Power and money: Designing statistically powerful studies while minimizing financial costs.

Abstract

Adequate statistical power is increasingly demanded in research design. However, obtaining adequate research funding is increasingly difficult. This places researchers in a difficult position. In response, the authors advocate an approach to designing studies that considers statistical power and financial concerns simultaneously. Their purpose is twofold: (a) to introduce the general paradigm of cost optimization in the context of power analysis and (b) to present techniques for such optimization. Techniques are presented in the context of a randomized clinical trial. The authors consider (a) selecting optimal outpoints for subject screening tests; (b) optimally allocating subjects to different treatment conditions; (c) choosing between obtaining more subjects or taking more replicate measurements; and (d) using prerandomization covariates.

Understanding statistical power in the context of applied research

“…Such calculations need careful planning and should be based on a clear understanding of the range of design factors that influence power. Particular emphasis should be placed on reducing measurement error, the 'dose' of the independent variable and the proportions in which participants are sampled from or allocated to conditions (Allison et al, 1997;McClelland, 1997).…”

“…In analysis of variance (ANOVA) conservative post hoc tests such as Tukey's HSD are often chosen by default, even when the costs of Type I errors might be negligible and when alternative procedures are readily available (Howell, 2002). Power can also be increased with repeated measures designs in situations where large individual differences are expected (Allison et al, 1997). Repeated measures analyses can also be used where participants have been matched (e.g., where each participant in one condition is paired with a participant in another condition on the basis of potentially influential confounding variables such as age, body mass index or social class).…”

“…The higher this correlation the more powerful the test (Howell, 2002). A similar outcome can be achieved by using analysis of covariance (ANCOVA) to control for confounding variables (Allison et al, 1997). These latter approaches are particularly useful for applied research.…”

“…This is not the case. Staged screening can also be used to sample optimal or near-optimal proportions of participants (Allison et al, 1997;McClelland, 1997).…”

Incorporating Cost in Power Analysis for Three-Level Cluster-Randomized Designs

“…In psychology, methodologists have discussed optimal allocation and power analysis in generalizability studies and measurement designs with budget constraints (see Marcoulides, 1993Marcoulides, , 1997. In addition, psychology methodologists have discussed optimal allocation methods for many aspects of experimental designs such as the number of individuals to different treatment levels, and the number of measurements within individuals (Allison, Allison, Faith, Paultre, & Pi-Sunyer, 1997;McClelland, 1997).…”