Below are worked examples that provide an overview of models and illustrate how the results are calculated once you click "Calculate Results" at the end of the seventh step (“Analysis”) of the stepped wedge group-randomized trial (SWGRT) sample size calculator. In addition, a document illustrating example calculations for the variance of continuous, dichotomous, and count outcomes is provided.
Worked Examples for Detectable Difference
The six worked examples reflect the six primary variations that are addressed in this sample size calculator. These six variations are a function of three analytic approaches (cross-sectional, open cohort, closed cohort) and two correlation structures (block-exchangeable, discrete-time decay) for each analytic approach.
Separately, users make selections for the: Type I error rate; power; expected distribution of the primary outcome; number of time periods; intraclass correlation within a time period; cluster and/or individual autocorrelations; number of members or participants per group or cluster; expected benefit, if any, from regression adjustment for covariates; and magnitude of the intervention effect. For discrete-time decay structures, users may also adjust autocorrelation estimates (if warranted). For open cohort designs, users will also enter an estimate of the churn rate. These choices provide the parameter estimates used in the formulae but do not dictate the structure of the formulae.
Clicking any of the six links below will take you to a PDF that presents the formula for the detectable difference for that variation and walks you through an example, showing how the detectable difference is calculated. You can open these files in any combination, save them, or print them, as they may be of help to you.
- Cross-Sectional Design
- Closed Cohort Design
- Open Cohort Design
Worked Examples for Outcome Variance
The distribution of the outcome – continuous, dichotomous, or count – can determine its variance. Clicking the link below will take you to a PDF that presents how outcome variance is calculated based on user inputs as well as example calculations.