IT IS NOTED THAT MOST RESEARCH ON RECIDIVISM IS MARKED BY SERIOUS METHODOLOGICAL FLAWS, SINCE IT IS DIFFICULT TO CONSTRUCT COMPARATIVE GROUPS EQUAL IN EVERY RESPECT BUT THE METHOD OF TREATMENT, I.E., INCARCERATION VS PROBATION OR SUSPENDED SENTENCING. THE RESEARCH STRATEGY ADOPTED HERE WAS DESIGNED TO OVERCOME THIS DIFFICULTY BY IDENTIFYING TRULY COMPARABLE GROUPS OF OFFENDERS WHO WERE NEVERTHELESS SUBJECT TO DIFFERENTIAL TREATMENT. THE IDEA WAS THAT IF CASES ARE ALLOCATED AT RANDOM TO JUDGES SITTING IN THE SAME COURT, AND IF SOME JUDGES ARE HARSHER THAN OTHERS, INCARCERATING A HIGHER PROPORTION OF THOSE WHO COME BEFORE THEM, THEN DIFFERENCES IN THE SUBSEQUENT RECIDIVISM RATES OF THE GROUPS COMING BEFORE THE VARIOUS JUDGES CAN BE ATTRIBUTED TO DIFFERENCES IN THE INCARCERATION RATES. TO FACILITATE COMPARISONS, A TECHNIQUE WAS DEVELOPED FOR CALCULATING RECIDIVISM PROBABILITIES FOR OFFENDERS WHO WOULD BE INCARCERATED BY A HARSH JUDGE BUT NOT BY A LENIENT. DATA FOR THE STUDY WERE COLLECTED FOR FIVE HARTFORD SUPERIOR COURT JUDGES TO WHOM IT APPEARED THAT CASES WERE ALLOCATED IN AN ESSENTIALLY RANDOM MANNER. UNFORTUNATELY THE OFFENDER GROUPS COMING BEFORE THESE JUDGES WERE FOUND TO BE SOMEWHAT DISSIMILAR. HOWEVER, IT PROVED POSSIBLE TO CONSTRUCT A 'COMPOSITE' JUDGE WHOSE GROUP OF OFFENDERS WAS COMPARABLE WITH THE GROUP COMING BEFORE ONE OF THE HARSHER JUDGES. COMPUTATIONS BASED ON THIS COMPARISON REVEALED THAT THE RECIDIVISM PROBABILITY OF BORDERLINE OFFENDERS (THOSE WHO WOULD BE INCARCERATED BY THE HARSH JUDGE BUT NOT BY THE LENIENT) WAS CONSIDERABLY HIGHER AFTER INCARCERATION THAN AFTER NON-INSTITUTIONAL TREATMENT. THIS CONCLUSION WAS QUALIFIED BY THE FACT THAT RELATIVELY SLIGHT VARIATION IN INCARCERATION RATES MADE THE COMPUTED VALUES QUITE UNSTABLE. (AUTHOR ABSTRACT MODIFIED)