How do I determine the number of subjects needed for my study?

As explained in Section 2.11, sample size calculations should be done prior to performing your analysis. Nonetheless, I have placed this section after the sections on statistical analyses because you need to know what type of analysis you will be doing (e.g., chi-squared, t-test) to calculate the needed sample size.

For each type of statistical analysis you will need different elements (e.g., expected proportion, standard deviation) to determine the needed sample size. These elements are explained in the sections below. Once you have these elements you can determine the needed sample size in one of the three ways:

1. use published tables;

2. use the formula;

3. use a software program.

Using published tables and formulas is adequate for calculating sample size for descriptive studies (univariate analysis). However, for more complicated designs it is better to use one of the available software packages.

Three software programs that are available free are:

1. Power and Sample Size (PS) by Dupont, W.D. and Plummer, W.D. (http://www.mc.vanderbilt.edu/prevmed/ps).

2. Statistical Considerations for Clinical Trials and Scientific Experiments by Schoenfeld, D. (http://hedwig.mgh.harvard.edu/sample_size/quan_measur/defs.html).

3. Simple Interactive Statistical Analysis (SISA) (http://home.clara.net/sisa/sampshlp.htm).

All the three software programs perform comparisons of unpaired means (t-test) and paired means (paired t-test). Of the three programs PS performs the widest array of calculations including sample size calculations for linear regression, survival analysis, and matched case control studies with an option of specifying more than one control per case. It requires downloading the software onto your computer; the other two are web based.

Unfortunately, none of these three packages perform sample size calculations for multivariable analysis.