DIX STATES THAT STUDIES WITH CAREFULLY MATCHED GROUPS OF PERSONS WHOSE COMMITMENT HAS BEEN SOUGHT, WHO HAVE BEEN DETERMINED TO BE DANGEROUS, AND WHO REFUSE TO SUBMIT TO TREATMENT VOLUNTARILY ARE NEEDED. PERSONS IN ONE GROUP MUST BE SUBJECTED TO INVOLUNTARY TREATMENT, WHILE PERSONS IN THE OTHER GROUP MUST SIMPLY BE LEFT UNTREATED. HE ACKNOWLEDGES THAT THESE STUDIES MAY INVOLVE SIGNIFICANT RISK OF PHYSICAL HARM, BOTH TO SUBJECTS AND TO MEMBERS OF THE COMMUNITY. ALTHOUGH THERE IS QUESTION AS TO WHETHER A DEFINITIVE EXPERIMENT TESTING THE ACCURACY OF SHORT-TERM PREDICTION IS ETHICAL, LEGAL, OR FEASIBLE, THE NEED TO COLLECT AS MUCH DATA AS POSSIBLE ON SUCH PREDICTION MADE IN THE CONTEXT OF EMERGENCY CIVIL COMMITMENTS IS URGENT. FOUR METHODS BY WHICH PRELIMINARY DATA CAN BE COLLECTED ON THE VALIDITY OF PREDICTING VIOLENCE IN EMERGENCY CIVIL COMMITMENT CASES ARE: (1) NATURAL EXPERIMENTS, TAKING ADVANTAGE OF EVENTS OVER WHICH RESEARCHERS HAVE NO CONTROL; (2) ASSESSING THE RELIABILITY OF PREDICTIVE JUDGMENTS; (3) CONDUCTING MODIFIED TESTS OF PREDICTIVE ACCURACY; AND (4) CARRYING OUT EXPERIMENTAL TESTS OF BORDERLINE CASES, REFUSING COMMITMENT ONLY FOR CASES IN WHICH THE PREDICTION OF VIOLENCE IS IN DISPUTE. NONE OF THE METHODS REPRESENTS A DEFINITIVE TEST OF THE ABILITY OF MENTAL HEALTH PROFESSIONALS TO PREDICT VIOLENCE IN EMERGENCY SITUATIONS. IF MENTAL HEALTH PROFESSIONALS CAN AGREE ON WHO THEY BELIEVE WILL BE VIOLENT AND IF A REDUCTION IN CONSEQUENCES FOLLOWING PREDICTION RESULTS IN AN INCREASE IN VIOLENCE, IT WOULD BE UNWISE TO EXPOSE SOCIETY TO THE RISKS POSED BY DIRECT EXPERIMENTATION WITH EMERGENCY COMMITMENT AS PROPOSED BY DIX. THERE IS REASON TO BELIEVE THAT SHORT-TERM PREDICTIONS MAY BE MORE ACCURATE THAN LONG-TERM PREDICTIONS. REFERENCES ARE NOTED. (DEP)