28

amount of variation explained by the same set of offense and prior record variables for both the sentence-length and time-served decisions. The relative amount of unexplained variation serves as a rough indicator of disparity after the offense and prior record variables are taken into account. Thus, a reduction in unexplained variation would be expected if the paroling authority was markedly reducing disparity in respect to similar circumstances of offense and prior record. Second, because the method of regression analysis used is limited in the extent to which significant subgroup differences that might exist are uncovered, a search for such subgroups was undertaken by means of tabular analysis, comparing disparity directly for specific groups of offenders with similar offenses and prior

record.

Results

29

Table 1 summarizes the sentence length and time-served data for the cases in the study. Whereas the standard deviation for sentence length is over twice as great as the standard deviation for timeserved, the coefficient of variation for time-served is thirteen percent smaller than for sentence length. The similarity between the two decisions on this statistic indicates that the relative variability of the two decisions is fairly similar; rather than sharply reducing the variability in sentence lengths, the time-served (i.e., parole board) decision apparently moves the individual cases down the scale of months served, thus affecting the variance of the two samples, but much less substantially affecting the relative variability about the mean." These data indicate that, on the average, these inmates served about fifty-two percent of their maximum sentences. The bivariate correlation coefficient between sentence length and timeserved is quite high (r +.85).

=

30

The findings that there is no large difference in the relative variability in time-served and sentence length and that sentence length and time-served are highly correlated casts some doubt on the hypothesis that parole board decisions substantially reduce sentence-length disparity. However, it could still be possible that a greater proportion of

See note 31 infra.

"The coefficient of variation standardizes the sample standard deviation to the sample mean. It is, therefore, useful in comparisons of relative homogeneity when groups have very different means.

30 Variances computed on logarithmic transformations of sentence-length and time-served yielded comparable results: time-served .46; sentence-length = .51.

the variation in time-served decisions is accounted for by offense and prior record items than is the variation in sentence length. Under the operational definitions set forth for this study, if this were the case, then regardless of the relative variability between the two decision points, time-served decisions would be less disparate. In order to address this question, both time-served in prison and sentencelength were regressed on the same set of offense and prior record variables. The large number of offense and prior record items available for analysis required that as a first step the number of independent variables be reduced. Thus, items with a significant bivariate association with either sentence-length or time-served were allowed to remain in the analysis. Under the operational definitions set forth here, if the parole board did reduce judicial disparity, then a greater proportion of the variance in time-served than in sentence length will be explained by these offense and prior record variables. If similar proportions of explained variance between the two decisions is found, or if less of the variance in time-served is accounted for relative to sentence-length, then this will be construed as evidence against the disparity reduction hypothesis. 31

31

32

The step-wise regression results for both sentence-length and time-served are presented in Table 2. For sentence-length, a total of six variables entered the equation in the step-wise analysis before additional variables added less than one percent to the explained variation (prior violent record and current violent offense, current conviction on assault, the seriousness of the charged offense, the number of counts on the current commitment, prior prison sentences, and the seriousness of the conviction offense). These items accounted for It should be stressed that the present study is not attempting to discover the specific factors most influential in determining sentence-length and time-served. Rather the criteria set for this analysis are the total amounts of variation explained in these decision outcomes by offense and prior record items. Obviously, some of these independent variables are highly interrelated. In addition to the items shown in note 22, supra, a limited search was undertaken to discover joint effects that would add additional explanatory power to the set of independent variables. Variables consisting of various combinations of present offense type and prior record experience, weapon use and prior record experience, and so forth were also constructed. These were added as additional variables. Thus, for example, attributes were constructed that placed persons with both a violent prior record and a current violent offense in one category and all others in another.

33

approximately forty percent of the variation in sentence-length. Thus, for this sample of cases and using the offense and prior record variables available for study, only a moderate amount of variation in sentence-length is accounted for. There is, therefore, under the criteria established for this study, considerable room for disparity (the unexplained variance) to be reduced by the parole board.

Also shown in Table 2 are the results of an identical analysis using time-served as the dependent variable. Seven variables, six of which were identical to those entering the sentence-length solution, entered before additional variables added less than one percent to the explained variation. Of interest is the fact that the two solutions accounted for almost identical proportions of explained variation. In fact, slightly less of the variation in timeserved is accounted for by offense and prior record variables. Thus, under the criteria established for this

havior categories developed by M. Warren and E. Reimer for use in a "Parole Movement Scale" in the Research Division of the California Department of Corrections. The scale values are the median scores obtained for each item in a decision game in which parole board members and correctional administrators were asked to indicate the probability of successful parole required for parole release after serving the average (median) time for the particular offense/behavior category. The score values range from 235 for bigamy to 887 for acts of deliberate, planned violence causing death of an adult. For the exact scale values and the rules used for coding offenses from case files, see D. Gottfredson & S. Singer, supra note 14. Clearly, this scale is only a rough measure of offense seriousness, defined by a restricted group of persons. It does, however, serve to classify offenses in terms of behavioral elements (e.g., value of loss, degree of injury, extent of monetary loss, and presence of weapons) and may therefore be preferable to a simple hierarchy based on statutory classifications.

33 A regression using the logarithm of sentence length as the dependent variable produced virtually identical results.

study, time-served decisions are no less disparate than are sentence-length decisions in this sample. The high correlation in this sample between time-served and sentence-length, when compared with the moderate associations between the offense and prior record variables and time-served, indicates that the best overall predictor of time-served is judicial sentence length. To some extent such a correlation is logically necessary; after all, the maximum and minimum amounts of time served in prison are constrained by where on the scale of punishment the judge sets the penalty. However, the large proportion of variation unexplained by offense and prior record found in this study for sentence-length left considerable room for disparity reduction, as defined here, on the part of the parole board. These results indicate, however, that for this sample the parole board decisions were highly consistent with the sentencing decision. Of course, it might be the case that other offense and prior record factors or their combinations not included in this data set could reduce the unexplained variation in sentence-length and thus reduce the apparent disparity found here.34

34

As noted in the introduction, it has been claimed that the disparity reduction function of the Federal Parole Board is most effective in those situations in which the Board is not constrained by the mandatory one-third minimum sentence of the "Regular Adult" sentence option. This possibility was examined here by employing identical procedures reported above for the total sample, but only within those cases sentenced under 18 USC § 4205 (b) (2)-cases for which the parole sets the minimum parole eligibility date up to one-third of the maximum (approximately one-third of the sample). For these cases (N=746) time-served and sentence-length were less strongly correlated than for the total sample (r = 68), however, the results of the regression analysis were also not supportive of the disparity reduction hypotheses. Thus, 45% of the variation in sentence-length was explained by three factors (current offense was robbery, simple versus consecutive sentences, and seriousness of the charged offense) before additional entrants explained

=

Forgery under $500, simple sentence, at least one prior conviction (N Car Theft (unplanned), simple sentence, at least one prior conviction (N = 217) Forgery over $500, simple sentence, at least one prior conviction (N = 89) Theft (planned), simple sentence, at least one prior conviction (N = 291) Car Theft (planned), simple sentence, at least one prior conviction (N = 284) Selling narcotic for profit, simple sentence, at least one prior conviction (N 54)

=

167)

decision points. Of most relevance to the disparityreduction hypothesis are the columns labeled "coefficient of variation". It can easily be seen that, overall, the differences in the coefficients of variation between sentence-length and time-served are not large; however, some subgroup differences do exist. In five of the seven comparisons there is virtually no difference in the two measures. In two groups, there is some evidence in favor of the reduction hypothesis. For cases involving forgery under $500 with simple sentences and at least one prior conviction, the standard deviation for sentence length is sixty-one percent of the mean, whereas for the identical cases the standard deviation for time-served is fifty-two percent of the mean, a reduction of fifteen percent. For cases involving selective service violations with simple sentences and at least one prior conviction, the standard deviation for sentence length is thirty-five percent of the mean whereas for time-served the standard deviation is twenty-two percent of the mean, a reduction of thirty-seven percent. Of interest is the finding that this latter classification,

36

The coefficient of variation is dependent on the standard deviation which in turn may be highly influenced by a few extreme scores. In order to ensure against the possibility that a few extreme cases were responsible for the apparent reduction in variability for the two classes discussed above, an identical analysis was performed using only cases falling between the 10th and 90th percentiles on the sentence-length distribution. The results were similar to those reported in the text.

which demonstrates the greatest disparity reduction as defined here, was the group with the smallest standard deviation on sentence length. These data indicate, therefore, that some subgroup differences in the extent to which parole boards affect judicial disparity may exist, and that such reductions may vary in magnitude according to the particular offense under consideration." These results also indicate that such reductions are not consistent for all categories of offenses.

The Influence of Institutional Behavior on Time-Served

These data clearly suggest that the major indicator of time-served in prison is judicial sentence length. Although there is considerable evidence that the parole board moves penalties down the scale of severity, indications are that this is done systematically with little reduction in variability

37 The analyses reported in Table 3 were repeated for the subgroups shown regardless of type of sentence (1.2., cases were included whether the sentence was simple, consecutive or concurrent). It might be thought that disparity arising from sentence type (ie., otherwise similarly situated offenders given different types of sentences) is reduced by the parole board. Disparity of this type is, of course, more difficult to operationalize. However, for the subgroups shown in Table 3 the results, in terms of differences between the coefficients of variation, are essentially the same. One exception are cases of planned theft with at least one prior conviction in which the coefficient variation for time-served is 11% less than for sentence length.

for similarly situated offenders. The question therefore arises whether factors uniquely known to the parole board appreciably affect the parole board's decision of time-served. As noted earlier, the ability to witness institutional behavior prior to setting a release date has served as a major justification for such delayed penalty-setting sentencing structures.

Perhaps the institutional behavior of most concern-both to post-release prognosis and to the theory of parole release as a mechanism of institutional control-is rule infraction. Consistent inability to comply with prison regulations can be viewed as evidence of increased probability of future law violation, and the threat of a longer prison stay might be an effective deterrent to institutional rule infraction. Several rule infraction items that were available to the parole board for their decisions were available for this part of the study. First, the number of prison punishments on the current stay was coded, defined as any action (other than dismissal) on charges of violations of prison rules resulting in withholding of privileges, segregation, isolation, loss of good time, any suspended sentence, or other deprivation. Second, prison assaultive infractions were coded, defined as any assault or threat to assault, resulting in a disciplinary infraction during the present confinement, unless there was a finding of not guilty. Thus, while the first item gives an indication of the frequency of rule-breaking, the second item gives some indication of the seriousness of such infractions. Third, whether or not the inmate had a record of escape or attempted escape from prison during the present confinement was coded.38

Institutional behavior items other than those relating to rule infractions, such as successful participation in treatment programs, are obviously of additional theoretical interest in a study of timeserved decisions. Unfortunately, data relating to treatment participation were only partially available for this study and therefore were not included in the analysis. The emphasis on disciplinary issues, however, most notably in contemporary reform proposals, suggests that there is considerable merit in ascertaining their influence in reality on time-served decisions. The bivariate correlations between the rule infraction items and time-served

39

38 There are limitations involved in using these items as indicators of prison rule infractions. Perhaps most important is the insensitivity of these items to the seriousness of the infractions, which is only partially overcome by the assault item. For this reason, the results of this phase of study should be viewed cautiously. See N. MORRIS, supra note 1.

in prison (n 2506; each significant at the p. < .05 level) were as follows: escape history, r = .10; assaultive infractions, r = .14; and prison punishment, r = .24.

The question posed for this portion of the study was whether institutional behavior had a significant impact on time-served for the persons in the sample once the judicial decision as to sentence length was taken into account. Essentially, the purpose was to discover the extent to which the parole board modified judicial decisions on the basis of knowledge about institutional infractions.

To address this issue, predicted scores for timeserved were formed on the basis of the linear regression equation obtained by regressing timeserved on sentence-length. From these predicted score values, residual scores were derived for each person in the sample. The total variation in these residual scores (which was twenty-eight percent of the total variation in time-served) was then treated as the dependent variable for a multiple linear regression that treated the institutional behavior items as independent variables. A step-wise solution was used with only items adding at least one percent of the variance in the residual scores permitted to enter. It should be stressed that the purpose of this analysis was not to determine which items, among those available for study, were most determinative of time-served. Rather, the purpose was to determine the proportion of variability remaining in time-served once the judicial decision as to sentence-length was taken into account that could be accounted for by these institutional behavior items. The results are presented in Table 4.

The two institutional behavior items entering the regression equation together account for less than ten percent of the residual variation in timeserved. Thus, there is evidence that institutional behavior of inmates may influence the time served in prison, but that the influence is not large. It