple count- ing of the same crime committed by several of- fenders and crimes that are not reported to the police are removed from total "prevented crimes" for inmates. A range of values of A for inmates is examined. Nationally, for robbery, when A is large, the incapacitative effect is esti- matecl to have prevented 33.4 percent of the robberies that would have been com- mitted in 1973 and 42.5 percent of those in 1982, a gain in the incapacitative effect of 27.2 percent. The number of crimes prevented through incapacitation is esti- mated to have increased by 89.3 percent (from 92.0 crimes per 100,000 prevented  in 1973 to 174.2 in 1982) because ofthe 94 percent increase in incarceration. The percentage of crimes averted through in- capacitation does not increase compara- bly, however, because the reported crime rate itself increased by 28.8 percent be- tween 1973 and 1982, perhaps reflecting increases in the number of active offend- ers or in their frequency, or increases in the rate of reporting crimes to the police. The magnitude of crime reduction is comparable but somewhat Tower for bur- glary: 25.7 percent in 1973 and 34.7 per- cent in 1982, for a 35.0 percent gain in the incapacitative effect. When lower values of A are used, the incapacitative effects for

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE robbery and burglary are estimated to be only in the range of 12 to 20 percent, with gains in the incapacitative effects of 37.6 percent for robbery and 44.4 percent for burglary (see Table 5-11.8 The procedure user] to estimate the ~ncapac~ranve effect (i.e., crimes that would have been committed by inmates had they been free in the community) highlights some important considerations in estimating incapacitative effects. The procedure relies fundamentally on the product of the number of inmates who would be active if not in prison times an estimate of their individual crime rate, A. The value of A for inmates is likely to be higher than that for street offenders, since prisoners are disproportionately drawn from the population of high-rate offenders with their greater exposure to arrest.9 The overrepresentation of the population of high-rate offenders among inmates wads . . . ~ , . 8These estimates are comparable to the earlier projections of 10 to 20 percent furler reductions in crime if prison populations were doubled from 1970 levels. Starting from a base level of 10 percent of crimes averted in 1970, a further 10 percent reduc- tion from the 90 percent of crimes actually commit- ted would yield an additional 9 percent of crimes prevented, for a total incapacitative effect of 19 percent. Likewise, from a base incapacitative effect of 20 percent in 1970, a furler 19 percent reduction from the 80 percent of crimes actually committed would add 16 percent to crimes prevented, for an incapacitative e£ect of 36 percent. The expected total incapacitative ejects from increased incarcer- ation in the range of 19 to 36 percent are comparable with the 18 to 20 percent estimated for low values of A in 1982 and the 35 to 43 percent estimated for large values of A (see Table 5-1~. 9In view of many chance factors involved in making an arrest, it is reasonable to anticipate that the arrest risk per crime is not likely to be very different across offenders except for a few highly skilled offenders. Thus, with a common arrest risk per crime for all offenders of 10 percent, the greater vulnerability to arrest for high-rate offenders is illus- trated with a simple example: an offender who commits only one offense per year will be arrested on average only once every 10 years, but an offender who commits 10 crimes each year will be arrested on average once every year. 125 be even greater if there is any selectivity in criminal justice system processing that increases the incarceration risk per arrest for high-rate offenders. Thus, estimates of the crimes that incarcerated inmates would otherwise be committing should rely on estimates of A that are representa- tive of inmates.  The two available estimates of A based on self-reports by inmates diverge consid- erably.~ Estimates of A for incoming pris- oners, for example, range from an annual average of 15 burglaries per offender (Peterson and Braiker, 1980:Table 12) to 204 burglaries per offender (Chaiken and Chaiken, 1982a:Table A.31. Because of the sensitivity of estimates of incapacita- tive effects to the value of A used, our analysis was done for a range of A values. With appropriate adjustments, especially for the sensitivity of mean annual rates to a few people with very high rates, the Inmate surveys provicle the basis for rea- sonable bounds on A for inmates (see Chapter 3 and Cohen, Appendix B). The first survey provides a Tower-bound esti- mate of A of 5 for robbery and 15 for burglary (Peterson ant! Braiker, 1980:Ta- ble 12~. The upper-bound values of A were derived from the panel's reanalysis of the data in the second inmate survey (Visher, Volume II). To avoid sensitivity to the extremely high rates reported by a few offenders, the mean A was estimated by assigning the 90th percentile value of A to all offenders with inclividual rates above that value: this 90th percentile value was 196 burglaries per year and 71 robberies per year. The resulting mean values of A are 14.3 for robbery and 36.7 for burglary (Cohen, Appenclix B:Table 61. In estimating incapacitative effects, the number of inmates multiplied by A must be adjusted by a number of factors that reduce the number of crimes prevented by incarceration: (1) only some inmates participate in any particular type of .

126 crime; (2) some criminal careers termi- nate during periods of incarceration; (3) some offenses are committed by groups of offenders; and (4) not all crimes are re- ported to the police or recorder! in the Uniform Crime Reports. These adjust- ments are explained below. Only a portion of all inmates would actively commit the crime type being examined if they were free in the commu- nity. In responses to the Ranc! inmate survey, just over 35 percent of prison inmates report committing robberies and 45 percent report committing burglaries (Chaiken and Chaiken, 1982a).~° Natu- rally, as the crime type examined is broadened, a larger percentage of in- mates would be active in that offense type. For the broad category of all index offenses, it is likely that only a small portion of inmates would not qualify as active offenders. Natural termination of criminal careers while incarcerated also recluces the num- ber of inmates currently active in an of- fense type. Such career termination is of less concern when residual careers are much longer than average time served. On the basis of analyses of career length for robbery and burglary in Michigan, 10Inmates from three states California, Michi- gan, and Texas were surveyed. The responses var- ied across the states, with responses in Michigan falling between those in California and Texas. In- mates in California appear to include a higher pro- portion of offenders who are active in serious crimes (49 percent for robberies and 54 percent for burglar- ies) and who report committing these offenses at high rates. Texas inmates, by contrast, include a higher proportion of less serious offenders. The responses for the three states combined were used to represent inmates nationwide. iiInmates not active in index offenses might include, for example, specialists in drug offenses or in white collar offenses like fraud, forgery, and embezzlement-offenses that together comprised 11.4 percent of current conviction offenses for U.S. inmates in 1979 (Bureau of Justice Statistics, 1982b). CRIMINAL CAREERS AND CAREER CRIMINALS Cohen (1985a, b) uses estimates of an average of 10 years remaining in either robbery or burglary careers. With an av- erage time served of 2 years for robbery, 83 percent of inmates who would commit robbery are estimated to remain active in robbery careers cluring periods of incar- ceration; for time served averaging 1 year for burglary, 91 percent of inmates who would commit burglary are estimated to remain active in burglary careers during periods of incarceration. Another adjustment relates to the indi-  viclual crime rate, A. Direct use of A yields all crimes committed by each indiviclual offender. Some offenses, however, are committed by groups of offenders, and the estimated number of offenders per crime must be used to eliminate multiple counting of the same crime. Data from national victimization surveys lead to es- timates of 2.3 offenders per robbery and 1.6 per burglary (derived from Reiss, 1980b:Table 2~; this adjustment reduces prevented crimes by 57 an`137 percent, for robbery and burglary, respectively. Finally, compared with reported crimes published in the FBI's UCR, pre- vented crimes should be discounted for crimes that would not be reported to the police. With typical reporting rates from victim surveys of about 50 percent for robbery and burglary, this adjustment represents another reduction by one-half to the estimate of prevented crimes (Bu- reau of Justice Statistics, 1983~. Combining the various adjustments, the estimate of crimes prevented for in- mates is given by: UCR Crimes Percent Who Prevented Number of x P ti i t for Inmates Inmates in Offense Percent not Adjusted , Terminating Individual Careers While Crime Rate Incarcerated (A*) where

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE Percent Individual A* = Crime Rate x /A\ Reported ~ ~to Police . Number of Multiple Offenders Per Crime. The combined adjustments are consider- able, reducing the number of prevented crimes obtained from the simple product of inmates times A by about 16-fold for robbery and 8-fold for burglary; this cal- culation yields estimates of prevented crimes that are comparable to UCR re- ported crimes. The final estimates for a range of values of A are presented in Table 5-1 (above). Failure to make the adjustments to prevented crimes would grossly overstate the incapacitative effect associated with changes in inmate popu- lations.~2 The increases in inmate populations experienced cluring the 1970s may have changed the incarceration rates per crime since the 1970 estimates provided by Cohen (19781. These chances would alter the estimates of the incapacitative effects that can be expected from further in- creases in incarceration. The technical note et the end ofthis chapter updates the estimates of potential incapacitative ef- fects to reflect both 1981 incarceration levels and the effect of career termina- tion. Despite the large increases in inmate populations nationally, the expected time served per index crime remains relatively low because of similar increases in the . . ~. number of reported crimes. In 28 of 31 i2The estimates of crime reduction from incapac- itation presented here for a recent 94 percent in- crease in inmate populations nationally are similar in magnitude to alternative estimates of crime re- duction for a smaller 50 percent increase in the California inmate population presented by Green- wood (1985~. The estimated crime reduction from each percent increase in prison population was thus higher in Greenwood's estimates for California, but his preliminary estimates of crime reduction were likely to be somewhat inflated because he did not take account of all the adjustments noted here. ]27 states examined, the expected time served per index crime was under .025 year in 1981: prison population increases of at least 100 to 200 percent would still be required to reduce reported index  crimes by 10 percent from 1981 levels (see Table 5-6 in the technical note). In nine of those jurisdictions, where the ex- pected time served per index crime was uncler .01 year, prison population in- creases well in excess of 200 percent would be required to reduce reported index crimes by 10 percent. The large increases in prison populations required in order to reduce crime reflect the fact that only a small fraction of active index offenders are in prison at any time. Even a small decrease in the number of of- fenders free in the community represents a large increment to the incarcerates! pop- ulation. The lower the current incarcera- tion risk per crime, the greater the in- crease in prison population required to attain a given percentage reduction in crime. Incorporating career length, and the associated possibility that some criminal careers end during periods of incarcera- tion, reduces estimates of the incapacita- tive effect and increases estimates of the prison population that would be required to achieve a 1 percent reduction from 1981 crime levels. This effect of career termination increases as time served by those incarcerated gets longer and as the fraction of offenders who have terminated their careers while incarcerated in- creases. For states where time served by those incarcerated is 1 year or less, incor- porating the effect of career termination increases the estimated growth in prison population by 20 percent; where time served is 1.5 to 2 years, the estimated growth increases by 40 percent (see Ta- ble 5-8 in the technical note). In summary, recent studies have exam- ined the crime reduction effects of collec- tive incapacitation from the substantial

128 increases in inmate populations in the United States during the 1970s. After ad- justments to A and to the proportion of active offenders among inmates, the near doubling of the inmate population be- tween 1973 and 1982 is estimated to have raised the aggregate incapacitative effect of prison for burglary from 25.7 percent in 1973 to 34.7 in 1982, a growth of 35 percent when A is large. The improve- ments in incapacitative effects for other cases shown in Table 5-1 are of similar magnitude. It is estimated that furler reductions in reported index crimes from 1981 levels would require at least 10 to 20 percent increases in inmate populations for each 1 percent reduction in crime. Despite large increases in inmate populations since 1970, the incarceration risk per index crime remained Tow in 1981, less than .02 year per crime in 24 of 31 states examined (see Table 5-6, below.) As was found in the earlier analysis of 1970 data (Cohen, 1978), when the prevailing incarceration risk per crime is low, even very large increases in inmate populations lead to only modest percentage reductions in crime through collective incapacitation. Selective Incapacitation The recent large increases in inmate populations continue to place pressure on available prison and jai! resources and to highlight the desirability of targeting in- carceration more narrowly to achieve greater crime reduction from the avail- able limited capacity. Increasing interest in a strategy of selective incapacitation has been further sparked by observations that inclividual offending frequencies dif- fer widely among offenders, with a small number of offenders committing crimes at very high rates (Peterson and Braiker, 1981; Chaiken and Chaiken, 1982a, b). Such a skewed distribution suggests the possibility of increasing the crime reduc CRIMINAL CAREERS AND CAREER CRIMINALS lion benefits from incapacitation by selec- tively targeting incarceration on the small number of high-rate offenders. This objective is one justification for habitual offender statutes, which existec! in 47 states as of 1981 (Cooper, Kelley, and Larson, 19821. Habitual offenders are generally defined by the length and seri- ousness of their prior conviction records, and sometimes by the nature of the cur- rent offense. While the statutory provi- sions vary widely from state to state, con- victed offenders who are designated "habitual" typically face an extra 1(~20 years in addition to the regular sentence,  or the possibility of a life term. Such enhancement of punishment could be consistent with a desire for increased ret- ribution against those who persistently violate. The incapacitative efficiency of these laws, however, depends on the re- sidual length of these offenders' criminal careers compared to their total time served. If the total time remaining in the career is less than the total time served, the time served after the career would have terminated represents a waste of prison space from an incapacitative per- spective. Almost certainly, life sentences represent such a waste. Criminal careers longer than 20 years are also very un- likely, but research is needed on the re- sidual career length of those meeting ha- bitual offender criteria. Because they attempt to identify a high-risk subset of convicted offenders for enhanced sentences, habitual offender statutes are an example of an attempt to achieve selective incapacitation. The po- tential usefulness of selective incapacita- tion rests on the ability to classify individ- ual offenders in terms of their projected criminal activity. Such policies also raise important ethical concerns. Some of those concerns relate to the quality of the cIas- sification rules, especially in terms of er- roneous classifications, and others relate to constraints on which classification vari

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE ables may be used; these issues are acI- dressed in Chapter 6. The discussion here focuses on the ethical concerns in justifying differential sentences. Ethical Concerns Under selective incapacitation poli- cies, Tong prison terms wouIc! be reserved primarily for offenders identifier! as most likely to continue committing serious crimes at high rates. Offenders not so cIassifiecI would be released from prison after short terms or would be subject to nonincarcerative sanctions. Selective in- capacitation thus permits, and indeed en- courages, different sentences for the same offense to accommodate the differences in crime potential among offenders. From a crime control perspective, such policies are attractive because of the promise they hoIct of preventing more crime for a given amount of prison space. Fundamentally, ethical concerns about the use of offender classifications to struc- ture criminal justice decisions for pur- poses of selective incapacitation are rooted in the challenge to the legitimacy of basing punishment of an individual on the possibility of future crimes rather than only on the crime already commit- ted. In the extreme, a pure just-deserts model of criminal justice would preclude all considerations other than the serious- ness of the instant offense and the of- fender's culpability in having committed it. That model is thus concerned primar- ily with the offenses committed rather than with offenders' criminal careers and it implies uniformity in sanctions for like offenses and proportionality among sanc- tions for different offenses. However, pro- tection of the community is also widely accepted, both philosophically and oper- ationally, as a legitimate objective of criminal justice decision making (Harvard Law Review Student Note, 19821. In this model, criminal justice cle |29 cisions, including imposition of prison sentences, are intended not only to pun- ish offenders for past committed crimes, but also to protect the community from future crimes that the offenders might commit if they were set free. Distinguish- ing between the justification for punish- ment in general and its precise allocation, this broacler view of sanctions encom- passes both retributive and utilitarian concerns (Hart, 1968), and it allows a limited degree of variation in the sanc- tions imposed for similar crimes on the basis of differences among offenders. Recent work has moved the just-  deserts and public protection positions closer to a mutually acceptable accommo- dation. In particular, public protection is posed not in opposition to just deserts; rather, the two positions are offered as matters of joint concern if. Monahan, 1982; Morris and Miller, 1985; van Hirsch, 19851. In these formulations, just deserts serves as a limiting principle in which the criteria of blameworthiness and the seriousness of the instant offense establish a range of acceptable penalties. The choice of a sentence within that range that enhances public protection is posed as a reasonable subsidiary princi- ple. In selective incapacitation policies, dif- ferences in anticipated future crimes for different offenders convicted of the same offense would be used to choose their respective sentences, but only within the constraints on the range of acceptable differences in sentences for high- and Tow-risk offenders. The desired range of differences in sentences varies from min- imal (von Hirsch and GottEredson, 1984), to a reasonably narrow range reflecting the imprecision in measuring the appro- priate "desert" if. Monahan, 1982; Morris and Miller, 1985), to largely uncon- strained ranges like those reflected in typically broad statutory limits (e.g., Greenwood, 19821. The degree of varia

130 lion recommended for sentences may be influenced in part by cleveloping knowI- edge about the criminal career, particu- larly by the accuracy of predictions about subsequent careers. I. Monahan (1981) suggests that the greater the accuracy of predictions, the greater the variation in sanctions that might be tolerated. Another consideration in evaluating the acceptability of a selective incapacita- tion policy is the tradeoff between the additional crime control advantage and the level of disparity resulting from dif- ferential sentences. For example, wouIc] a potential 10 percent reduction in crime be sufficient to justify sentences as clispar- ate as 10 years for high-risk offenders and 1 year for all other offenders? WouIcI 20 percent-or 50 percent-crime recluc- tions be enough? Alternatively, would a 10 percent reduction in crime justify a more limited sanction differential, for ex- ample, of 2 to 1? Ethical questions cannot be resolved on empirical grounds, but empirical re- sults exploring the differential outcomes associated with alternative policies can be valuable in providing a factual base for consideration. While the ethical con- straints are absolute for some, others ar- gue that improvements in identifying the most serious offenders warrant some ex- pansion of a permissible sentencing range. Sentences Based on Offender Characteristics: The Rand Inmate Survey Scale Interest in exploring selective incapac- itation policies was one stimulus for the seconc! Rand inmate survey, in which 2,190 incarcerated offenders from three states were asked for the information needed to estimate their indivi(lual crime rates for robbery and burglary. When analysis of the survey data confirmed very high values of A for a small number of CRIMINAL CAREERS AND CAREER CRIMINALS offenders (Chaiken and Chaiken, 1982a), further interest was stimulate<] in clevel- oping statistical instruments that were ca- pable of distinguishing among offenders in terms of A. Using bivariate correlations, Greenwood (1982) developed a seven- factor scale reflecting prior criminal record, drug use, and employment to dis- tinguish Tow-, medium-, ant! high-rate of- fenders among the inmate sample. For C;aTifornia inmates, mean indiviclual crime rates were reported to vary sharply and in the expected direction, with A for inmates cIassifiec! as high rate by the scale being 12 to 14 times higher than the  mean for those classified as low-rate in- mates. The difference in mean rates was less dramatic for Texas and Michigan in- mates: the mean A for high-rate offenders was 6 to 9 times higher than that for Tow-rate offenders in Texas and less than 3 times higher in Michigan (Greenwood, 1982:Table ES.11.~3 Greenwood (1982) estimated incapaci- tative effects on crime volume ant! prison population for alternative incarceration policies that targeted the pre(lictecl high- rate inmates for Tong prison terms while limiting sentences for other inmates. These estimates relied on mean crime rates for inmates iclentified as low-, me- dium-, and high-rate offenders by the scale, Ai, together with statistics from other sources on incarceration risk for offenders, I'*. To estimate I`*, other data the number of crimes reported, the fraction of all crimes that are reported to police, the number of offenders per crime incident, the number of arrests were used to estimate the conditional probabil- ity of arrest, conviction, and incarceration after an offense, 7, which was assumed to be invariant across the three offender groups. Also Si, the average time served for an incarceration sentence, was esti i3The derivation and predictive accuracy of the Rand scale are discussed in Chapter 6.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE mated separately for each offender group i, using the expected terms reported by the Rand respondents. Both.l and Si en- tered the estimate of l'*. The total num- ber of active adult offenders both incar- cerated and not in each group, N`, was then estimated from the number of in- mates, Ri*, and the varying risks of incar- ceration for each offender group, Ii*, with Ni equal to Ri*/Ii* (see technical note for details).~4 With these parameter estimates, Greenwood analyzed the effects of a se- lective incapacitation policy on crime rates and on prison population in CaTifor- nia (using Equations 5-1', 5-2', 5-9, and 5-10 ofthe technical note). The total num- ber of crimes and inmates associated with the prevailing sentencing practices serve as a benchmark for comparing the out- comes of alternative selective incapacita- tion policies. Although Greenwood ana- Tyzed six alternative policies separately for robbers and burglars, one of the anal- yses his most selective "Policy 6" for robbers has attracted particular atten- tion, probably because it emerged from his analysis as the most effective and because robbery is the more serious crime. Therefore, the remainder of this discussion is focused on that analysis. In the analysis of Policy 6, Greenwood suggested that a selective incapacitation policy of increasing the time served in prison for incarcerated robbers classified as high-rate offenders, while limiting the jai] terms of other incarcerated robbers to 1 year, would produce a more advanta- geous tradeoffbetween crime control and prison population effects than could be achieved by collective incapacitation pol- icies. Greenwood (1982:xix, 79) reports selective incapacitative effects for a range of increases in prison terms: a 20 percent Greenwood (1982:11~116) uses a complex procedure to estimate the numbers of inmates in prison and jail in each offender group, Hi* |3] reduction in robberies by adults is associ- ated with no change in the number of robbery inmates; with a smaller increase in prison terms for high-rate robbers, it would be possible to achieve a 15 percent reduction in robberies by adults with a 5 percent reduction in robbery inmates. Greenwood's methods and results have been closely scrutinized by Cohen (1983, 1984b), by Spelman (1984), and by van Hirsch and Gottfredson (19841. These other analyses of the same data suggest that the original estimates overstate the effects of the proposed selective incapac-  itation strategy. Because their work and related work by Chaiken and Chaiken (1982a) raised questions about the vaTid- ity of Greenwood's results, the panel commissioned a reanalysis of the original survey data and recomputation of the incapacitative effects estimated by Greenwood. The reanalysis (Visher, Vol- ume II) reproduced fairly closely the Chaiken and Chaiken estimates of the A distribution from the survey instrument and the power of the scale developed by Greenwood to distinguish among con- victed robbers in the California inmate sample in terms of offending rates (Visher, Volume II:Tables 9, 171. How- ever, it raised important questions about the ability of the scale to discriminate among offenders in the subsamples of burglars and robbers from other states (see Chapter 61. Some differences also emerged in the estimated magnitude of the effects that might be expected from pursuing selec- tive incapacitation strategies (see Table 5-21. While Greenwood does not specify the exact prison terms for the alternative estimates of selective incapacitative ef- fects, the sentencing changes are re- ported to involve no more than doubling the length of prison terms for high-rate robbers (Greenwood, 1982:791. In reanal- yses of the original data, however, a pol- icy of doubling time served to about 8

132 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE 5-2 Alternative Estimates of Crime Reduction and Prison Populations Projected for Highly Selective Incapacitation Policy Applied to California Robbery Inmates Percent Reduction Percent Change in Source in Adult Robberies Robbery Inmates No Career Termination Greenwood (1982:xiv, 79) 20 0 Greenwood (1982:xix, 85) 15 -5 Cohen (1983:Figure 2; Visher, 13 -8 Volume II:Figure 3) Visher (Volume II:Figure 3) 13 +1 With Career Termination Technical Note TR = Ha 14 - 1 TV, = 15 years 10 +17 To = 10 years 9 +23 To = 5 years 6 +38 NOTE: Alternative outcomes reported in Greenwood (1982) are for selective policies of increased prison teens for high-rate robbery inmates and 1-year jail teams for all other robbery inmates; the effect varies with the length of prison terms for high-rate robbery inmates. All other estimates in this table are based on a policy of sending high-rate robbers to prison for just over 8 years, while all other robbery inmates are sent to jail for 1 year. aAssuming that residual career length is infinite is analogous to ignoring the effect of career termination. Slight differences in the methods used to calculate the crime reduction and prison population effects explain the differences in these results compared with those reported by Visher (Volume II:Figure 31. years by inmates classified as high-rate offenders, while reducing time served for all over inmates to 1 year, was estimated to reduce crime by 13 percent for Califor- nia robbers, with an 8 percent decrease in the number of robbery inmates (Cohen, 1983:Figure 2; Visher, Volume II:Figure 31. Thus, with the assumptions and pa- rameter values employed by Greenwood, the 15 and 20 percent reductions in rob- beries that he reports cannot be achieved without making sentences even more dis- parate than a ratio of g to 1. The estimated incapacitative effects are potentially sensitive to the parameter val- ues substituted into the equations de- fining the mode} (see the technical note at the end of the chapter). The distribution of A computed in the Visher reanalysis (Volume II:Table 9) reflected both a lower mean value of A anc! somewhat less skewness. In the incapacitation model, however, lower values of A imply higher projected numbers of active offenders and similar reductions in crime. Substi- tuting her revised parameters into the incapacitation model used by Green- wood, Visher estimates that under Policy 6, doubling prison terms for inmates cIas- sified as high-rate robbers would result in a 13-14 percent reduction in adult rob- beries, with the number of robbery in- mates essentially unchanged (Volume II:Figure 31. Visher reports that achieving a 20 percent reduction in robberies in California would require tripling the time  served in prison by robbers classified as high-rate offenders by the scale, which would lead to an estimated 5 percent increase in robbery inmates. To assess the robustness of the esti- mates of incapacitative effects, an analysis of the same classification scale and a sim- ilar sentencing policy was done for the

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE Michigan inmates. However, the inmates sampled from Michigan prison and jails apparently were very different from those in the California sample. In her reanalysis of the inmate data, Visher (Volume II) reports that Michigan inmates had lower scores than Califomia inmates on the sev- en-point classification scale, largely be- cause fewer Michigan inmates reported serious drug use or juvenile incarcera- tions. Therefore, while 48 percent of the California inmates convicted of robbery were classified as high-rate offenders by the scale, only 15 percent of robbery in- mates in Michigan were so classified. According to several accuracy measures, the scale was much less accurate in dis- tinguishing high-rate inmates from other inmates in Michigan than in California (see Visher, Volume II:Table 171. Finally, the sentencing policies for convicted rob- bers in Michigan and California are ap- parently quite different: Michigan nearly always incarcerates convicted robbers for prison terms averaging 5 years, while Cal- ifornia sentences 50 percent of these of- fenders to jail for terms of no more than 1 year (Greenwood, 1982:Table B.21. These differences between California's and Michigan's inmate characteristics and existing sentencing policies are likely to substantially affect estimated incapacitative effects. Visher (Volume II) showed that using the selective sentenc- ing policy defined earlier (doubling prison terms for inmates classified as high-rate offenders and giving 1-year jai! terms to all others) would increase rob- beries by adults in Michigan by an esti- mated 30 percent while the number of robbery inmates would decline by almost 50 percent. The substantial increase in crime in Michigan would be clue to the large reduction in sentences for the many predicted medium- and low-rate offend- ers; they would serve only 1 year in jai! under the policy rather than the average of 4 or 5 years in prison under current ]33 policy. In addition, few inmates in Mich- igan would be classified as high-rate of- fenders and thus subject to long prison terms -using the same classification scale. The analysis of Michigan inmates illustrates that incapacitative effects of any policy are likely to vary dramatically across jurisdictions because of variations in the predictive accuracy of the cIassifi- cation scale and in existing sentencing . . pot .lcles. Recently, Cohen (1984a) pointed out that the projections by Greenwooc! and  Visher (based on Equations 5-1'. 5-2'. , , 5-9, and 5-10 in the technical note) as- sume career length to be much longer than time served. The long incarceration times of ~ years or more analyzed in Greenwood's policies make that approxi- mation particularly inappropriate and warrant including career length in the analysis to reflect termination of some careers during these long periods of in- carceration. Using Equations 5-1, 5-2, 5-3, and 5-5 of the technical note with Visher's parameter values, Cohen reesti- mated the effects on crime and prison population, assuming that residual ca- reers average 5, 10, and 15 years (see Table 5-21. Properly accounting for career termination during periods of incarcera- tion decreases the estimate of crime re- duction through incapacitation because only a portion of total time served by inmates reduces time free for active of- fenders; some portion of incarceration time is served after careers have termi- nated. The magnitude of the decrease is larger when time served is Tong or resid- ual career lengths are short, because ei- ther increases the likelihood that careers te'-~inate while offenders are incarcer- ated (see Table 5-31. In the illustrative policy considered here, time served in prison would be doubled to just over 8 years for high-rate robbers. If residual careers for robbers averaged 15 years, then 35 percent of

134 TABLE 5-3 Percent of Prison Inmates Who Terminate Careers While Incarcerated Mean Time Served, S (years) Residual Career Length, TR (years) 5 10 15 1 2 s 8 16.7 9.1 6.3 28.6 16.7 11.8 50.0 33.3 25.0 61.5 44.4 34.8 NOTE: The likelihood that careers terminate during incarceration is given by S/(S + TR)- (See Equation 5-8 in the technical note at the end of this chapter.) inmates serving average 8-year terms would be expected to end their careers before their release. This would lower the estimated crime reduction effect for such a policy to 10 percent. When re- maining careers are assumed to be even shorter, the reduction in incapacitative effect is even more substantial~own to a 6 percent reduction in robberies for residual careers that average 5 years (see Table 5-2, above). Accounting for career termination in Greenwood s incapacitation mode! has an opposite effect on the estimates of inmate populations under alternative incarcera- tion policies. An underlying assumption of this steady-state mode! is an active offender population of constant size N. which can be determined from knowI- edge of A and the total crime rate, C. A shorter career length implies that a larger fraction of offenders end their careers each year; these ax-offenders are as- sumed to be replaced by new offenders to maintain the value of N. The new of- fenders, in turn, become vulnerable to incarceration under the new policy and eventually add to prison population and serve a larger share of this incarceration time after their careers have terminated. Accounting for these termination and re- placement processes results in estimates of the inmate population that are larger . . , CRIMINAL CAREERS AND CAREER CRIMINALS than those that fail to adjust for these processes. If the remaining careers of CaTifomia robbers average 15 years, the impact of doubling time served by high-rate rob- bers, accounting for termination and re-  placement, results in an estimated 17 per- cent increase in robbery inmates (Table 5-2, above). If remaining careers are only 5 years, the impact of accounting for ter- mination and replacement is even more substantial, resulting in a 38 percent in- crease in the prison population of robbers (see the technical note at the end of this chapter). These prison population esti- mates would represent 4 and 10 percent increases, respectively, in the total in- mate population (for all offenses) in Cali- fornia.~5 Table 5-2 summarizes the various esti- mates of the effects in California of a highly selective incapacitation policy of doubling prison terms to 8 years for in- mates classified as high-rate robbers while imposing 1-year jail terms on all other robbery inmates. Depending on as- sumptions about career length, the policy is estimated to reduce the number of robberies by adults by 6 to 14 percent, with effects on robbery inmates ranging from essentially no change to a 38 percent increase. These alternative crime-reduc- tion estimates are all smaller than the 15 to 20 percent decreases in adult robberies estimated by Greenwood. The impact of career length on the estimates is substantial. Therefore, fur- ther research estimating the length of criminal careers, and especially the way residual career length varies during crim i5Convicted robbers represented approximately one-quarter of Me total prison population in Califor- nia in 1980 (California Department of Corrections, 1980~. A 17 percent increase in robbery inmates would thus represent a 4.25 percent increase in total prison population. Likewise, a 38 percent increase in robbery inmates would increase total population by 9.5 percent.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE inal careers, is very important for evaluat- ing alternative incapacitation sentencing policies. While Tong incarceration terms may be clesirable for deterrent or retribu- tive reasons, they make less efficient use of prison resources from an incapacitative standpoint. For any sentencing policy, the tradeoffs between the benefits of incapacitative crime reduction ant! the costs of prison population are most favor- able when time server] is short ant! resicI- ual career length is Tong. Basec! on the estimates in Table 5-2, moclest crime reduction is possible from a selective incapacitation policy applier! to California robbers. These estimates, how- ever, are themselves optimistic; in actual use, several factors wouIc! diminish the crime reduction achieved: · The incapacitative effect will be less if the offenses of incarcerates! inmates persist in the community, perhaps be- cause the inmate is replacer! by a new recruit or because incarceration of some members of offending groups floes not disrupt the groups' crimes. · The predictive power of the scale I35 suggests that the system may adapt to precliction-basec] clecision rules in ways that recluce the consiclerable sentence disparity of 8 years for preclictec] high-rate robbers ant] 1-year terms for all others. Because the magnitudes of these effects conic] be measurer] only following imple- mentation of the policies, they are un- known at present ant! can be expecter! to vary by crime type. However, because the effects wouIc! all operate to recluce incapacitative effectiveness, the prececI- ing estimates of the effects on crime ant] on prison population shouIc! be treater! as best-case estimates. Parole Release Basect on Offender Characteristics: The Salient Factor Score Perhaps the most common use of cliffer- ential treatment for purposes of selective incapacitation is fount! at parole, when ., , ,. . · · 1 community prorecuon IS a mayor consla- eration in early-release decisions. In the fecleral parole system, this consideration has involves! the use of an explicit preclic- tion scale known as the Salient Factor Score as an air! in assessing the risk of future crimes of cancliciates for parole re- lease. Relying on factors reflecting an inmate's age, prior criminal record. ant! . . ~.  user] to distinguish offenders may climin- ish as the scale clevelopec! on inmates is applier! to a broacler and potentially clif- ferent population of all convictec] offencI ers, especially in a different jurisdiction . ~. .. . . .~ . . . . . history of drug use, an 1ncilvlclual s Sa Wlt ~ ~ luerent existing crimma Justice c e- . . . . . Ilent Factor Score IS computed; on the clslon practices. basis of that score parole cancliciates are · Self-reports of predictor variables . .'. . .l1 1 1l ~ 1' 1 .l1 1 assigner! to a reclcilvlsm risk category. W111 provably not oe usable ana W111 DrOD ably be clistortec! if they are used, ant! official records of those variables may have less predictive power because of recorc] inaccuracies or gaps. · The offending rates of Tow-rate of fenders may increase in response to the shorter sentences imposer! on those incli vicluals uncler a selective policy because of a clecreasec] deterrent effect. · Finally, experience with efforts to re cluce criminal justice system discretion _ These risk levels vary from a 49 percent chance of recidivism a return to prison or outstanding warrant in the first 2 years after release for poor-risk inmates to a 12 percent chance of recidivism for very- goocI-risk inmates (Hoffinan, 19831. When a broader recidivism criterion of rearrest in the first 3 years after release is used, Tithe scale factors and predictive accuracy of the federal Salient Factor Score are discussed in Chap- ter 6.

136 the risk levels are a 66 percent chance of recidivism for poor risks and 22 percent for very good risks (Janus, 19851. While considerable attention has been given to evaluating the predictive ade- quacy of the Salient Factor Score in dis- tinguishing among different recidivism risk Velvets for federal inmates (see Chap- ter 6), the selective incapacitation effects from using the scale have been assessed only recently. Janus (1985) examined a 50 percent sample of releasees from federal institutions during the first half of 1978. To control for the impact of offense sever- ity on time served, only inmates who scored "very high" in the "severity of offense behavior scale" were used.~7 Ta- ble 54 summarizes the differences in time served and the impact of these dif- ferences on arrests prevented for the clif- ferent recidivism risk levels. The data show that the time served does vary with differences in the risk of future crimes: poor-risk offenders serve 10 months more than the average term for offenders, and very-good-risk offenders serve 3 months less than the average. These differences in time served need not arise solely from differences in parole release decisions; they might also reflect differences in the length of sentences imposed in court. The different risk levels also display dif- ferent arrest rates during the follow-up period. The monthly arrest rate of poor- risk offenders (.035) is 3.5 times that of very-good-risk offenders (.0101. Janus (1985:122) notes that even this difference across risk levels is probably understated because of likely differences in exposure during the follow-up for the different risk levels; poor-risk offenders with their higher number of arrests-are more likely i7This offense category includes robbery; break- ing and entering or burglary of armory, residence, or involving hostile confrontation with victim; coun- terfeiting currency; drug sales; extortion; explo- sives; and property offenses involving more than $100,000 but not more than $500,000 (U.S. Parole Commission, 1981~. CRIMINAL CAREERS AND CAREER CRIMINALS to be reincarcerated and thus to be on the street for less time during the 3-year fol- low-up period. An adjustment for differ- ential exposure time would therefore fur- ther increase the monthly arrest rate of the high-risk group compared with that of other risk levels. The monthly arrest rates can be used to assess the increment to arrests prevented as a result of the differences in the time served by inmates in the different risk levels (Table 541. Janus reports that,  compared with the average time served for the 620 inmates examined, the 10 months' longer time served by poor-risk offenders prevents 20.03 arrests. Also, be- cause poor-risk inmates serve longer than the average term of 26.9 months and very good risks serve less than the average term, a total of 19.63 arrests are prevented for the 620 inmates examined. Thus, the differential terms result in a net 7 percent increase in the number of arrests pre- vented with no change in prison popula- tion. One could also consider the maximum possible increment in crime reduction by using a different baseline for time served. The time served by the very good risks might be regarded as the lower end ofthe acceptable range of sentences for inmates convicted of very-high-severity offenses. The longer time served by each of the higher-risk groups would have selective incapacitative effects. Because offenders in higher-risk categories serve longer than those characterized as very good risks (Table 54), an additional 52.17 ar- rests are prevented, but there is also a 13 percent increase in the time served. Thus, an increment of 37.1 inmate- months in prison is required for each aci~i~onai arrest prevented. [anus's data also permit examining the degree to which the Salient Factor Score distinguishes among offenders in terms of individual arrest rates for active offenders or in terms of the continued participation by active offenders observed during the

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE 137 TABLE 5-4 Selective Incapacitative Effects of Federal Parole Release Decisions Using the Salient Factor Score for "Very High Severity" Offenders . . ~ ~. Salient Factor Score by Risk Level . Total or Very Average, Poor FairGood Good All Risk Variable 0-3 4-5 6-7 8-10 Levels - Number of Inmates (N) 59 102 136 323 620 Current Time Served Average time served at 1978 release, in months (S) Arrests per inmate per month in a 3-year follow-upa (it) Number of arrests prevented by current time served (I = N x S x ~) All Serve Average Time Served Number of arrests prevented if all served average time served of 26.9 months (I* = N x 26.9 x ,u) Increment (decrement) to arrests prevented by current time served versus average time served (I - l*) Percent change in arrests prevented by current time served versus average time served [~1 - I*)tl*] 36 Increment (decrement) to time served from current time served versus average time served 36.6 28.7 .035 .027 28.8 .018 23.8 .010 75.58 79.04 70.50 76.87 26.9 .026 301.99 55.55 74.08 65.85 86.89 282.37 20.03 4.96 4.65 7 7 (10.01) 19.63 (11) 7 [P* = (S - 26.9) x N] 572.3 183.6 258.4 (1,014.3) 0.0 Percent change in time served from current time served versus average time served [(P* - 26.9 x N)/26.9 x N] 36 7 7 (11) 0 All Serve "Very Good Risk" Time Served Number of arrests prevented if all served "very good risk" time of 23.8 months (1' = N x 23.8 x ,u) Increment to arrests prevented by current time served versus "very good risk" time served (I - 1') Percent change in arrests prevented by current time served versus "very good risk" time served [(1 - 1')11'] Increment to time served from current time served versus "very good risk" time served [P' = (S - 23.8) x N; Percent change in time served from current time served versus "very good risk" time served [(P' - 23.8 x N)123.8 x N] 49.15 65.54 58.26 76.87 249.82 26.43 13.50 12.24 0.0 52.17  54 21 21 0.0 21 755.2 499.8 680.0 0.0 1,935.0 54 21 21 0.0 13 NOTE: Offense types in the "very high severity" category include robbery; breaking and entering or burglary of armory, residence, or involving hostile confrontation with victim; counterfeiting currency; drug sales; extortion; explosives; and property offenses involving more than $100,000 but not more than $500,000 (U.S. Parole Commission, 1981~. aAssumes that all inmates were free in the community and still active in criminal careers for the entire 36-month follow-up. Time served as a result of subsequent commitments or career termination will increase the average monthly arrest rate reported here. SOURCE: Derived from data presented in Janus (198S:Tables 1 and 3~. follow-up. Table 5-5 compares the entire sample in terms of various aspects of individual criminal careers for the dif- ferent risk-level groups identified by the Salient Factor Score. The aggregate monthly arrest rate estimates provided by Janus combine offenders who are not rearrested after release and those who are

138 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE 5-5 Variations in Criminal Careers for Different Risk Levels Identified by the Salient Factor Score Salient Factor Score by Risk Level Variable Poor Fair 0-3 4-5 Number of released inmates 493 495 476 Total arrests in 3-year follow-up 757 638 377 Average arrests per month per offender (aggregate rates) .043 .036 .022 Percent of offenders with at least one arrest in 3-year follow-up 65.5 57.4 41.4 Average time to first arrest for those arrested (months) 12.62 13.66 14.79 Average arrests per month free for active offenders with at least one arrest in 3-year follow-upa Average arrests per month free for all active offenders (correct ed rated Percent of all active offenders expected to have at least one arrest in 3-year follow-ups Percent of offenders who remain active during 3-year follow up~ Very Good Good 6-7 8-10 722 285 .011 22.0 16.49 .079 .073 .073 .066 .060 93 91 88 70.4 63.1 47.0 .068 .061 .05 84 26.2 Total or Average, All Risk Levels - 2,186 2,057 .026 44.1 14.01 .071 .064 90 49.0 aThis arrest rate is estimated from the reciprocal of the time to the first arrest for offenders who are arrested during the follow-up (see text). bThe corrected rate is estimated by adjusting downward the rate for active offenders who are arrested in the follow-up; see Chapter 3 and Cohen (Appendix B) for the estimation procedure used. CFor a follow-up t months long and individual arrest rate, ,u, the probability that an individual who  remains active during the follow-up will be arrested is estimated as 1 - e-~. The total fraction of active offenders among releasees is estimated by dividing the percent of offenders with at least one arrest in the follow-up period by the percent of all active offenders expected to be arrested in that period. SOURCE: Derived from data presented in Janus (1985:Tables 1 and 3~. rearrested! in the 3-year follow-up. These aggregate rates vary substantially across risk levels, from a high of .043 arrest per month for poor-risk offenders to a low of .011 for very-good-risk offenders. The dif- ference in these rates, however, is largely due to differences in arrest participation during the follow-up period: poor risks are much more likely to be rearrested (66 percent) than are very good risks (22 per- cent). The average time to the first arrest in the follow-up period (for those ever ar- rested) provides a basis for estimating the individual arrest rate for the demonstra- bly active offenders. Since releasees are on the street until their first arrest after release, the rate calculated by using those intervals provides an estimate of the ar rest rate while free for offenders who remain active. However, this estimate is biased upward somewhat by the Toss of observations of arrests for offenders who remain active but whose first rearrest would occur after 3 years. The rate esti- mated only for the demonstrably active offenders is thus conditional on having at least one arrest during the follow-up pe- riod. The rate corrected for this bias re- ported in Table 5-5 ranges from .073 ar- rest per month free for poor risks to .051 arrest per month free for very good risks. i8For it*, the biased frequency rate, the corrected frequency, I, is estimated from the relationship: ,u* = ,uI(l - end; see Chapter 3 and Cohen, Appendix B. for the details of this adjustment

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE This estimate of average arrest rates while free for all active offenders pro- vides a basis for estimating the total size of the active offender population during the follow-up, including active offenders who are not yet arrested. For individual arrest rate, A, the probability that an indi- vidual who remains active during a fol- low-up period of length t will be arrested can be estimated as 1 - emu; see Table 5-5. During a follow-up as long as 3 years, very few active offenders will fail to be arrested. Even among the very good risks, who have the lowest indiviclual arrest rate, 84 percent of all active offenders can be expected to be arrested at least once in a 3-year follow-up period. Applying this 84 percent estimate to the 22 percent of offenders who are rearrested, total active offenders are 26.2 percent of the very good risks. Similarly, 70.4 percent of poor risks are estimated to remain active throughout the 3 years. The percentage of poor risks who are arrested in the 3-year follow-up period (65.5) is 198 percent higher than for very good risks (22.0 per- cent), and there are an estimated 169 percent more total active offenders in the poor-risk group. In contrast, the corrected individual monthly arrest rate (,u) for ac- tive offenders is only 43 percent higher. The Salient Factor Score is thus more effective at distinguishing subsequent participation as measured by arrests than at distinguishing arrest frequency. Charge-Based Sentencing Policies Partly in response to the ethical con- cems associated with some forms of se- lective incapacitation, Cohen (1983, 1984a, b) has explored the degree of crime control that could be achieved by targeting incapacitation on offenders de- fined solely in terns of their convicted offenses, and perhaps prior record. This type of policy could avoid the use of more ethically controversial variables to clishn ]39 guish among offenders convicted! of the same offense type. A charge-based inca- pacitation policy would augment current . . . . . . sentencing practices ay assigning m1nl- mum sentences uniformly to all offenders convicted of the same charge who have similar prior records (Cohen, 1984b). Un- der such a policy, the sentences of all offenders convicted of the targeted charge would receive at least the same minimum term of incarceration, while sentences already above that minimum would remain unchanged. The minimum sentence imposed would be selected  from within the range of acceptable pen- alhes for the targeted offense type; whether the sentence is at the high or low end of that range would depend on the predictecI subsequent criminal activity associated with that offense type. While the offender-based selective incapacita- tion policies examined above have fo- cusec! on differences among offenders, charge-based incapacitation policies ex- tend incarceration uniformly to all of- fenders convicted ofthe same offense and focus on differences between offenses. The basis for this approach to incapac- itation is recent evidence from several different studies that important differ- ences may exist in indiviclual frequencies of offending for offenders charged with different offense types. Chaiken and Chaiken (1982a, 1984), for example, found that "violent predators" inmates who reported committing the combina- tion of robbery, aggravated assault, and drug dealing-committed not only those offenses but also burglary at the highest rates of all survey respondents. Cohen (1984b) has estimated the po- tential incapacitahive effects that could be achieved by targeting increased incarcer- ation on offenders convicted of specific offenses. Her data show that charge- base(l incapacitation policies directed at different offense types would differen- tially reduce crime levels and increase

140 prison populations. For example, in Washington, D.C., in 1973, the index crime and sentence offering the best tradeoff between crime reduction and prison impact was minimum 2-year prison teas for defendants convicted of robbery: robberies committed by adults would have decreased by 8 percent and adult "safety crimes" by 5 percent, while increasing the total prison population by 17 percent.~9 The same 2-year minimum prison term imposed on all burglary con- victions would have reduced burglaries committed by adults by 6 percent and adult safety crimes by 3 percent, while increasing the total prison population by 25 percent. A likely judicial response to this type of charge-based sentencing policy might be to impose the mandatory minimum 2- year prison terms only on offenders with previous convictions for the targeted of- fense. Restricting prison terms to repeat convictions, however, would seriously re- duce the expected crime reduction. Even extremely long sentences for persons pre- viously convicted of robbery or burglary would have reduced those offenses by only 1 and 4 percent, respectively. The reason for this low incapacitative effect is that only about 25 percent of convicted robbers and burglars have previous con- victions for those offenses. Many clefen- dants were never previously arrested for the target offense, and among those who were, robbery and burglary charges were oRen reduced or dropped during plea bargaining. The effects of charge-based policies are estimated from inclividual arrest histories. Those estimates are most accurate for the portion of arrests that might have been averted through alternative sentencing policies imposed at previous convictions. Extrapolating from arrests to crimes de 19"Safety crimes" are murder, rape, robbery, and burglary (Avi-Itzhak arid Shinnar, 1973~. CRIMINAL CAREERS AND CAREER CRIMINALS pends on assumptions of arrest probabil- ities across different offender subpopula- tions: those arrested in a sampling year are assumed to have the same arrest risk per crime as those not arrested that year; and among arrestees, those with arrests averted are assumed to have the same arrest risk per crime as those without arrests averted. Violations of these as- sumptions will lead to biases in the esti- mates of crime reduction from charge- based policies: if offenders with prior records are more likely to be arrested for Weir future crimes (perhaps because they  are better known to the police), then the crime reduction from charge-basec! inca- pacitation is overestimatecl; the opposite will occur if prior records lead to greater adeptness in avoiding arrest. Adequately addressing these potential sources of bias requires empirical research into the vari- ations in arrest risk per crime during in- dividual criminal careers. In addition to empirical concerns about the accuracy of the estimates, charge- based incapacitation policies raise their own ethical concerns. The basic strategy of ignoring differences among offenders convicted of the same offense is intended to avoid some of the ethical problems raised by differential sentences based on individual attributes. But charge-based sentencing policies impose a minimum sentence on all offenders convicted of the same offense and ignore variations in the blameworthiness of criminal conduct fall- ing within the same convicted offense type. While sentences above the mini- mum prison terms are possible to accom- modate aggravating circumstances, the mandatory minimum sentence represents a floor on the length of prison terms. Mitigations both in the harm done and in the culpability of the offender that are not adequately captured by variations in the convicted offense type will not be possi- ble under a uniform mandatory minimum sentence policy.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE Operational Issues in Assessing Alternative Selective Incapacitation Policies Analysis of alternative selective inca- pacitation policies raises several opera- tional issues. One of these relates to the choice of the policy variable that is used in the assessment. Throughout the dis- cussion so far, the costs and benefits of selective incapacitation have been as- sessecl by seeking the largest crime re- duction while minimizing the increase in prison population. Operationally, how- ever, one must consider such constraints as prison capacity and limits on the length of time to be served for any particular offense emerging from just-deserts con cerns. The effects associated with such con- straints can be illustrated with data from Cohen (1984b) assessing the effects of various charge-based incapacitation poli- cies applied to arrestees in Michigan. If the increase in the total prison population is constrained to a maximum of 20 per- cent, the alternative policies of targeting charge-based incapacitation on robbery, burglary, or auto theft convictions all emerge as best and are indistinguishable in their effects. Targeting any one ofthese offense types for mandatory minimum terms is estimated to result in a 12 to 13 percent reduction in the selected target offense and a 20 percent increase in the total prison population. These effects, however, result from different sentences for each target offense: 5-year minimum terms for either robbery or auto theft and 2-year terms for burglary. A different best policy emerges w. hen ~ _ ~ ~ O Me prison population constraint is dropped and a constraint on sentence length is considered. If the mandatory minimum terms are constrained to be no more than 2 years, for example, targeting charge-based incapacitation on robbery convictions emerges as best, requiring 14' the smallest increase in total prison pop- ulation (2 percent) to achieve the largest reduction in the targeted offense, an 8 percent reduction in robberies commit- ted by adults. In comparing alternative constraints on the minimum sentence, Cohen also notes that shorter prison terms are generally more attractive op- tions because they involve the lowest marginal costs in terms of prison popula- tion increases for each estimated 1 per- cent reduction in crime. Offenders are often diverse in their offending, engaging in several different  offense types during their criminal ca- reers. Thus, another consideration in choosing among alternative selective in- capacitation policies might be the impact of a policy in preventing crimes other than the offense targeted for minimum prison terms. With a 20 percent constraint on increases in total prison population and considering reductions in safety crimes by adults, targeting charge-based incapacitation on burglary emerges as the best policy: a 20 percent increase in total prison population from 2-year minimum terms for burglary convictions results in an 11 percent reduction in adult safety crimes. In comparison, 5-year minimum prison terms for robbery convictions re- sult in only a 6 percent reduction in total safety crimes by adults with the same 20 percent increase in total prison population. The strategy of targeting particular of- fense types for longer prison terms also raises other operational considerations. In the analysis of charge-based incapaci- tation policies, We impact of alternative policies varied across jurisdictions. This makes it especially important that any policy analysis not be applied indiscrimi- nantly to other sites. In Washington, D.C., robbery emerged as a clearly dom- inant policy target because it offers the most favorable tradeoff between large crime reduction and small prison popula

42 40 Q ~5 O ~"~.m 30 u' ~ ~ ._ at. ~ a) ~ m a) ~ t'' llJ Q Cat o o 20 4- ~ ~ Q a, O ~ 4- U' _ ._ I Q ~ O O 4 - ~ ._ .O ~ Cat ~ o ~ t~ - a, LL ~ ~ CRlMlNAL CAREERS AND CAREER CRIMINALS Detroit white Detroit black Southern Michigan white Southern Michigan black 1 _ _ _ / ok/' ark' /' ok' Ok:, ~ I I I I 0 2 4 6 .K~ ,~ ok' 8 10 Minimum Prison Terms, S (years) White (AJS = .089) a/' (AJS=.082)  Black (US = .1 00) (AS = .108) FIGURE 5-1 Variations in incapacitative effects for different population subgroups from uniform minimum prison terms imposed for burglary in Michigan. Source: Derived from data provided in Cohen (1984b). tion increases in that jurisdiction. There was more ambiguity in the results for Michigan: if the increase in total prison population is constrained to 20 percent, burglary offers the largest reduction in safety crimes committed by adults; if sen- tence length is constrained to 2-year min- imum prison terms, robbery is a better target offense. These variations in dif- ferent settings and with different policy constraints make it essential that jurisdic- tion-specific analyses be conducted when considering alternative selective incapac- itation policies. Charge-based incapacitation policies are consciously designed to avoid the use of extra-legal predictor variables, but they may have a larger impact on some groups than on others. In particular, any group that is disproportionately involved in the target crime would be so affected. For example, Figure 5-1 shows Cohen's (1984b) estimate of the percentage in- crease in prison population for each 1 percent reduction in adult burglaries for different race subgroups and jurisdictions in Michigan; the largest prison popula- tion increases occur among white offend- ers. These differences in impact reflect differences in individual frequency rates, A, and in the prevailing levels of expected time served per crime, [S. for different subgroups: in general, the higher AJS is for any group under prevailing policy, the Tower wflT be the increases in prison pop- ulation resulting from aTtemative selec- tive incapacitation policies. Summary: Incapacitation Strategies While incarceration leads to some re- duction in crime through incapacitation, none of the various incapacitation strate- gies examined provide dramatic reduc

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE lions. Even interventions that have only modest impacts on crime can involve siz- able costs, especially when they are ap- plied broadly to large target populations. In the case of collective incapacitation, for example, general increases in prison use to achieve a 10 percent reduction in crime may require more than doubling existing inmate populations, ant! even those estimates depend on the crimes of incarcerated offenders not being replaced or continued by other offenders still in the community. The estimates also must be viewed as preliminary. They involve estimates of A that may not be applicable to the particu- lar population that would constitute the new prisoners. The estimates also ignore any effects that incarceration might have on the future criminal careers of the in- carcerated offenders; it is possible that it might lengthen or shorten those careers. Improving the estimates of incapacitative effects requires continued research char- acterizing individual criminal careers, particularly efforts to identify variations in those careers and in their interaction with the criminal justice system. Further efforts should also be directed at assess- ing the influence of those variations on estimated incapacitative effects. Because of the wide variations in of- fending rates, with only a small number of offenders committing serious crimes at high rates and having extended careers, the potential exists to achieve similar re- ductions in crime by focusing interven- tions more narrowly on smaller high-risk populations, i.e., by selective incapacita- tion strategies. This cost advantage was illustrated in the analyses of selective incapacitation. For example, it is esti- mated that selective incapacitation poli- cies could achieve 5 to 10 percent recluc- tions in robbery with 10 to 20 percent increases in the population of robbers in prison, while much larger increases in prison populations are associated with collective incapacitation policies. To ]43 achieve even these small reductions in crime, however, targeted interventions require effective classifications to identify those offenders at highest risk of contin- ued serious offending. The current status ant] future prospects of classification in criminal justice decisions are the subjects of the next chapter. TECHNICAL NOTE This note explores some technical de- taiTs in estimating incapacitative effects. The first section presents the details of  the general incapacitation moclel. The next section applies the mode! to estimate the crime reduction and prison popula- tion effects of a collective incapacitation policy of general increases in the use of incarceration from 1981 levels. The final section applies the mode! to estimate the effects associates] wit-in a selective inca- pacitation policy of differential sentences applied to inmates convicted of robbery in California. A Mode} of Incapacitation The crime control and prison popula- tion effects estimated by Greenwood (1982) and others arise from a model of individual criminal careers derived by Avi-Itzhak and Shinnar (1973) and Shin- nar and Shinnar (1975), and summarized by Cohen (1978, 19831. In the steacly-state moclel, all crimes are assumed to be com- mitted by a fixed population of N individ- ual offenders who, if left free, would com- mit crimes at an average rate A per year throughout the time they remain active in their criminal careers, i.e., their mean residual career length, TR 20 The effects of the criminal justice system are described in terms of I, the probability that an of 20The population of active offenders remains sta- ble in size as offenders who end their criminal careers are replaced by new offenders with identical criminal career attributes.

144 Sense will be followed by arrest, convic- tion, and incarceration, and S. the average time served by those incarcerated. Under these conditions, active offenders are in- capacitated during a fraction of their crim- inal careers, I, while during a fraction (1 - I), active offenders are free to commit crimes on the street. It can be shown (Avi-Itzhak and Shinnar, 1973:Appendix; Shinnar and Shinnar, 1975:Appendix) that the incapacitated fraction of careers, I, is given by AJS(T S ~ ~ J . (5-1 ) 1 + AJS( ) The fraction of careers that active offend- ers are not incapacitated, i.e., when they are free on the street, is given by 1 - 1 = T 1 + AJS T,{ + S In both Equations 5-1 and 5-2, the frac- tion TR/(TR + S), which takes account of the reduction in crimes avoided because of spontaneous career termination during incarceration periods, S. reflects the like- lihood that an offender is still active in a career after serving a sentence; the re- maining fraction, S/(TR + S), is the likeli- hood that a career terminates while a sentence is being served. 21 2iThe expression for the likelihood that a career terminates while a sentence is being served de- pends on how career lengths and sentence lengths are distributed among offenders. The result re- ported here applies when career length is distrib- uted exponentially with mean residual length TR and when sentence length is also distributed expo- nentially with mean length S. Such distributions are generally compatible with observed patterns of ca- reer termination and with the distribution of time served by offenders. CRIMINAL CAREERS AND CAREER CRIMINALS Equations 5-1 and 5-2 can be used to estimate the annual numbers of crimes committed by active offenders who are free and of crimes averted for active of- fenders who are incarcerated. Crimes committed, C, are calculated as C = N X A X (1 - 1) , (5-3) while crimes avoided through incapacita- tion, CA, are calculated as CA = N X A X 1. (5 4) Combining Equations 5-3 and 5 4 yields the total number of potential crimes. The reduction in crimes through incapacita-  tion, CA/(C + CA) is equal to I in equation 5-1. Likewise, Equations 5-1 and 5-2 can be used to estimate the size of the annual inmate population, R. The average daily inmate population in a year (reflected in person-years in prison)22 is given by (5-2) R = NO AX (~1-1) HIS. (5-5) Some careers terminate during prison terms, and so not all inmates reflected in R are active offenders. When focusing on a single offender, the expected time dur- ing that offender's residual career that he spends incarcerated is given by rA = TR X 1. (5-6) Similarly, in any year, the number of active offenders found in an annual in- mate population is given by RA = N X 1. (5-7) When Equations 5-5 and 5-7 are com- bined, the fraction of total inmates (R) who would be active if free (RA) is given by 22Each unit of R is equivalent to one inmate incarcerated for a full year. Use of average daily population allows for population turnover during a year as some inmates are released and new commit- ments take their place in the inmate population.

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE RAIR= NXI = TR N X A X (1 - 1) X JS TR + S While the total number of inmates (R) is easily observed, the total number of active offenders (N) both in ant! out of prison cannot be observed directly. The size of the active offender population, however, can be estimated from the in- mate population. First, by applying the fraction in Equation 5-8 to the total num- ber of inmates in a year, the number of active offenders among inmates (RA) can be estimated. Then, with an estimate of I in Equation 5-7, the total number of ac- tive offenders, N. can be estimated. When TR is much larger than S. spon- taneous career terminations are less likely, and the portion of a sentence dur- ing which an offender would be inactive becomes negligibly small. For reasons of analytical convenience and because reli- able estimates of TR have only recently become available, researchers have ig- nored career termination among inmates. With this approximation, the fractions of careers spent incapacitated and free are given by the simpler formulas: AJS 1 + AJS and 1 (5-1') 1 1 - 1* =, , ~ ' 1 - 1. (5-2') 1 + AJS Ignoring career termination, the annual count of crimes committed is given by the product of the offender population, the indiviclual offending rate, and the portion of a year offenders are free to commit crimes: C*=NxAx(l-I*~. (5-9) This value of C*, which is less than or equal to C and which fails to take account 145 of spontaneous career terminations dur- ing incarceration, understates the number of crimes that are committed and so over- states the number of crimes that are avoided by incapacitation. Because of spontaneous career termination during incarcerations, only a portion of the total time served during a career can be counted as reducing crimes through inca- pacitation; the time spent incarcerated after a career ends has no incapacitative effect. Failure to exclude the time served after career termination thus overstates  the portion of a career that an offender spends incarcerated (I* 2 I), and under- states both the time he is free on the street t(1 - I*) _ (1-I)] and the estimate of the number of crimes he commits (C* c C). When time served (S) is longer relative to residual career length (TR), the under- estimate of crimes committee] is also larger. If there were no career terminations, the average daily incarcerated population would be the product of the number of crimes committed annually (C*) times the expected time served per crime (IS): R* = C* X JS = N x A x (1 - 1*) x [s. (5-10) The underestimate of crimes committee! when career termination is ignored leacis to a similar underestimate of the size of the inmate population, and the value of R* ' R. While the fraction of a career during which an active offender is incar- cerated is overestimated (1* 2 1), the total time spent incarcerated is underesti- matecl. This underestimation occurs be- cause as an offender commits more crimes during a career (C 2 C*), his vulnerability to incarceration also in- creases, but increasing portions of the incarceration time are server! after his career terminates. As time served (S) in- creases relative to resiclual career length (TR), the time spent incarcerated after the ens! of a career also increases, and the

146 underestimate of the average daily incar- cerated population gets larger (R* C R). When career termination is ignored, all inmates are assumed to remain active in their criminal careers, ant! PA* = R*. In this case, the total number of active of- fenders, N. is estimated from Equation 5-10, which reduces to R* = RAM = N x 1~. (5-10') Applying the Mode! to Estimate the Effects of a Collective Incapacitation Policy Equations 5-3 and 5-5 can be used to estimate the responsiveness of crime lev- els to changes in incarceration levels from incapacitation alone. The elasticity of crime (E), or the percentage change in crime accompanying a 1 percent change in the expected prison stay per crime US)' is given by dC JS d(JS) C -A2NTR2/(TR + S)2 = [1 + AJSTR/(Ta + S)]2 JS CRIMINAL CAREERS AND CAREER CRIMINALS change in the expected prison stay per . . crime IS given Dy dR JS d(lS) R AN[1 + AJS2TR/(TR + S)2] = [1 + AJSTR/(TR + S)]2 x_ JS ANJS 1 + AJSTRI(TR + S) 1 + AJS2TRI(TR + S)2 (5-12) 1 + AJSTRI(TR + S) Likewise, the elasticity of the annual in- mate population to changes in ~ or S alone is given by the same expression, since ER]S = ER] = ER,S. The elasticities in Equations 5-11 and 5-12 can be combined to form a cost/benefit ratio reflecting the percent- age change in the annual inmate popula- tion required to achieve a 1 percent change in the volume of crimes. This cost/benefit ratio is given by  ER]S 1 + AJS2TRI(TR + S) (5-13) X I- Ecus -AJSTR2/(TR + S)2 AN 1 + AJSTRI(TR + S) AJSTR2/(TR + S)2 (5-11) +AJSTRI(TR+S) The expression in Equation 5-11 also gives the elasticity of crime with respect to changes in J or S alone, since it can be shown that EC IS = Ec ~ = Ens. The elasticity of the total annual inmate population with respect to a 1 percent The expressions in Equations 5-11 to 5-13 differ significantly from those cle- rived by Cohen (1978:Appendix B), which ignored the role of career length. The importance of taking career length into account is examined here by upciat- ing the Cohen (1978:Appendix C) esti- mates of the relationship between inmate populations ant] crime reflection with new estimates that rely on more recent data from 1981 and that include the effect of career length. The upciated estimates rely on (lata on all new commitments from court to state prisons in 1981, me

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE clian time server] by first-time releasees from state prisons in 1981, and the num- ber of index offenses known to the police in 1981. In estimating the cost/benefit ratio in Equation 5-13, residual career length, TR, is set at 10 years remaining in a career. An adjusted value of five crimes per of- fencler per year is used for the individual crime rate, A. This corresponds to a value of A of about 20 crimes per year. How- ever, since the same crime may involve more than one offender, the rate for indi- viduals must be discounted by the aver- age number of offenders per crime. On the basis of data from the national victim- ization surveys, it is estimated that there are an average of two offenders per crime (Reiss, 1980b); this adjustment recluces A by half Also, since crime recluction is measured in terms of crimes known to police, the rate at which victims report crimes to the police (also estimated from national victimization surveys) is used to reduce the value of A by half again to reflect only the reduction in crimes that are known to police (Bureau of Justice Statistics, 19831. The updated data and estimates are presented in Table 5-6 and Figure 5-2. As was previously found (Cohen, 1978), the upciated cost/benefit ratios displayed in Figure 5-2 depend strongly on the value of expected time server! per crime, IS. If sanction levels are already high (say, IS above .05 for A = 5 and TR = 10), the additional costs in terms of the percent age increase in prison population re- quired to achieve a 1 percent reduction in crime are not unreasonable. Below IS = .05, however, the increases in prison pop- ulation associated with a 1 percent de- crease in crime are much larger. On the basis of 1981 data, only one jurisdiction, Washington, D.C., has a sanction level above IS = .05. The esti- mated prison increase to achieve a 1 per 147 cent reflection in index crimes known to the police in Washington is 3.7 percent when career length is ignored and 6.8 percent when career length is included. In contrast, several states (Hawaii, Mas- sachusetts, Montana, New Hampshire, New Jersey, New Mexico, Norm Dakota, Rhode IsTanc3, and Utah) have values of.7S less than .01, and they require increases in prison population of more than 20 per- cent to achieve only a 1 percent reduction in crime. The prison population increases will be larger if A and TR are smaller than assumed here and smaller for larger val- ues of A and TR.  Several large changes in.IS between 1970 and 1981 are also noteworthy. IS clecreased by at least 40 percent between 1970 and 1981 in seven states (Georgia, Maine, Mississippi, Montana, New Hampshire, New Mexico, and North Da- kota) and increased by a similar amount in two states (Hawaii and New York). The cost/benefit ratio increased between 1970 and 1981 in states where IS declined and clecreased in the states where IS in- creased. The differences in cost/benefit effects for different states are due to the consid- erable differences in the base levels of crime and prison populations for different values of.IS. At high values of IS, the system is operating effectively and pre- sumably aIreacly has a sizable proportion of offenders in prison and a fairly low crime rate. Any increase in.lS wit! only marginally increase the initially large prison population, while making a large fractional dent in an already low crime rate. At Tow values of JS, with a high crime rate and relatively small prison population, the changes that result from increasing IS will have a greater propor- tional impact on an initially small prison population. If one includes finite career length and the associated possibility that offenders

148 ._ _4 c) C) o ·> C. .4o ~ ~ to C. Ed C) ~ h_ Cot ·4- C) ~ o C) o o 4~ C. ah. oo .= C. Cal I U. o U. . - C. Cat a, .~ 5- Cat CS ~ o Ed ~ .= ~ ~ ~ be ~ ~ ~ ~ Ce .U) C,) ~ C) C: ~ ~ ~ ~ ~ . ~ ~ :i ~ ~ 0 ~ ~ > 0 ~ ~ °- ~ =~ ~ 11 ~ o c.) ~ ~ .= ll ~; ~ ~ o ~ d '3 ~ a' a ~=,~>,= r~ ~ c~ =.= ~ ~ ~ .= ed ~ O ~o d U)  4 - ~o _ _ U' 3 a ~ 0 ~ ~ ~ O ·S~. O ~ z C) ~o ~, (, O o 4 - C~ ~ ~ ~ d ~ ° 4~ et C~ ^^^^^^~____~ c~ 0 c~ ~ c~ oo 0 0 _ c~ ~ 0 oo oo 0 0 ~n C~ C~ C~ C~ C~ _ ~ C~ C] C~ _ _ ~4 + + + + + + + + + + + + + + + + + _ _ _ _ _ _ _' ~ ~ oo cD ~ o c~ c] ~ ~ ~ ~ oo - ~ 0 o co u: c~ u: c~ ~ oo 0 ~ 0 ~ o ~ c~ o ~ 0 cO ~ ~c~ co 00 ~ oO 0 ~ c~ oo ~ 0 oo In ~ ~ c~ ~ 0 O o ~ o cD U: C'] CD C';1 ~14 - 0 . . . . . . . . . . . . . . . . . . - ~ ~ 0 ~ d4 c~ 0 C~ ~4 00 di - co C~ ~ oo ~ ] 0 CS ~ ~ oO O ~-' C~ ~D ~ 0 ~ C~ O ~ _ c~ _ oo c~ u, ~ d~ oO ~ C~ 0 ~ ~ ~ ~ 0 - - - c] o - - c~ - o c~ - o - - o o o o o o oo o o o oo o o o ooo o o . . . . . . . . . . . . . . . . . . 'o o ~ o o o '~ o o '~ o o ~ o o ~ ~ o c~ o - o ~ o c~ ~ u) c~ o ~ - ~ ~ - - o oo u) 0 o ~ ~ oo ~ ~ oo o c~ 0 c~ c~ 0 0 o . . . . . . . . . . . . . . . . . - - c] 0 ~ - CD - O O O ~ ~ C~ 0 c~ 0 C~ _ c~ ~ O ~ ~ _ C~ 0 ~ C] 00 ~ ~ O O- C~ CD ~ - c] o - c~ o o - - o o c~ - - - - o o o oooooooooooooooooo C~ 0 0 ~ C~ C~ ~ CO O 0 0 CO CD U) ~ oo 0 C~ C~ ~ 0 C~ ~ C~ 0 ~ ~ oo 0 oo ~ ~ ~ oo 0 C53 ~ ~ _ C] ~ ~ ~ ~ _ 00 ~ ~ 0 C~ ~ C~ -^ C~ C~ _^ C1^ C1^ _ Cg^ C1^ C~^ _ d4 ~ c~ co O oo 0 C~ ~ -~ O O oO c~ 0 ~ C] oo 0 ~n ~ c~ oo ~ C~ 0 0 ~ ~ C~ O O CO oO O ~ O CD C~ ~ O ~ ~ C] ~ 00 (D ~ C~ C~ _ C~ 0 C~ ~ C~ 0 C~ oo ~ ~ ~ 0 ~ C~ 0 ~ C] 0 ~ ~ _ Co CV) C~ C] ~ C~ oo CO C':} C~) ~- =4 w) oo _ oo C~ ~ _ C~ CO C~ di - Cd ~ >~ ~ ~ ,x, c, o 3 3 ~ o .~ ._ o . - ~, ~ ~ =.~=a~ .= .= o ~ ~ 3 Z Z Z Z Z

149 _ ~ ~ _` ~ '_ ~ '_ ~ ~ _ _` ~ _ ~ oo o C~ C~ ~ ~ C] C~ ~ o C~ _ C~ C~ oo CO C~ + + + + + + + + + + + + + _ _ _ ~ _' ~ _ . ~ ~ _' . ~ C~ ~ o o C~ ~ _ U) _ ~ ~ ~ CD _ . . . . . . . . . . . . . ~ ~ CO o ~ oo o ~ ~ C~ C~ o ~ _ C~ _ _ C~ _ C~ . _ _ ~ _ CO C~ ~ _ ~ oo o C~ C~ ~ ~ _ ~ 0 ~ C] _ ~ o _ _ o _ C~ C] o ~ _ o o o o o o o o o o o C . . . . . . . . . . . , oo , C~ 1 _ ~ o ~. C~ ~ C~ C~ o o~ o~ o C~ C~ C~ ~ C~ C~ o o ~ o _ o C~ C~ Oo _ oo co ~ u, _ ln ~4 IO C~ C~ CS . . . . . . . . . . . . . __ _______C~__ o C~ ~ o oo ~ oo oo _ CD ~4 C~ C] oo C~ ~ C~ =4 _ ~ _ ~ ~ oo o CS _ ~ C~ C~ ~ oo oo ~ _ ~ oo 0 ~ oo C~ C~ ~ _ _ - oo ~ ~ ~ ~ o ~ ~ o o _ _ c~ cn oo o CC o ~ U: C~ ~- ~14 O) o 0 0 ~ 0 0 ~ ~ ~ ~ ~ u, 0 0 0 ~ cs ~ 0 oo c~ ~ 0 _ _ co ~ oo c~ 0 0 oo ~ ~ c~ _ oo c~ - _~ d ~' 0,_ O ed O a~ ~I,) ° ~ > _ _ ~ ~ ~ a) ~4 ~ cd  =~0~0~0~m - 0 0 3m =~= ~ ~ _ ~_ ~ ,~ _ ~ ~.,, (,, 0 ~ 0 ~ ~ ~ ~ ~ _ 0~ ~ ~ 0 0 ~ (,, 00 0 ,, ~:, ~ . ,,: ~ ~ ,= ~ , < e E c 0 ~' E E- ~ e 5 E _ ;) ~ ~ t° ~ 5 e · - :, ° O ·= <~) ~ ~ ~ ~ ·o .Cd ~, 0 C24 ~ ~ ~ ~ o~ . ~4 _ ~4 00 ~ 0 C] U: ~ ~ ~ ~ ~ ~ ^ ~ ~^ ~ ~ ~ ~ ~ 0 00 0 ~ ~ ~ CD ~ ~ ~ CS ~ 00 ~ O ~ ~ ~ ~ ~ O ~ ~ ~ ^ o C4 0 0 0 0 0 0 0 0 0 0 0 ~ ~ 3 ~, , ~ ~ ~ ~ E ~ _ 5 _ 5 5 E 5 O 5 ~ ,_ '5 9 ~ ~ E `~. E ° 5 ti ~ B E :, O ~' ..Q--~ ^= ~ o Q) ~ ~ =0 ~ =0 c,, ,=- ~ ~ ~ = to ~^ s0 ~ ~ O .C) ~ ~ ~ ~ ~o~ ~ ~ ~ 4 .= ~.~ ~ ~ ^.= O - ~ 5 ~ X~ ~ 5 ~ °

150 CRIMINAL CAREERS AND CAREER CRIMINALS 45 40 35 ._ o In ~30 .o - {L , ~ Q 25 0 ~ ~ c ~ ._ c ~ ~ x ~ - ~ ' ._ - o ._ 20 . _ . - . - m c, ~ ~10 15 o . _ _ .  \ .~. i \ ~ \ ~- _ - - - - . ~ 1 1 1 1 1 1 0 0.01 0.02 0.03 0.04 0.05 0.06 Expected Time Served per Crime fJS) in 1981 (years) FIGURE 5-2 Cost/benefit ratio with career length for selected states as a function of imprisonment sanction levels, IS, prevailing in 1981. Note: Cost/benefit ratio with career length is derived using Equation 5-13 with A = 5 and TR = 10. end their criminal careers while incarcer- ated, the estimates of prison population increases for each 1 percent reduction in crime will always be larger (see Table 5-61. The magnitude of this increase in the cost/benefit ratio varies with the length of time servecl. As indicated in Table 5-7, the longer the time served relative to resiclual careers of 10 years, the greater is the impact of including career length in the analysis. As found in nine states, for time served of 1 year or less, the

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE TABLE 5-7 Effect of Including Career Length in Cost/Benefit Ratio (A = S. TR= 10) Median Time Served (years) > 1.0 > 1.5 and and c 1.0 c 1.5 s 2.0 > 2.75 Mean percent in- crease in cost/ benefit ratio Standard deviation of percent in- crease in cost/ benefit ratio Number of states in 1981 19.6 30.5 40.5 75.5 1.6 2.8 2.7 16.3 9 14 6 2 cost/benefit ratio increases an average of 20 percent; for time served in six states of between 1.5 and 2.0 years, the cosUbene- fit ratio increases an average ot 4u per- cent. The impact of including career length is largest in two jurisdictions with very Tong time served: the cost/benefit ratio increases 87 percent in Washington, D.C., and 64 percent in Hawaii, where time served was 3.3 years and 2.75 years, respectively. Applying the Mode! to Estimate the Effects of a Selective Incapacitation Policy The model of a criminal career pre- sented in this section was also used to generate altemative estimates of the ef- fects of a selective incapacitation policy applied to predicted high-rate robbers found among Califomia inmates. The analysis used the basic approach devel- oped by Greenwood (1982), but with re- sults derived from the reanalysis of the original inmate survey data done for the panel by Visher (Volume II). I51 (See Chapter 6 and Visher EVolume II] for discussions of the scale itself.) On the basis of reanalysis of the survey data, Visher (Volume II) estimated Tower val ues for the mean individual crime rates in the three offender groups. Several other values of parameters for the incapacita tion model shifted slightly when she reestimated the incapacitative effects in Califomia (see Visher, Volume II:n. 23; Appendix A, Table 41. The Visher inmate distribution for the  three offense-rate groups is presented in the top row of Table 5-8. These estimates do not account for the effect of career termination on the inmate population. In particular, as careers get shorter, the in mate population wit! include increasing numbers of inmates who have ended their criminal careers while incarcerated; if they were released, these inactive in mates would no longer commit crimes. The total inmate population (R) can be partitioned into active inmates (RA) and inactive inmates (RI) using the relation ships developed in Equations 5-5 and 5-8. Estimates of the numbers of active inmates found in the total inmate popula tion are presented in the three bottom rows of Table 5-S for residual robbery careers of 15 and 5 years, using the cur rent average time served found in each offender group (see Table 5-9, below). As time served increases from low- to me dium- to high-rate offenders, the propor tion of active offenders in the total inmate Population for that offender group de creases: for residual careers of 5 years, the proportion of active offenders is 79 per cent among Tow-rate inmates, but it is only 67 percent among high-rate inmates. The numbers of active inmates in Ta ble 5-8 provide the basis for estimating the total number of active offenders in California inmates convicted of rob-each offender group, whether incarcer bery are partitioned into three groupsated or free in the community. Using the low-, medium-, and high-rate offendersfraction of careers that offenders spend based on their scores on a seven-factorincarcerates! (1) from Equation 5-1, the scale developed by Greenwooc! (19821.number of active offenders (N) is esti . . .

152 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE 5-S Distribution of California Inmates Convicted of Robbery by Predicted Robbery Frequencies Low Rate Medium Rate High Rate Total Percent Percent Percent Percent Number Active Number Active Number Active Number Active Total Inmatesa 2,865 4,942 5,942 13,749 Active Inmatesb TR = ~2,865 100 4,942 100 5,942 100 13,749 100 TR = 15 years 2,629 92 4,361 88 5,118 86 12,108 88 TR = 5 years 2,257 79 3,532 71 4,007 67 9,796 71 aData from Visher (Volume II:Note 23~. bThe number of active inmates (i.e., inmates who do not end their criminal careers while incarcerated) is estimated from Equations S-5 and 5-8, where SL.OW = 1.347, Smell = 1.997, and SHigh = 2.415 years (Table 5-9, below). When there is no career termination (TR = 00), the number of inactive offenders is negligibly small, and so all inmates are considered active. mated from N = RA/I. This procedure is used to estimate the size of the offender population under current incarceration policies; see Table 5-9. Section A of Table 5-9 presents the estimates of the current values of the various sanction risks faced by robbers in California. In the absence of any reason- able empirical basis for distinguishing sanction risks, the same probabilities of arrest, conviction, and incarceration for a crime are used for all three offender groups. The basis for these estimates is described in Greenwood (19821. Average time served is estimated separately for each group on the basis of the inmates' reports of their expected release dates. Using the relationships developed in Equations 5-3, 5-5, 5-9, and 5-10, Table 5-9 provides estimates of the number of robbery inmates and the number of rob- beries committed by adults in California under both the current policy and an alternative, more selective incarceration policy. The estimates are developed sep- arately for each offender group and Men combined to yield estimates of total in- mates and total crimes. The more selec- tive policy reduces time served to 1-year jai} terms for all predicted Tow- and me- dium-rate offenders' while increasing time served for all predicted high-rate robbers to twice the current prison term found among high-rate robbers. Alterna- tive estimates are provided for the cases when there is no career termination (Ta- hle 5-9. Sections A and B) and when the residual robbery careers are 15 years (Sections C and D) and 5 years (Sections E and F). Within any offender group (e.g., high- rate offenders), failure to account for ca- reer termination leads to the same per- centage errors for the expected number of crimes anc! for the expected inmate pop- ulation under any incarceration policy.  For example, if the crimes committed under some policy are underestimated by 10 percent, the number of inmates under that policy is also underestimated by 10 percent. If an incarceration policy varies across offender groups, however, the magnitude of the estimation errors will vary across the groups and depend on the relationship between S and TR in each group. This variation across offender groups contributes to the much larger impact of career termination on the estimates of prison population than on crime reduc- tion. In Greenwood's illustrative policy, high-rate offenders are assigned the long- est time served. Failing to account for their career termination under the pro- posed policy leads to large underesti- mates of crimes committed and of inmate

CRIME CONTROL STRATEGIES USING CRIMINAL CAREER KNOWLEDGE TABLE 5-9 Estimates of Effects on Crime and Inmates of a Highly Selective Incapacitation Policy for California Inmates Convicted of Robbery Values of Parameters for Offenders 1S3 Parameters of the Model Predicted Predicted Predicted Low Medium High Rate Rate Rate Total Current Policy- No Career Terminationa Average annual robbery frequency (A) 0.9 8.1 20.8 Probability of arrest and conviction .03 .03 .03 Probability of incarceration, given conviction .86 -.86 .86 Probability of incarceration, given a crime U) .0258 .0258 .0258 Probability of prison, given incarceration (p) .12 .27 .46 Average jail tells in years (s) 1.0 1.0 1.0 Average prison term in years (S) 3.892 4.692 4.075 Average time served in years [S = (1 - pus + pS] 1.347 1.997 2.415 Fraction of time free (1 - 1*) .96967 .70555 .43554 Number of inmates (R* = RA*) 2,865 4,942 5,942 13,749 Number of offenders (N = RA*II*) 94,461 16,784 10,527 121,772 Number of crimes [C* = N X A X (1 - 1*~] 82,437 95,920 95,367 273,724 Selective Policy No Career Terminationb Average time served in years (S) 1.0 1.0 8.15 Fraction of time free (1 - 1*) .97731 .82714 .18609 Number of inmates [R* = N X A X (1 - l*) X US)] 2,144 2,901 8,568 13,613 ~-l~o)C Number of crimes [C* = N X A X (1 - 1*~] 83,086 112,450 40,748 236,284 ~-14%)C Current Policy TR = 15 Years Fraction of time free (1 - 1) .97210 .73083 .47253 Total inmates (R = RA + RI) 2,865 4,942 5,942 13,749 Number of active inmates [RA = TR/(TR + S) X R] 2,629 4,361 5,118 12,108 Number of inactive inmates [R.= S/(TR + S) X R] 236 581 824 1,641 Number of offenders (N= RA/I) 94,229 16,202 9,703 120,134 Number of crimes [C = N X A X (1 - 11] 82,440 95,912 95,367 273,719 Selective Policy TR = 15 Years Average time served in years (S) Fraction of time free (1 - 1) Total inmates [R = N X A X (1 - 1) X JS)] Number of active inmates [RA = TR/(TR + S) X R] Number of inactive inmates [RI= Sl(TR + S) X R] Number of crimes [C = N X A X (1 - 1~] - 1.0 .97870 2,141 2,008 1.0 .83618 2,831  2,654 8.15 .26083 11,069 7,172 133 177 3,897 4,207 82,999 109,737 52,642 16,041 (+17%) 11,834 245,378 (-10%)

154 TABLE 5-9 Continue] CRIMINAL CAREERS AND CAREER CRIMINALS Values of Parameters for Offenders Predicted Predicted Predicted Low Medium High Parameters of the Model Rate Rate Rate Total Current Policy TR = 5 Yearse Fraction of time free (1 - 1) Total inmates (R = RA + RI) Number of active inmates rLRA = TR/(TR + S) X R] Number of inactive inmates [RI = Sl(TR + S) X R] Number of offenders (N= RA/I) .97595 2,865 2,257 .77028 4,942 3,532 608 1,410 93,846 .53365 5,942 4,007 1,935 15,375 8,592 13,749 9,796 3,953 117,813 Number of crimes [C = N X A X (1 - 1~] 82,430 95,929 95,370 273,729 Selective Policy TR = 5 Yearse Average time served in years (S) Fraction of time free (1 - 1) Total inmates [R = N X A X (1 - 1) X IS] Number of active inmates [RA = TRI(TR + S) X R] Number of inactive inmates [RI= Sl(TR + S) X R] Number of crimes [C = N X A X (1 - 1~] 1.0 1.0 .98102 .85168 2,138 1,781  357 82,858 2,736 2,280 456 106,066 8.15 .37552 14,111 5,366 8,745 67,111 18,985 (+38%) 9,427 9,558 256,035 (-6%) aThe variable values are based on a reanalysis of the original survey data available in Visher (Volume II). No adjustments for career termination are made. bThe selective policy involves 1-year jail terms for all inmates convicted of robbery and predicted to commit robbery at low or medium rates, and doubling the length of prison terms from 4.075 years under current policy to 8.15 years for all inmates convicted of robbery and predicted to commit robbery at a high rate. No adjustments for career termination are made. CThe difference between a decrease of 1 percent in the inmate population compared to a 1 percent increase estimated by Visher and the difference between a reduction of 14 percent in robbery and a 13 percent reduction estimated by Visher (Volume II) arises from differences in rounding the outcomes of various intermediate calculations. Remaining robbery careers (TR) are estimated to last 15 years. eRemaining careers (TR) are estimated to last 5 years. population for high-rate offenders. With their higher values of A and greater incar- ceration risk per crime, however, high-- rate offenders also contribute a much larger proportion to the inmate total than they do to the crime total for the three offender groups. As shown in Part F of Table 5-9, when remaining careers are 5 years long, high-rate robbers (both active and inactive) under the Greenwood pol- icy contribute 75 percent to total robbery inmates, but only 26 percent to total rob beries by adults. Because of the greater representation of high-rate robbers among inmates, their larger underesti- mates have more influence on the results for all inmates. The differences in the size of errors found across offender groups combine with differences in the contribu- tion of different offender groups to total crimes and total inmates to result in finite careers having a larger effect on estimates of inmate populations than on crime re- duction.