Abstract
Ethical research with experimental systems (animals or humans) requires a rationale for the number of subjects to be included in a study. Standard methods for estimating sample size are not fit-for-purpose when the experimenter cannot predict the effect size/outcome with any certainty. These types of studies are often designated "pilot study"; however, there are few guidelines for sample size needed for a pilot study. Here we seek to address this issue. Concerning survival analysis it is noted that the experimenter can adjust the parameters of the experiment to improve the power. We propose that the experimenter needs to consider the "limit of interest" needed to represent an effect that the experimenter would be prepared to defend in terms of scientific or medical interest. Conventional power analysis is then used to estimate the n to deliver an alpha (false positive rate) of p < 0.2. This approach provides a balance that can inform a future study, demonstrate a strong effect or dismiss if no effect was observed. Where weight change or infection burden is considered, parametric analysis can be used. Here the main requirement for the pilot study is to establish a meaningful estimate of variability for subsequent power analysis. When considering the confidence intervals for standard deviations, it can be noted that a turning point is reached for n of four to six, beyond which we observe diminishing returns, suggesting that sample sizes should be greater than four. Finally, we discuss both the importance in statistical blocking and repeated measures in maximising the usefulness of the pilot study; and the importance of considering and outlining analysis techniques prior to performing the experiment. These findings are intended to be useful in the design of experiments in further prospective research.