QM Speaker: Nataly Beribisky

When:
November 25, 2019 @ 10:00 AM – 11:30 AM
2019-11-25T10:00:00-05:00
2019-11-25T11:30:00-05:00
Where:
Endler Room (BSB 164)
4700 Keele St
North York
ON M3J 1P3
Cost:
Free

Title: What do I need to know to estimate sample size? An introduction and re-examined use of precision-based power analysis.

Abstract: Traditional sample planning procedures such as power analysis still heavily rely on the dichotomous decision-making used within the null hypothesis significance testing (NHST) framework. Specifically, the procedure estimates the sample size required to detect the presence or absence of an effect, rather than plan for the width of an interval. In contrast, precision-based power analysis allows researchers to estimate how many participants they need to detect a confidence interval of a given size. This may be advantageous when researchers do not know a minimally meaningful effect size (which is required for traditional power analysis) or are at different phases of the research process (e.g., confidence intervals may be wider for novel research but narrower for confirmatory studies). After discussing the MBESS R package that allows users to perform a precision-based power analysis, this talk explores whether certain difficult-to-know parameters are necessary to implement precision-based power analysis effectively. Specifically, the present study uses simulations to investigate how much of the variability in sample size can be attributed to different arguments (e.g., confidence interval width, effect size) within three functions in the MBESS package (standardized mean difference, standardized regression coefficient, RMSEA). The results differ across the three functions but suggest that, at least for the standardized mean difference, the effect itself need not be specified to conduct the procedure efficiently. The preliminary results of this study suggest that precision-based power analysis may be a more informative and accessible tool than its traditional counterpart.