Open RossKen opened 1 year ago
Potentially relevant image from this Slack discussion (including other recommendations for this PR)
In addition to the m, u and match weight charts, there used to be a "proportion of comparisons" chart showing how all comparisons were distributed among the levels. This shows where levels rarely or never appear, even if their model parameters appear sensible, or explaining why model parameters aren't sensible (#1434)
Is your proposal related to a problem?
Linked to #1442 and #1434
It would be helpful to have a quick reference to the number of pairwise comparisons considered in each comparison level in the match weights chart. This could be particularly useful when dealing with model training issues as a result of small samples.
Describe the solution you'd like
Add the count and percentage of pairwise comparisons considered by each comparison level.
Describe alternatives you've considered
Additional context