Yes, I will definitely post the data. Best to be fully transparent. Expect updates sporadically though, because I'm doing other things. For now, I prepared what could be considered a "sensitivity run", where I replaced two very "explanatory" posters in the initial PCA, DiscoBiscuit and Floki, with two other runner-up members (others with decent amounts of data), @
Red Herring and @
Beorn. (Note, based on how I have my input and output structures, this took all of 2 minutes to generate, the more difficult part is writing about it. To produce other results/data would take longer).
Some things to note on the results below:
First, ignore the absolute position of members in the graphs. The way PCA works, the actual location of members (i.e. lower right, lower left, etc.) isn't really important at all. Different groups of data will produce slightly different orientations of the principal components, but it isn't meaningful. Rather, focus on how closely groups members are to one another, and how far groups of members are from one another.
Second, a way to interpret the change in the results is to consider who was removed; i.e. DiscoBiscuit and Floki. Both those members were quite explanatory in the original data set. In other words, many groupings were defined at least in some part by how they felt about those two members, or, how those two members felt about them ("felt" ~ likes received and given). Now that they're gone, different groupings will naturally arise.
Third, remember that these groupings really are couched in recent political discussions on the forum. An analogy might be, imagine if I grouped the forum based on their responses to MBTI questions meant to discern where you fall on the T-F spectrum only. members would be grouped based on their relative T-F preferences, but it would say nothing about their I-E, N-S, or P-J preferences, so the groupings would naturally miss alot. INTP's and ISFJ's would be seen as closer than INTP's and ENTP's, for instance.
