 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Services Research Outcomes Study (SROS)
|
|
|||||||||||
|
|
| |||||||||||
|
Services Research Outcomes Study (SROS) | ||||||||||||
CORRELATES OF TREATMENT OUTCOMES
In the preceding section, group and subgroup before/after treatment differences were compared. However, the presence of differences does not necessarily indicate which variable or set of variables explains a sufficient proportion of variance to be of value in a predictive sense and does not specify the relationships among several variables. This section presents regression models to predict outcomes of the treatment episode.
This analysis seeks to find covariation among the variables—that is, to assess the extent to which demographic characteristics (i.e., age, sex, and race/ethnicity) and behaviors (i.e., drug use and criminal behavior) coincide with outcomes in the five years after treatment, as measured by behavior (i.e., drug use and criminal behavior). The variables used in these regression analyses are not intended to be exhaustive of the relationships that could be assessed from this rich data resource.
The four assumptions that structure the regression models are the following: (1) the presence, absence, or level of a characteristic before a client enters treatment is predictive to some degree of the same characteristic after treatment; (2) other pretreatment circumstances, such as education and reasons for seeking treatment, may have sustained effects on the outcome; (3) measured aspects of the treatment episode will contribute to predicting the outcome; and (4) post-treatment circumstances and factors will further contribute to predicting the outcome. The pre-, in-, and post-treatment variables are tested cumulatively, as a model is built up successively in somewhat the same way that the client experiences the sequence of conditions. This sequencing is summarized in Figure 3-17: pretreatment variables (Model 1); pretreatment and in-treatment variables (Model 2); and pretreatment, in-treatment, and post-treatment variables (Model 3). Model 1 predicts outcome after treatment as a direct test of the pretreatment variables. Model 2 assesses the increase in predictive power (R squared) and the directions of the variables in the model. Finally, Model 3 is entered, again simultaneously, to assess the strength and direction of the relationships. The models attempt to predict behavior in the five years after the treatment episode; the component variables are described below. Each model adds to the previous model (e.g., Model 2 adds treatment variables to the pretreatment variables contained in Model 1).
Outcome Variables
The outcome variables being predicted represent two key sets of outcomes: drug use and criminal behavior in the five years after the SROS treatment episode.
·510·Drug use: alcohol, marijuana, cocaine, crack, and heroin. The first set of analyses assesses the simple dichotomous classification of reported use or nonuse of the drug during the five-year period following the SROS treatment episode. For this set of measures, logistic regression was used—a technique that generates "odds ratios" estimates for each predictor. Such estimates are easily interpretable probabilities that indicate how much more likely it is that an outcome would be observed if, all other elements being the same, the predictor occurs instead of a comparison condition. For example, all other things being equal, an odds ratio would estimate how much more (or less) likely an older client is to use crack after treatment than a younger client. In addition to these logistical results, a second set of regressions uses continuous variables in standard least squares linear regression analyses for the average number of days per month of drug use reported after the SROS treatment episode.·510
·510·Criminal Behavior: selling drugs, prostitution/procurement, larceny (shoplifting, theft, etc.), and breaking and entering. These crimes were chosen based on their overall frequency of occurrence relative to the other crimes assessed by the SROS questionnaire. The investigation—whether a particular type of crime was committed by the respondent in the five years after treatment—used logistic regression as described above and elsewhere.·510
Figure 3-17. Models explaining behavior after SROS treatment episode
Predictor Variables
The variables used in each logistic regression model to predict drug use and criminal behavior are as follows:
Model 1— Before Treatment
·510·Demographic characteristics of the respondents: gender, race/ethnicity, and age at the time of discharge from treatment in 1989–1990, represented by both a linear and nonlinear term to reflect possible curvature in the relationship of age to the target variables.·510
·510·Behavior before treatment: self-reported use of each major drug, including the target variable, during the five years preceding the 1989–1990 treatment episode, and self-reported commission of specific crimes in the five years before treatment.·510
·510·Reason for entry into treatment: in particular, whether criminal justice pressure ("legal pressure" in the shorthand of the table headings) was a factor in seeking treatment.·510
Model 2—Before and During Treatment
·510·Behavior during treatment: self-reported use of drugs or alcohol during treatment. ·510
·510·Treatment characteristics: treatment type (inpatient, residential, outpatient methadone, or outpatient nonmethadone, using inpatient as the common comparison group), whether the client completed the treatment plan, length of the treatment episode, and treatment revenues (costs) per patient.·510
·510·Relationship with the principal clinician (counselor): specifically, whether the counselor understood the client’s problems. ·510
Model 3—Before, During, and After Treatment
·510·Additional treatment episodes and use of 12-step programs: such as Alcoholics Anonymous (AA), Narcotics Anonymous (NA), and Cocaine Anonymous (CA) during the five-year post-treatment period. This may have occurred at any time during the five-year period and may be presumed to have a responsive or reciprocal relationship with drug use after the index treatment.·510
Prediction of Alcohol and Drug Use
Alcohol Use
Appendix Table B-41 presents the set of regression models that attempt to predict any alcohol use (five times or more) in the five years after the 1989–1990 treatment episode. For each predictor in each model, the table reports the estimated odds ratio, a statistic that describes the association between the predictor variable and the target variable, and indicates which odds ratios were statistically significant at the conventional level of p < 0.05. In other words, the model estimation procedure calculated how strongly the individual predictors were associated with client use of alcohol five or more times after treatment. Although each of the stated predictors was used in the respective models, only some of the predictors yielded coefficients that were significantly different from 1.00—that is, only some predictors increased or decreased the chances of observing post-treatment alcohol use compared with the odds of such use occurring in the comparison condition.
In the simple pretreatment model (first column), the two age variables were significant, showing that older individuals were less likely to use alcohol than younger individuals and that the relationship was not strictly linear. In the strongest correlation, those who used alcohol five or more times across the five years before treatment (versus those who did not) were almost 11 times (10.91) more likely to use alcohol after treatment, all else being equal. Alcohol use after treatment was not predicted by the use of any illicit drug before treatment. Men were almost one-and-one-half (1.43) times more likely than women to use alcohol after treatment, and Hispanics were one-half (0.54) as likely as whites. Finally, those who entered treatment under legal pressure were nine-tenths (0.88) as likely to use alcohol after treatment as those under no legal pressure.
Model 2 (second column) added in-treatment variables to the predictors in Model 1. The addition of the in-treatment variables yielded only minor changes in the strength and significance of the Model 1 predictors. Specifically, when in-treatment as well as pretreatment variables were taken into account, clients who used alcohol before treatment (versus those who did not) were still found to be 11 times (11.13) more likely to use alcohol after treatment; males were almost one-and-one-half (1.42) times more likely to drink than females, and Hispanics were one-half (0.57) as likely as whites to do so. Only the odds ratio attached to legal pressure faded from a significant odds ratio (0.88) to a nonsignificant one (0.96).
Among the in-treatment predictors, clients who used drugs during treatment (versus those who did not) were twice (2.03) as likely to use alcohol after treatment. Clients who remained in treatment for one to six months were about one-half (0.55) as likely to use alcohol after treatment as those who were in treatment less than one week, and clients who remained in treatment six months or longer were only one-third (0.33) as likely to drink after treatment as those who remained in treatment for less than one week. Clients who believed their primary counselor understood the client’s problems well were only three-quarters (0.74) as likely to usealcohol after treatment as those who reported that their counselor did not understand their problems well. Finally, clients discharged from methadone treatment were less than one-half (0.40) as likely to use alcohol after treatment as those discharged from inpatient facilities. The revenue-per-patient variables did not predict alcohol use after treatment.
In the final model (third column) the two post-treatment variables were added to the predictor side of the equation. These additions yielded only slight changes to the odds ratios and significance levels calculated in Model 2. In particular, clients with one- to six-month lengths of stay were slightly more likely than in Model 2 (0.64 instead of 0.55) to be post-treatment drinkers, and the odds ratio slipped from a significance of p _ 0.01 to p _ 0.08, just out of the p < 0.05 range. Clients who had additional treatment episodes after treatment were about 30 percent more likely to use alcohol after treatment than those who did not return to treatment (their use of alcohol is probably associated with a return to treatment); AA/NA/CA attendance was not significantly associated with any change in the likelihood of drinking. The multiple r2 for Model 3 was 0.19—that is, all the elements in the combined model accounted for 19 percent of the total variance in the target variable. Model 2 accounted for 18 percent, and the pretreatment variables alone (Model 1) accounted for 15 percent of the variance. In other words, the pretreatment variables in the model were able to account for about six times as much of the predicted outcome as the in-treatment variables did when they were added.
Any Illicit Drug Use
Appendix Table B-42 presents results of regression models for any illicit drug use in the five years after treatment. Unlike alcohol, the age variables were not predictive for any illicit drug use. However, use of any of the principal drugs before treatment was highly predictive of post-treatment use, with odds ratios of about 1.7 for pretreatment cocaine and crack use and more than 5 to 1 for pretreatment users versus nonusers of marijuana or heroin. Those who used alcohol before treatment were one-half as likely to use illicit drugs as those who did not, and men were 60 percent more likely than women to use any illicit drug after treatment. Clients with legal pressure were 12 percent more likely to use any illicit drug after treatment than those without legal pressure, a small but statistically significant difference.
According to Model 2, clients who completed their treatment plans were 61 percent as likely (that is, 39 percent less likely) to use any illicit drug after treatment than those who did not. Those whose lengths of stay in treatment were one week to less than one month were about 59 percent as likely to use any illicit drug after treatment as those who stayed less than one week, and clients who remained in treatment six months or longer were only 42 percent as likely. Further, clients who used drugs during treatment were more than three times as likely to use any illicit drug after treatment as others who did not.
Model 3 indicates that those who had additional treatment episodes after treatment were 58 percent more likely to use any illicit drug after treatment as those who did not return to treatment. Model 3 accounted for 42 percent of the variance in any illicit drug use aftertreatment, Model 2 for 39 percent, and Model 1 for 35 percent. The pretreatment variables accounted for about ten times as much of the predicted variance in modelled outcomes as the in-treatment measures did when added to the model.
Marijuana Use
Appendix Table B-43 covers marijuana use in the five years after treatment. Clients using marijuana before treatment were nearly 25 times (24.29) as likely to use marijuana after treatment as those who did not. Pretreatment use of cocaine increased the odds of post-treatment marijuana use by 48 percent. Males were twice (2.1) as likely to use marijuana after treatment than females; neither legal pressure nor race/ethnicity predicted marijuana use after treatment.
Clients who remained in treatment six months or longer were one-half (0.50) as likely to use marijuana after treatment as those who remained in treatment for less than one week, and those who used drugs during treatment were almost four times (3.71) as likely to use marijuana after treatment as those who abstained. Model 3 indicated that clients who received additional treatment were 14 percent more likely to use marijuana after treatment than those who did not return. Model 1 accounted for 43 percent of the variance in marijuana use after treatment; Models 2 and 3 both accounted for 48 percent; thus, pretreatment characteristics accounted for about ten times as much of the variance in outcomes as the in-treatment elements did when added to the model.
Cocaine Use
Appendix Table B-44 covers cocaine use in the five years after treatment. Use of cocaine before treatment added more than 750 percent to the odds of using this drug after treatment (8.58); and pretreatment heroin or marijuana use made it 2.4 and 1.9 times as likely, respectively, that cocaine would be used after treatment. Differences in age, sex, race/ethnicity, and legal pressure did not predict cocaine use after treatment.
According to Model 2, clients who remained in treatment six months or longer were only one-third as likely to use cocaine after treatment as those who stayed for less than one week. In addition, clients who used drugs during treatment were three times as likely to use cocaine afterwards as those who did not use drugs during treatment. There were no statistically significant differences among the treatment types or revenue variables, but subsequent treatment was associated with post-treatment cocaine use at an odds ratio of 1.09. The variance accounted for by the three models was 22, 25, and 26 percent, respectively; most of the predicted variance was due to pretreatment variables.
Crack Use
Appendix Table B-45 reports on regression models of crack use after treatment. Client use of crack before treatment made crack use after treatment more than 8 times as likely. Cocaine use before treatment added 58 percent to the likelihood of post-treatment crack use, and blacks were almost twice (1.93) as likely as whites to use crack after treatment. After controlling for pretreatment use, client age did not predict post-treatment crack use.
Adding the in-treatment variables with Model 2, longer stays did not yield a statistical difference in crack use after treatment (the odds ratio for a six-month length of stay was not significant), although a further analysis among clients in inpatient, residential, and methadone facilities separately did indicate a length-of-stay effect in each type. Overall, clients who used drugs during treatment were more than twice (2.3) as likely to use crack after treatment, and clients discharged from outpatient nonmethadone treatment were one-half (0.47) as likely to use crack as those discharged from inpatient treatment.
According to Model 3, clients who had additional treatment episodes after the 1989–1990 episode were about one-quarter (1.26) more likely to use crack after treatment than those who did not return to treatment. The variances accounted for by the three models were 27, 29, and 32 percent, respectively.
Heroin Use
Appendix Table B-46 summarizes the models predicting heroin use in the five years after treatment. Using heroin before treatment was associated with nearly a fiftyfold increase (48.91) in the odds of using heroin after treatment. Using cocaine before treatment almost doubled (1.95) the likelihood of heroin use after treatment. Age, sex, and race did not generally predict heroin use after treatment, but Hispanics were twice as likely as whites (2.27) to use heroin after treatment. Legal pressure did not predict heroin use generally, but in a separate analysis, methadone clients with pressure from the criminal justice system were 30 percent more likely than those without it to use heroin.
Length of stay (Model 2) had no significant effect on post-treatment heroin use, but clients who used an illicit drug during treatment were more than twice (2.34) as likely to use heroin after treatment as those who did not use any illicit drug during treatment. Finally, clients discharged from outpatient nonmethadone treatment were less than one-third (0.29) as likely to use heroin in the post-treatment period compared with inpatients, and facility revenues per patient had a small but significant association with heroin use after treatment..
There was little difference in post-treatment heroin use between those with and without additional treatment episodes and between those who did or did not attend 12-step programs. Model 1 accounted for 44 percent of predicted variance, whereas the other two models accounted for 47 percent; none of the detailed measures of treatment effects was nearly aspowerful a predictor as pretreatment heroin use, which had a stronger effect on post-treatment heroin use than any other predictor-target pair in all the analyses. This result may serve as testimony toward Kaplan’s (1983) nomination of heroin as "the hardest drug."
Summary
Statistical models of post-treatment drug use outcomes were able to account for close to one-half the variance in heroin, marijuana, and any illicit drug use; one-third of the variance in crack use; and one-quarter and one-fifth of the variance in cocaine and alcohol use, respectively. In every instance, the strongest predictor of post-treatment drug use was the use of the same drug (or any illicit drug in predicting the same global variable) in the pretreatment period, a relationship that was especially strong for heroin. Cocaine was also predictive of both crack and marijuana use, and marijuana and heroin were predictive of cocaine use, which perhaps indicates a degree of commonality of use that is somewhat distinctive of cocaine. Using any illicit drugs during treatment, even controlling for pretreatment use, was also predictive of every type of drug use (including alcohol) after treatment. The power of past drug use to predict future drug use is perhaps another way of stating the chronic, habitual, addictive, and in short, the highly persistent nature of what treatment programs are treating.
Even controlling for levels of pretreatment use, males and younger clients were more likely to use alcohol and marijuana after treatment, blacks were more likely to use crack, and Hispanics were less likely to drink. The strength of these demographic associations was roughly similar to the strength of length of stay as a predictor of alcohol, marijuana, cocaine, or any illicit drug use after treatment. Length of stay was not strongly associated with post-treatment crack or heroin use; however, for these two drugs, outpatient nonmethadone treatment substantially reduced the odds of post-treatment use compared with inpatient treatment, which is generally much shorter term. Thus, the length-of-stay variable was masked by its collinearity with these differences in type of treatment. Counselor understanding of the client’s problems was significantly associated with lower alcohol use and (in one of two models) heroin use, but not other drugs—a finding that invites further work.
Finally, small but significant associations were found between every measure of post-treatment drug use and additional treatment during the post-treatment period. It seems most likely that the return to treatment followed rather than (or as well as) preceded drug use in most of these instances, but the SROS data do not permit definitive conclusions to be drawn about the sequence of these post-treatment events.
Average Number of Days Used Drugs per Month: Ordinary Least Squares Regressions
The previous section described the results of logistic regression analysis predicting the use of alcohol and several drugs during the five-year period after treatment. A parallel series of analyses was conducted using the technique of ordinary least squares (OLS) regression and measuring the average number of days per month of drug use as the target variable.
Alcohol
Appendix Table B-47 presents the results of OLS regression analyses using the same predictor variables as in the logistic series to model the average number of days per month of alcohol use after treatment as the dependent variable. For Model 1 (pretreatment), the significant predictor variables were the following: The number of days per month the respondent reported using alcohol before treatment was a significant predictor of reported use after treatment. The regression weight predicted a 0.3 increase the number of days per month of reported alcohol use after treatment for each one-day increase in the number of days of reported use before treatment. Although the number of days per month of pretreatment reported use of marijuana, crack, and cocaine were unrelated to the number of days per month of post-treatment reported alcohol use, there was a small, but significant, relationship with heroin (B = 0.14). The strongest variable in Model 1 was sex: Controlling for the other variables in the model, males reported use of alcohol 1.6 days per month more than females after treatment.
Model 2 added in-treatment variables. The length-of-stay variables showed the strongest coefficients. Compared with those who stayed in treatment for less than one week, clients who stayed in treatment for one week to less than one month reported an average of 2.2 fewer days per month of alcohol use after treatment; clients whose treatment stays were one to less than six months reported an average of 3.3 fewer days per month of alcohol use after treatment; and a stay in treatment of six months or more resulted in an average of 5.3 fewer days per month of alcohol use after treatment. Clients who reported using drugs during treatment were estimated to average three days per month more alcohol use after treatment, compared with those who did not use drugs during treatment. The regression equations for each of the three models explained less than 20 percent of the variance in days per month of alcohol use after treatment.
Marijuana
Appendix Table B-48 presents the regression analyses for the average number of days per month of reported marijuana use after the treatment episode. Age was found to be a significant explanatory variable for post-treatment marijuana use, and each additional year of age predicted a lower average of four-tenths of a day per month in reported use of marijuana after treatment. Each day per month of pretreatment marijuana use predicted an additional 0.3 days of post-treatment marijuana use per month. The number of days per month of alcohol, crack, cocaine, and heroin use before treatment were not associated with changes in the number of days per month of marijuana use after treatment. Males reported using marijuana 1.6 days per month more than females after treatment.
Model 2 added treatment variables. The length-of-stay variables were not significant. The use of drugs during the treatment episode was associated with an average of 3.1 more days per month of reported marijuana use after treatment, compared with those who reported not using drugs during treatment. The complete (Model 3) regression equation predicted 28 percent of the variance in days per month of marijuana use after treatment.
Crack
Appendix Table B-49 contains regression coefficients predicting the average number of days per month of reported crack use after the treatment episode. The age variables were not significantly related to the target crack variable. A significant Model 1 (pretreatment) variable was the average number of days per month of crack use before treatment; each additional day of reported use before treatment was associated with a 36-percent increase in the average number of days per month of reported crack use after treatment. Pretreatment days per month of marijuana, cocaine, and heroin use were not related to reported post-treatment crack use. Black (non-Hispanic) clients used crack 1.38 days more per month after treatment compared with whites. There was a weak effect of 0.18 days per month more reported crack use after treatment for those who were under legal pressure to enter treatment.
In Model 2, the length-of-stay variables predicted lower crack use after treatment for longer stays, but none of these coefficients was statistically significant. Use of drugs during the treatment episode was associated with an average of 1.5 days per month more of reported crack use after treatment, compared with those who reported not using drugs during treatment. Model 3 predicted 29 percent of the variance in reported crack use after treatment.
Cocaine
Appendix Table B-50 models the average number of days per month of reported use of cocaine after the treatment episode. Age was not related to the number of reported days per month of cocaine use after treatment; likewise, the number of pretreatment days per month of reported alcohol, marijuana, and crack use also were not related. Statistically significant Model 1 variables included the number of days per month of reported pretreatment use of cocaine and of heroin. Each additional day of reported cocaine use before treatment was associated with a 0.3-day increase in the average number of days per month of reported cocaine use after treatment, and each additional day of reported heroin use before treatment was associated with a 0.14-day increase in the average number of days per month of reported cocaine use after treatment. Sex and race/ethnicity were not statistically significantly related to days per month of reported cocaine use after treatment.
Model 2 added treatment variables, and the length-of-stay and drug-use variables showed statistically significant results. The shorter length-of-stay variables showed negative coefficients, but they were not statistically significant. However, clients who reported a treatment stay of six months or more also reported an average of 1.6 fewer days per month of cocaine use after treatment than did those with the shortest reported treatment stay. Finally, those who reported using drugs during the treatment episode reported an average of 1.6 more days per month of cocaine use after treatment than did those who reported not using drugs during treatment. Model 3 predicted 26 percent of the variance in reported cocaine use after treatment.
Heroin
Appendix Table B-51 models the average number of days per month of heroin use after the treatment episode. Age was not related to this outcome variable. Each day of reported heroin use before treatment was associated with an increase of 0.58 days per month in reported heroin use after treatment. Pretreatment alcohol, marijuana, crack, and cocaine use were not related to the number of days per month of post-treatment heroin use. Neither sex nor race/ethnicity predicted the number of days per month of heroin use after treatment.
Model 2 length-of-stay variables did not predict the average number of reported days per month of heroin use after treatment. Clients who reported completion of treatment plans reported 1.2 fewer days per month of heroin use after treatment than did those who reported not completing treatment. Clients who reported using drugs during the treatment episode reported six-tenths of a day per month more of heroin use after treatment than those not using drugs during treatment.
The heroin OLS Model 3 predicts 46 percent of the variance in days per month of heroin use after treatment.
Summary
The results of the OLS regressions reproduced many of the logistic model results, but they departed in ways that invite a cautious reminder about the ever-present potential for specification error in building statistical models. The main cure for this type of error is continued careful and intensive study of the data and comparison of results with other data sets as these become available. The power of drug use before and during treatment to predict drug use after treatment was confirmed by these analyses, as was the association of crack use with black clients and alcohol and marijuana use with males. The associations between length of stay in treatment and post-treatment drug use were much weaker in the OLS than in the logistic models, reaching significance for individual length-of-stay coefficients only for post-treatment alcohol and cocaine use. The firm association in the logistic regressions between later treatment episodes and post-treatment drug use was also not evident in the OLS models.
Prediction of Criminal Activity After Treatment
Selling Drugs
Appendix Table B-52 displays the results of logistic regression models that predict drug trafficking after treatment. Selling drugs before treatment was the major predictor, increasing the odds of post-treatment drug selling by about 10 to 1 in every model. Older clients were less likely—and male and black clients more likely—to sell drugs after treatment, and clients who had committed theft or larceny before treatment were 75 percent more likely to sell drugs after treatment. Of the treatment variables assessed in Model 2, only length of stay and whether the respondent had used drugs during treatment had effects on selling drugs after treatment: Oddswere one-third as likely for longer stays and three times as likely for drug users during treatment. Clients with additional treatment episodes were 11 percent more likely to sell drugs than those without additional episodes. Model 1 accounted for 31 percent and Models 2 and 3 accounted for 35 percent of the variance in selling drugs after treatment.
Prostitution/Procurement
The odds of engaging in prostitution or procurement of sex for money (see Appendix Table B-53) were much higher (30.27 in Model 3) after treatment if this activity had preceded treatment, and clients who were black or committed burglary (breaking and entering) were about three times as likely to engage in post-treatment prostitution or procurement.
Length of stay was an important variable; clients who stayed in treatment for six months or more were only one-sixth as likely to commit these offenses. Drug use during treatment was also predictive, doubling the odds. Those with additional episodes of treatment were slightly more likely to be involved in prostitution/procurement. Model 1 accounted for 33 percent of the variance, whereas Models 2 and 3 accounted for 40 and 41 percent, respectively.
Larceny/Theft
There were no significant demographic predictors of post-treatment larceny and theft (see Appendix Table B-54). Only pretreatment larceny or theft (odds ratio about 7) or burglary (slightly more than 2) increased the odds, and lengths of stay in excess of six months decreased them (0.41). Drug use during treatment increased the odds (2.12), as did subsequent treatment or AA/NA/CA attendance (1.23 and 1.02, respectively). The respective models accounted for 23, 25, and 29 percent of the variance.
Breaking and Entering
The final criminal activity modeled by logistic regression techniques was breaking and entering, or burglary (see Appendix Table B-55). Pretreatment commission of burglary and prostitution/procurement were strong predictors (8.58 and 2.66 in Model 3, respectively), but clients who were black or male were also more likely to commit burglary after treatment (2.20 and 2.72, respectively). In addition, subsequent treatment and AA/NA/CA attendance were also correlated at higher odds (1.27 and 1.02, respectively). Only medium-term length of stay was a significant predictor in Model 2, and this model accounted for only an additional one percent of variance over the 27 percent accounted for by Model 1.
Summary
The patterns in the logistic models of post-treatment criminal activities closely followed the patterns seen for drug use after treatment. The models accounted for about one-third of the variance in criminal outcomes, and specific criminal behavior before treatment was the strongestpredictor of specific criminal behavior after treatment. Moreover, pretreatment prostitution/procurement and larceny further increased the odds of other kinds of post-treatment criminal activities. Even controlling for criminality before treatment, males were more likely than females to sell drugs and commit burglaries after treatment, while females were more likely to engage in prostitution/ procurement after treatment. Longer lengths of stay in drug treatment reduced the likelihood of each kind of criminal behavior, although the relationship was less robust for burglary; outpatient nonmethadone treatment also predicted lower larceny rates relative to inpatient treatment. Drug use during treatment increased the odds of post-treatment criminality for three out of the four crime types analyzed, and return to treatment was moderately associated with criminality during the same post-treatment period.
|
|
|
 |
Click to Return to OAS Home Page |
|
This page was last updated on August 15, 2003.
|
|
