Statistical Power of Educational Psychology Research in Japan

Abstract

Statistical power was calculated for the articles published in the Japanese Journal of Educational Psychology between 1992 and 1996. The median power to detect small, medium, and large effects was. 15,. 68, and. 96 respectively, and was higher than the results of previous power surveys, such as Cohen (1962). The power was different, depending on the research area and the research method. The power of tests of null hypotheses as research hypotheses was relatively high, but was not high enough to justify the use of nonsignificant results to support research hypotheses. Sample effect size was also calculated and used to examine Cohen's (1992) criteria of effect size. The correlation between sample effect size and Nwas negative, and the correlation between sample effect size and 1/Nwas stronger. Despite critical views on the limitations of power analysis, more attention should be paid to power as long as significance testing continues to be the major tool of data analysis.