Will the “Top-Two Primary” End Polarization?

by John Sides on March 24, 2010 · 2 comments

in Campaigns and elections,Political science

bq. Want to get serious about reducing the toxic levels of hyper-partisanship and legislative dysfunction now gripping American politics? Here’s a direct, simple fix: abolish party primary elections.

So says Phil Keisling in the New York Times. A recent USA Today story examines the topic as well. Unfortunately, it waits until the very end to cite the, oh, evidence about the effect of primaries on polarization:

bq. Not everyone agrees that changing the primary system will draw voters and centrist candidates. A study by the non-partisan Public Policy Institute of California concludes that the proposed new primary system would only have “a noticeable but modest effect.”

Here (pdf) is the PPIC study, authored by Eric McGhee. (NB: Eric is a friend from graduate school.) I want to delve into its conclusions, which suggest very limited effects indeed. These are particularly evident in the technical appendix (pdf) that accompanies the report.

Basically, Eric looks at the effect of primary rules on polarization in various contexts, leveraging variation across states to look at polarization in the US House and variation across time within states (e.g., when the CA blanket primary was in place for two election cycles) to look at polarization in state legislatures or House delegations. When you count up all the different models — across these different contexts, with different measures of ideological preferences and polarization, with different measures of primaries — there are 40 coefficients that indicate the effect of primary type on preferences. (My count from Tables 1, 2, 5, 7, 8, and 9 of the appendix.) How many of those coefficients are statistically significant?

Eight.

In other words, there is really no robust direct effect of primaries on polarization at all. Moreover, when those effects emerge, they are not very large. Consider the effects of primary type on moderation using one measure of ideology in the US House:

primariespolarization.PNG

At best, these forms of primaries might make Democrats and Republicans a few points more moderate, but that is a far, far cry from “reducing toxic levels of dysfunction.” Small wonder, then, that the appendix concludes:

bq. This study has examined the available roll-call evidence to determine the effect that open primaries have on representation. The results suggest that most of these systems have little effect on moderation.

To be sure, in the report itself, Eric is properly cautious about the effects of California’s proposed “top-two primary,” particularly over the long term. But the report still concludes that any effect is likely to be “modest.”

I’ll go further and define “modest” as: (1) not big enough to do anything near what the proponents of primary reform claim, and (2) not big enough to affect the day-to-day dynamics of policymaking in state and national legislatures.

In short, don’t expect primary reform to spawn a magic new race of centrist representatives.

{ 2 comments }

Matt Jarvis March 24, 2010 at 9:15 pm

BTW, just to keep piling on:
8 statistically significant coefficients out of 40 is even less than it seems, when randomly, if there were absolutely no effect, you still would have expected 2 of those to be statistically significant.

Martin Blom March 25, 2010 at 5:57 pm

Dude. You don’t need every candidate to have a moderate view. You just need one or two from each party’s moderate fringe to be more inclined to vote with the other guys to have a huge effect. The health care bill would have been “bipartisan” if a single republican had voted for it, and if a single republican had felt he had more to fear from the general election than the primary that could have happened.

And as Matt Jarvis points out, 8 out of 40 is a fourfold enrichment over random, ie 300% more than what you would expect by chance. “300% more” sounds much better than “only 8 in 40″ does … :P

Comments on this entry are closed.

Previous post:

Next post: