Reporting PGI results: types of metrics for prediction performance

Common & rare variant association
1

When reporting in a paper PGI results for a set of target phenotypes that are both continuous and binary we would have two different metrics of prediction performance (incremental-R2 and nagelkerke-pseudoR2, respectively).

If we want to report the PGI results for all these phenotypes within one figure, my questions are:

  1. Is there an overall term that englobes both metrics (i.e. for the Y axis of such plot)?
  2. Is it even a good idea to combine different metrics in one plot, with this I mean, are these two metrics comparable? Or should we report instead a different metric that is neither incremental-R2 nor nagelkerke-pseudoR2?
  3. I have also seen in some papers other metrics such as McFadden’s R2, is there a preference to report a specific metric over the other? Do you know of any blog or paper that further explains these different types of prediction performance metrics in PGI analyses.
2

I would have separate plots for those, they are not really comparable metrics. Another way I like to present prediction results for binary phenotypes is plotting the prevalence of the trait for PGI deciles like in here:
https://www.nature.com/articles/s41588-022-01016-z/figures/2

I think whether to use McFadden or Nagelkerke is a personal choice. McFadden will typically be lower than Nagelkerke and the interpretation is slightly different, but I wouldn’t necessarily prefer one over the other. Here’s a good blog post about this if you’d like to read a detailed discussion:
What’s the Best R-Squared for Logistic Regression? | Statistical Horizons