A 10-YEAR LONGITUDINAL FOLLOWUP DESIGN WAS EMPLOYED IN THE STUDY TO RECORD INFORMATION FROM FILES MAINTAINED BY THE WISCONSIN DIVISION OF CORRECTIONS ON 31 PREDICTOR VARIABLES. ALL PREDICTOR VARIABLES USED IN THE ANALYSIS OCCURRED PRIOR TO A JUVENILE'S FIRST RELEASE. TWO SAMPLES WERE UTILIZED, ONE CONSISTING OF 432 BOYS FROM ALL FIRST ADMISSIONS TO WISCONSIN SCHOOLS FOR BOYS IN 1965 AND THE OTHER OF 500 BOYS REPRESENTING ALL FIRST ADMISSIONS IN 1967. STATISTICAL PROCEDURES EMPLOYED IN DATA ANALYSIS INVOLVED PREDICTIVE ATTRIBUTE ANALYSIS, LOGISTIC REGRESSION, AND A BURGESS-LIKE PROCEDURE. THESE PROCEDURES WERE COMPARED BY EXAMINING DIFFERENCES BETWEEN EXPECTED FAILURE RATES (BASED ON RATES OBSERVED IN ONE SAMPLE) AND OBSERVED RATES (BASED ON RATES OBSERVED IN THE OTHER SAMPLE). PERSONS WERE CLASSIFIED WITH REGARD TO FAILURE AND SUCCESS IN ONE SAMPLE BASED ON THEIR PROBABILITY OF FAILURE, AND AN EFFORT WAS MADE TO SEE HOW WELL PREDICTIONS COULD BE MADE IN A SEPARATE SAMPLE. A JUVENILE RECIDIVISM RATE OF 60 PERCENT WAS FOUND, AS COMPARED TO AN ADULT RATE OF 35 PERCENT. SMALL BUT COMPARABLE TOTAL DIFFERENCES BETWEEN EXPECTED AND OBSERVED FAILURE RATES WERE OBSERVED WITH ALL THREE STATISTICAL PROCEDURES FOR JUVENILE AND ADULT RECIDIVISM. CAUSAL OR ASSOCIATIONAL VARIABLES IN RECIDIVISM WERE PRIOR CRIMINAL RECORD, AGE, DELINQUENT BEHAVIOR PATTERNS, FAMILY, RACE, SCHOOL, OFFENSE TYPE, WORK RECORD, MENTAL DISORDERS AND PERSONALITY, RESIDENCE, FAMILY MOVES, FRIENDS, AND SOCIOECONOMIC STATUS. IN ESTABLISHING RELATIONSHIPS AMONG THESE PREDICTOR VARIABLES, IT WAS DETERMINED THAT INSTITUTIONALIZED MALE DELINQUENTS ARE NOT A HOMOGENEOUS POPULATION WITH REGARD TO FURTHER CRIMINAL BEHAVIOR AS ADOLESCENTS OR AS ADULTS. POLICY IMPLICATIONS OF THE STUDY FINDINGS ARE DISCUSSED, AND SUPPORTING DATA ARE TABULATED. APPENDIXES CONTAIN THE STUDY QUESTIONNAIRE AND ADDITIONAL INFORMATION ON STATISTICAL PROCEDURES. (AUTHOR'S ABSTRACT MODIFIED--DEP)