Multivariate Methods and Small Sample Sizes

摘要

Behavioural biologists face an important choice at the onset of any research project: what sample size should be used to address the questions being posed. The sample size selected – regardless of the research question being asked or the statistical approach used – necessarily impacts the precision of parameter estimates (e.g. means or effect sizes). In addition, for conventional null hypothesis testing, sample size also affects the power available to reject a null hypothesis. Unfortunately, behavioural researchers are often constrained in the sample sizes available. In a recent review, Taborsky (2010) found that behavioural studies most often had sample sizes likely determined by logistical constraints rather than power and precision. Sample sizes necessarily impact the conclusions drawn from data and small sample sizes may thus lead to improper inferences, most typically a failure to identify biological differences that are present but have small effects (Taborsky 2010). These concerns are particularly important with multivariate analyses where large sample sizes are often considered necessary (Budaev 2010).