1

I was performing a power analysis of articles published in a journal of management using the pwr package in R. However, it seemed to be impossible to compute power for small, medium and large effect sizes for multiple ordinal logistic regression. I have tried using G*power, but it only seemed to be useful when we have simple logistic regression output. Thus, I have tried to simulate to calculate the power based on the answers by @GregSnow and @gung here: Simulation of logistic regression power analysis - designed experiments. How can I get power for small, medium and large effect sizes in multiple ordinal logistic regression?

mdewey
  • 16,541
  • 22
  • 30
  • 57
A. B. Bonache
  • 65
  • 4

1 Answers1

3

Simplified notions of effect size just get us into trouble, and there are no cutoffs that work universally. I usually specify an odds ratio and distribution for a predictor for ordinal regression. For the proportional odds case, power of a simple unadjusted 2-sample comparison can be computed using the R Hmisc package popower function.

Frank Harrell
  • 74,029
  • 5
  • 148
  • 322
  • Thank you very much for this helpful answer. I have solved my problem. Using G*power, the main point is to compute R². The answer resides in the correlation matrix: this matrix permits to calculate the R² among the X. To determine X distribution, the descriptive statistics of the article may be helpful. Hope this help people! Thank again! – A. B. Bonache May 12 '16 at 17:41