Stats on a "need-to-know" basis

As covered in my textbook's Chapter 3 (biostatistics), one needn't know very many details to select and perform most statistical tests. Often based on prior studies involving similar variables, it is possible to project the likely character or distribution of the data in a trial. When doing so, we typically need to know whether data are likely to be: 1) categorical, or continuous; 2) paired or unpaired (independent); and 3) normally or non-normally (e.g. skewed) distributed. What about projecting a sample size for a future study? Apart from alpha and beta thresholds--which are the proportions of false-positive (for alpha) and false-negative (for beta) results that we are willing to accept--and 1 - (minus) beta, which is the statistical power (often 0.80), we need to know only two properties: effect size and variance. "Shameless plug": this lesson and so many others come to you in an attractive and affordable textbook you can order at: