Using a proportional odds model instead of logistic model to improve power

[Elias_Eythorsson]

Professor Frank Harrell had previously suggested using popower from the Hmisc package. Despite reading as much as I could and trying to implement myself, I was unable to parse how to do this or interpret the answer. If anyone could build upon this suggestion I would be much obliged.

Sample Size Calculation for Ordinal Logistic Regression

Wow a lot of good stuff here. As in the R Hmisc package popower function I would compute the factor k such that the effective sample size for ordinal Y is the same as that of a continuous variable with kN observations. Then use a Richard Riley-type sample size calculation for continuous Y and apply the factor 1/k to it.