You're obviously a glass-half-empty person - maybe. That means there's room for more beer though...

More seriously though it could be argued that the data-filtering & reduction makes the dataset more statistically significant as only songs which have date tags in the MusicBrainz database are included, which could arguably be the more popular or well known songs. As it's a study of popular music this seems valid.

Be interesting to know what an analysis of the "less popular popular" half million songs would reveal though. Arguably this might be the more interesting side of pop? I suppose that was what I was trying to get at...