NCJ Number
14172
Date Published
1974
Length
117 pages
Annotation
'REGRESSION ARTIFACT' IS THE NATURAL TENDENCY FOR THOSE SUBJECTS SELECTED AS MOST DEVIANT ON AN INITIAL MEASURE TO AVERAGE NEARER TO THE MEAN ON A SECOND MEASUREMENT.
Abstract
THE IMPLICATIONS OF THIS EFFECT FOR PROGRAM EVALUATION ARE THAT THE MERE COMPARISON OF A 'BEFORE' MEASURE TO AN 'AFTER' MEASURE MAY NOT ALWAYS BE A TRUE INDICATION OF THE EFFECTIVENESS OF A PROGRAM. IGNORING THE ARTIFACT STIUATION CAN CAUSE A PROGRAM TO LOOK GOOD WHEN IT MAY ACTUALLY HAVE HAD NO EFFECT WHATSOEVER ON THE TREATED POPULATION. HOWEVER, A REGRESSION ARTIFACT MAY BE PREVENTED IF A GROUP IS ORIGINALLY CHOSEN BY A RANDOM PROCESS, EVEN THOUGH THE GROUP MAY TURN OUT TO HAVE AN EXTREME MEAN. THIS STUDY DEVELOPS ANALYTICAL TECHNIQUES FOR ESTIMATING THE MAGNITUDE OF THE ARTIFACT. AS A RESULT, IT ALSO DETERMINES A MORE ACCURATE MEASURE OF THE EFFECTIVENESS OF SUCH PROGRAMS. FOR SIMPLE CASES, ORDER STATISTICS ARE USED. A COMPUTER SIMULATION WAS DEVELOPED FOR MORE COMPLEX SITUATIONS. USE OF THE ANALYTIC TECHNIQUES TO STUDY AN ACTUAL EVALUATION SITUATION, THE ST LOUIS POLICE FOOT PATROL PROJECT, SHOWS BY WAY OF A SPECIFIC EXAMPLE THAT THE REGRESSION ARTIFACT IS ALMOST CERTAINLY RESPONSIBLE FOR SOME OF THE APPARENT CRIME REDUCTIONS ATTRIBUTED TO THE PROJECT. (THIS PROJECT WAS DESIGNED AND IMPLEMENTED TO DETERMINE THE CRIME REDUCTION EFFECT OF INTENSIVE POLICE FOOT PATROLS IN HIGH CRIME AREAS.) THE VARIABLES USED IN THE COMPUTER PROGRAM APPEAR IN THE APPENDIX. A BIBLIOGRAPHY IS PROVIDED.