A NECESSARY CONDITION FOR A SOCIAL EXPERIMENT IS THAT THE AMELIORATIVE TREATMENT IN QUESTION BE IN SHORT SUPPLY, A CONDITION WHICH GUARANTEES THAT THERE ARE MORE EXPERIMENTAL UNITS (PERSONS, SCHOOLS, CTIES, ETC.) AVAILABLE FOR TREATMENT THAN ARE GOING TO BE TREATED. THE BEST EVALUATION STRATEGY UNDER SUCH CIRCUMSTANCES IS A TRUE EXPERIMENT IN WHICH THERE IS RANDOM ASSIGNMENT OF UNITS ACROSS THE FULL RANGE OF ELIGIBILITY FOR THE PROGRAM IN QUESTION. HOWEVER, THERE ARE MANY SITUATIONS IN WHICH RANDOM ASSIGNMENT OF ELIGIBLE PERSONS IS NOT REGARDED AS ACCEPTABLE. SUCH SITUATIONS ARE COMMON: AN AMELIORATIVE PROGRAM IN SHORT SUPPLY; A BASIC DECISION THAT THE PROGRAM'S SERVICES SHOULD GO TO THE MOST NEEDY (AS OPPOSED TO BEING DISTRIBUTED ON A FIRST-COME, FIRST-SERVED BASIS); AND A DESIRE TO SEE THE PROGRAM EVALUATED SCIENTIFICALLY. IN SUCH SITUATIONS, THE REGRESSION-DISCONTINUITY DESIGN -- A QUASI-EXPERIMENTAL ANALYSIS -- IS USEFUL. THE USE OF THE DESIGN REQUIRES A QUANTIFIED ELIGIBILITY CRITERION AND OTHER SPECIAL ADMISSION PROCEDURES AND CONTROLS. THIS ADDITIONAL PROCEDURAL BURDEN USUALLY EQUALS OR EXCEEDS THAT IMPOSED BY RANDOMIZATION FROM A POOL OF ELIGIBLE APPLICANTS LARGER THAN THE PROGRAM'S CAPACITY. RANDOMIZATION IS PREFERABLE ON GROUNDS OF BOTH STATISTICAL EFFICIENCY AND THE NEED FOR FEWER BACKGROUND ASSUMPTIONS. WHERE RANDOMIZATION IS PRECLUDED, THE REGRESSION-DISCONTINUITY DESIGN IS RECOMMENDED. DETAILS OF REGRESSION-DISCONTINUITY ANALYSIS AND EXAMPLES OF ITS APPLICATION IN PROGRAM EVALUATION ARE PROVIDED REFERENCES ARE INCLUDED. (AUTHOR ABSTRACT MODIFIED--LKM)