Sample Size for Pilot Studies and Precision Driven Experiments

C. O. S. Sorzano,D. Tabas-Madrid,F. Núñez,C. Fernández-Criado,A. Naranjo,C.O.S. Sorzano
DOI: https://doi.org/10.48550/arXiv.1707.00222
2017-07-01
Applications
Abstract:Pilot studies are highly recommended in experiments with animals when little is known about the anticipated values of the mean of the variable under study, its variance or the probability of response. They are also recommended to test the feasibility of the animal model or the experimental technique. However, the sample size required for a pilot study has received little attention and typical sizes practically used go from 5 to 20 animals disregarding any statistical consideration. Pilot studies are a particular case of precision driven experiments in which the sample size is designed according to a desired precision. In this article we provide some statistical guidance on the selection of the sample size of a pilot study whose driving force is the accuracy desired for the determination of the unknown parameters. We provide formulas and design tables for the sample size when trying to determine the standard deviation of a population, its mean, the probability of a certain feature or event, the correlation between two variables, and the survival time before an event occurs. All the calculations performed in this article can also be freely performed on the web through the online calculator available at http://i2pc.es/coss/Programs/SampleSizeCalculator/index.html. An immediate consequence of our analysis is that typical pilot sizes (5-20) are normally result in a very small precision and that the researcher should be aware of this fact before carrying out a pilot study.
What problem does this paper attempt to address?