Statistical Power and Mixed Effect Models

Timothée Bonnet
June 1st 2018

What is power?

Probability to detect an effect that exists for real

= 1 - false negative rate (type II error)

Why should you think about statistical power?

Low power means:

  • Statistical tests non-significant whether an effect exists or not
  • Your results will be inconclusive
  • Data collection and analyses are wasted (maybe not completely, there is a twist later on)

High power means:

  • Test probably significant when an effect exists
  • Test rarely significant when an effect does not exist
  • You learn something about the world

Think about power early and late

BEFORE planning experiment or data collection:

  • Can I get enough data to answer my question? Is it worth it?
  • How to improve my chances of detecting an effect if it exists?

But also AFTER doing an analysis:

  • Is this non-significant result due to a lack of power or to the absence of an effect?
  • What is the maximal likely value of the effect? Is it still important biologically?

What power depends on

1. Sample size