This representativeness metric𝑅𝑣(𝑔, 𝑐) for group g, comment c and vote v estimates how much more likelyparticipants in group g are to vote v on said comment than those outside groupg.

This results from the first step in PCA analysis being the “cen-tering” of the data matrix, by subtracting the mean of each column from everyentry in that column, leaving a row corresponding to someone with no voteswith mostly 0 entries.

To correct for this, a simple procedure is applied thatscales the projected positions of participants by the factor √𝐶/𝐶𝑝, where C isthe total number of comments, and 𝐶𝑝 is the number of comments voted on byparticipant p.1

. Participants who voted on fewer than seven com-ments are removed from the conversation to avoid the “clumping up” of partic-ipants around the center of the conversation.

Missing values,corresponding to comments the participant in question did not see, are im-puted by taking column-wise means of the non-missing values associated withthe given comment, a common method of dealing with this issue in downstreamanalyses (Dray & Josse, 2015

As participants vote, a vote matrix is derived, where rows correspond toparticipants, and columns to comments.

keeps its habitable zone much closer than larger, hotter stars like our Sun do