This program estimates statistical power for testing hypotheses about categorical moderating effects using moderated multiple regression (MMR) using a theory-based approximation. It can be used for moderators with between two and 10 levels (e.g., subgroups) and it also allows you to specify variable reliabilities as well as predictor range restriction (i.e., truncation), making it the most comprehensive, and therefore recommended, power estimator.
INSTRUCTIONS: Specify the number of subgroups (i.e., moderator-based categories), desired a priori Type I error rate (e.g., .05., .01), and whether you wish to conduct the power analysis based on inputting observable (i.e., as observed and affected by measurement error) or true (i.e., for relations between constructs at the theory/conceptual level and free of measurement error) SDs and correlations. Then, the program shows an input table where you will enter information for each moderator-based subgroup. Regarding truncation (i.e., range restriction) values, this is related to the proportion of population scores included in the sample. For example, .75 means 25% of the population scores are included, and .10 means that 90% are included (if you enter 0, the power calculation assumes no range restriction). After clicking the "Calculate Power" button, the program displays the information you entered (so you can confirm that all the values you entered are accurate) and the statistical power estimate. By entering different values in the input table (e.g., increasing sample size, changing the truncation proportion), the program shows you interactively and in real time how statistical power is affected by research design and measurement characteristics. You can use this information to design studies that have sufficient power to detect existing moderating effects. Also, by entering values reported in published research, the program allows you to learn whether those studies, especially the ones that did not find a hypothesized moderator, have been conducted with insufficient statistical power—and this may be the reason for the lack of support for a moderating effect hypothesis.
Source for program's algorithms: Aguinis, H., Boik, R.J., & Pierce, C.A. (2001). A generalized solution for approximating the power to detect effects of categorical moderator variables using multiple regression. Organizational Research Methods, 4, 291-323. [pdf]
