Module 7 Statistical Power and Design Diagnosands
Before we run a study, we would like to know whether a particular design has the statistical power to detect an effect if it exists. It is difficult to learn from an under-powered study, since it would be unclear whether a null result indicates that there was no effect or just that we failed to detect a non-zero effect that exists. A power analysis can help you improve your design and allocate your resources better; it may even help you decide against conducting the study.
In this module, we introduce statistical power, core approaches to calculating power through analytical calculations and through simulation, and how design features such as blocking, covariate adjustment, and clustering impact power.
7.1 Core Content
Statistical power is the ability of a study to detect an effect given that it exists.
Power analysis is something we do before a study. It helps you figure out the sample you need or what effects you can detect. It is an essential step in research design and helps you communicate your design.
Common approaches to power calculation:
Analytical power calculations (using a formula)
Simulations (for example, using DeclareDesign)
Covariate adjustment and blocking can increase power.
For clustered designs you need to take account of the intra-cluster correlation (the within-cluster variance relative to the overall variance).
Below are slides with the core content that we cover in our lecture on power. You can directly use these slides or make your local copy and edit.
You can also see the slides used in previous EGAP Learning Days:
7.3.1 EGAP Methods Guides
7.3.2 EGAP Policy Briefs and PAPs
Some examples of power analysis in designs: