NCJ Number
92343
Date Published
1983
Length
14 pages
Annotation
Statistical power is a more important criterion than statistical significance in determining the substantive significance of research results and their implications for both decisionmaking and the design of new research.
Abstract
However, social science research tends to rely almost exclusively on tests of statistical significance. In addition, statistics textbooks generally give minimal attention to the subject of statistical power. Whereas statistical significance focuses on the null hypothesis and on determining whether or not an effect exists, statistical power focuses on the research hypothesis and on whether or not an effect is trivial or substantive. Working with statistical power is much the same analytically as working with statistical significance, but it requires a shift in focus. Statistical power increases with increasing sample size, with the size of the effect being sought, with the reliability of measurement of both the independent and dependent variables, with the degree of linear independence among the set of independent variables, and with the overall ability of a regression to account for the variance in the dependent variable. If a study has a low statistical power, it should be largely discounted. To design new research with a high probability of finding a substantively important effect if it exists, researchers should specify the minimum size of the effect considered to be substantively important, use prior research to estimate reliability values, specify the desired level of statistical power, and incorporate this information into the choice of sample size and other research features. Reanalysis of studies suggesting the existence of racial bias in sentencing calls into question these studies' conclusions. Four notes and 17 references are provided.