NCJ Number
81012
Journal
Evaluation Review Volume: 5 Issue: 6 Dated: (December 1981) Pages: 834-850
Date Published
1981
Length
17 pages
Annotation
The application of statistical power analysis (determining the probability that a significant effect can be found when an effect actually exists) to field experiments in which subjects 'trickle in' through a caseflow process is discussed.
Abstract
Experience with a national evaluation of a juvenile justice innovation (restitution) is presented to illustrate problems that may be encountered in experiments based on caseflow designs. Statistical power is the probability of reaching the correct conclusion when treatment produces an effect. The planning procedure for evaluation begins by stating desired levels of power and significance for a specific test statistic and the minimum size of the anticipated effect of the experimental treatment. Previous research can serve as a guide to expected effect size. When these factors are specified, the required number of cases can be calculated. By observing at least this number of subjects, the researcher fixes the probability of detecting a significant effect. Used in this manner, power analysis may indicate it is not feasible to observe enough cases to have a reasonable probability of detecting a significant effect, or that it is not worth the effort to observe the larger number of cases required to provide a reasonable probability of finding a small effect. Statistical power analysis can also be used to determine how long the experiment will run to provide a viable study; decisions about the analysis design in relation to unequal group sizes, such as inefficient test, correlated factors, and attenuated coefficients due to skewed independent variables; and the decision to discontinue random assignment or stop the experiment. Tabular data and 16 references are provided.