Monday, April 8, 2013

Do endorsements actually work?

When political scientists make claims that party insiders are influential in primary or general elections (and we do that from time to time), there's an assumption built in there that endorsements matter. Endorsements are the primary means by which an influential person or group can actually be influential. But we don't have very many great tests of the power of an endorsement.

Thad Kousser, Eric McGhee, and I are presenting a paper (PDF) later this week at the Midwest Political Science Association conference that uses several approaches to measure the impact of a party endorsement on voters. We take advantage of a convenient natural experiment provided by the state of California, which switched to a top-two primary system for the 2012 election cycle. In response to this very open primary system, the parties issued endorsements for their preferred candidates in state legislative and congressional races in the first-round election, and the state publicized those endorsements in the ballot booklet it mailed to voters.

Our first approach was to conduct a survey of 1,000 Californians right around the time of last June's primary. We briefly described to the respondents three fictitious candidates: a "traditional" Democrat, a "business" Democrat, and a Republican. We then randomized an endorsement by the Democratic Party between the two Democrats, and asked voters who they preferred. The endorsement made a big difference, particularly for the traditional Democrat, whose vote share increased by around 10 points when respondents heard that the party backed him.

We attempted a second approach by examining the actual results of the June 2012 election, looking to see whether endorsed candidates out-performed others. We were aided by a data collection consisting of the internal party endorsement votes conducted within local Democratic party committees. This way, we could see if candidates who had a lot of support within the party but fell just short of the endorsement threshold did notably worse than those who just passed the threshold. (This form of analysis is known as regression discontinuity.) Some results can be seen in the figure below. The internal party endorsement votes are depicted along the horizontal axis (with zero the point at which the endorsement occurs), and the primary vote share lies along the vertical axis.

What the graph suggests is that the endorsement mattered. Candidates who just barely eke out endorsements are presumably very similar to those who just fell short of it, but the former got a substantially larger chunk of the primary vote than the latter. The effect looks even larger than ten percent here.

What this means is that parties are more than just cheerleaders for particular candidates; they actually shape events. The reason party insiders are usually happy with the sorts of nominees they get is because they play a large role in picking them. 


  1. Not too surprising. Candidate sourced information is often too equivocal or managed to make it a reliable way to judge a candidate (especially intraparty, where you can't rely simply on a party identification to distinguish candiates). Endorsements frequently say much more.

    On the other hand, the difference between a party endorsement in an open primary and an individual endorsement in an ordinary candidate race is huge - indeed, the use of the word "endorsement" to describe the former is almost misleading.

  2. I have a paper (as yet unpublished, alas) on congressional primaries between 1990 and 1998 that shows very similar results. Controlling for other factors, having the support of party leaders (rather than just formal endorsements as such) is worth about 6 points in a contested primary without incumbents.