Regression Artifacts in Correctional Program Evaluations (From Corrections at the Crossroads - Designing Policy, P 27-47, 1981, Sherwood E Zimmerman and Harold D Miller, eds. - See NCJ-85837)

• NCJ Number: 85838
• Author(s): D McDowall; R McCleary; A C Gordon; M D Maltz
• Date Published: 1981
• Length: 21 pages
• Sale Source: Sage Publications, Inc
• Publication Type: Statistics
• Language: English
• Country: United States of America

Annotation

Regression to the mean in a one-group, pretest-postest quasi-experimental evaluation design used with subjects selected for treatment due to an abnormally high rate of deviant behavior will make the treatment appear more effective than it actually is.

Abstract

Study data were taken from an evaluation of the Unified Delinquency Intervention Services (UDIS) program of the Illinois Department of Corrections. An evaluation of UDIS conducted by the American Institutes for Research (AIR) claimed that the program had a powerful, long-term inhibiting effect on subsequent delinquent activity of chronic juvenile offenders. This study demonstrates that if a juvenile court judge selects only the 'worst' delinquents of a cohort for a correctional intervention, regression to the mean will guarantee a before/after difference. Also, if the delinquents are selected for intervention at a time when it appears that their behavior is worsening, regression to the mean will guarantee a before/after difference. While the analysis of the AIR evaluation data does not prove the existence of a regression artifact, there is a strong relationship between preintervention arrest rates and percent reductions, which is the precise relationship to expect if a regression artifact is operating. The likelihood of regression is so great in a before/after quasi-experiment, such an evaluation design should not be used for the evaluation of any correctional intervention. Tabular and graphic data and 10 notes are provided.