I’ve been very impressed by the speed with which the Labour party put up full breakdowns on constituency Labour party preferences, MPs and MEPs’ voting records, and voting in affiliated organizations. I guess hiring ERS really does pay off.

I got very excited when I saw that MPs’ full preference listings are public. Preference data like this can easily be turned into a map of candidates’ positions in political space. This is called multidimensional scaling.

Unfortunately, a common problem cropped up. MPs haven’t really given their sincere preferences. Some did: there’s a group of 21 Milibandians (D) who graciously ranked all candidates (David>Ed>Balls>Burnham>Abbott). But others didn’t: the 34 Milibandians who gave their first preference to David, and ranked no other candidates.

This means it’s quite hard to interpret the results from multidimensional scaling as if they really revealed preferences. The group of Milibandians who gave a complete preference structure looks very far to the left in this sunflower plot, but the bloc of 34 — who probably share the same preference structure but didn’t share it — look far more centrist.

Maybe you might know of a multidimensional scaling routine that doesn’t treat missing orderings as indicating ambivalence — the one I’m using (SMACOF for R) doesn’t. I might try and reshape the data into paired comparison data, and see if I can get at it that way. But I’ve spent too long on this for today.