Tim Besley sent me this paper by himself and Ian Preston on the relation between bias in an electoral system and the policies of political parties. They write, “The
results suggest that reducing electoral bias leads parties to moderate their policies.”
Their model goes as follows: they start with a distribution of voter positions (in a one-diemensional ideology space) in each district, which, in combination with party issue positions, induces a two-party vote result in each district and thus a seats-votes curve. Properties of the seats-votes curve correspond to properties of the distribution of voter ideology distributions across districts.
They then run regressions on a dataset of local governments in Britain and find that, in the places with more electoral bias (thus, one party having a lot of “wasted votes” and the other party winning districts with just over 50% of the vote), the parties take on more extreme positions. The first way I would understand these results is geographically: areas with high bias must have the voters of one party or the other highly concentrated in districts. I’d like to know what these areas look like: are they inner cities favoring the Labour Party? Suburbs and rural areas favoring the Conservatives? I don’t know enough about U.K. politics here, but I’d like to see a more detailed local analysis to complement the theoretical model. In particular, is it the districting, or is it just that, in the areas in which it is natural for districtings to be biased, the voters as a whole are more extreme? To start with, I’d think they could account for some of this by considering %vote for the parties as a predictor in their model.
Picky, picky
Tables 1 and 2 should be presented graphically. I mean, are numbers like “0.3152” of any relevance whatsoever?
Table 3 should be presented graphically, as data and fitted lines. Tables 4-8, I don’t even want to think about! but at the least you could do some rounding… I know this sort of presentation is standard but I don’t like it (even on the occasions that I’ve done it myself!) See this paper for some suggestions on how to do this sort of thing graphically.
Figure 1 is mysterious. What is each dot, what are parties a and b? Also, why does the x-axis extend so far in both extremes? Figure 2: similar comments. Figure 3: simply unreadable. Also, it’s completely inappropriate to list the districts alphabetically. It would be much better to list in increasing support of one of the parties.
(Also, of course, I’d suggest they use the JudgeIt estimates (as in Gelman and King, 1994) of bias and responsiveness, since these will be more stable than the between-year regression estimates that they use. The between-year regressions require much less in the way of raw data, however, so I can see why they went that route.)