A better way to do this analysis would have been to create an extremely sparse matrix with one column for every possible endorsement category with the value being the number of endorsements (normalized). Then try to predict various aspects of coding performance.

Definitely wouldn't endorse the author for machine learning :)