Estimating sample sizes for repeated measurement designs

https://doi.org/10.1016/0197-2456(94)90015-9Get rights and content

Abstract

Formulas for estimating sample sizes that are required to provide-specified power for analysis of variance (ANOVA) tests of significance in a two-group repeated measurements design are presented and evaluated. Power and sample size requirements depend on the pattern of treatment effects and the pattern of correlations among the repeated measurements, as well as on parameters common to sample size estimation for cross-sectional comparisons of treatment effects in simple randomized designs. Simplifying assumptions permit generation of these numerous parameter estimates from predictions of the magnitude of the standardized “effect size” at end of trial and the single correlation between the baseline and endpoint measurements. Monte Carlo methods are used to verify the actual power of different tests of significance for treatment effects in repeated measurement designs using sample sizes estimated by the formulas. The sample size implications of different patterns of treatment effects, levels of correlation, and numbers of repeated measurements are evaluated.

References (12)

There are more references available in the full text version of this article.

Cited by (102)

  • Cost structure effects of horizontal airline mergers and acquisitions

    2020, Transport Policy
    Citation Excerpt :

    Table 2 illustrates the 19 M&As considered in our analysis, including information on the year of the transaction, the geographical regions involved, and the number of years under observation. To estimate the necessary sample size for our study we used an approach suggested by Overall and Doyle (1994) taking into account its longitudinal design. In our sample size analysis we assumed a range of statistical power levels (Cohen, 1992), two groups, 6 time-points, and a 95% probability level, resulting in a minimum required sample sizes of 30 firms for power level 0.70, 35 for level 0.75, and 40 for level 0.80.

View all citing articles on Scopus

This work was supported in part by grant MH32457 from the National Institutes of Health.

View full text