Saturday, July 25, 2009

Numb3rs: Incumbency

After a quick look at competitiveness last week, I wanted to touch on incumbency effects today.

First of all, the technical part - if this bores you, just scroll down to the bolded line below. I won't think any less of you.

To see how important an incumbent is, we need a way to calculate an expected result. After spending a lot of time tinkering with various projection models, I've come to the conclusion that a simple arithmetic transfer of votes works the best - at least for the 2006 and 2008 election, which is what I'm looking at (if you want to see residual charts and regressions and such to prove this, just e-mail me). So I carved the country into 30 geographic regions and projected accordingly - if the Liberal vote dropped 2 points in Calgary last election (it did!), I subtracted 2% from the 2006 result from every Calgary riding.

How accurate is this? Well, half the results fall within +/- 2% of the actual result, with 90% falling within +/- 6%. The predicted values correlate 96.3% with the results. Keep in mind, we don't expect the results to be spot on because, after all, the candidates and campaigns should make a difference.

For those interested in the fine print, by election winners were considered to be incumbents, so Bob Rae was an incumbent MP last election. And Garth Turner became a Liberal incumbent, with Wajid Khan as a Tory incumbent (for what it's worth, Turner's impact was negligible, while Khan, not unsurprisingly, undershot the expected Tory vote by 3%).

So what does the data show?

For both 2006 and 2008, in ridings without an incumbent, the incumbent party took a 4.1% hit compared to their expected result. Since the totals have to sum to zero, that means incumbents performed about 1% above expected, for an overall benefit of around 5%. These results were largely identical both elections and mesh well with the academic literature on the subject so I don't have any reason to doubt them.

What's also interesting, as an aside, is that MPs were more likely to call in quits in regions where their party was heading for a drop in support (think of the Tory Newfoundland MPs before last election). So "spending more time with my family" loosely translates to "I can see the writing on the wall" quite often.

Now, I do think it's important not to read too much "cause and effect" into this. As Andrew Steele comments here (citing freakonomics...a book I adore), incumbency effects are often overblown once other factors are controlled for. And when we're talking about "incumbency", we're really talking about resources, organization, and "name recognition" - Gerard Kennedy and Justin Trudeau weren't incumbents last election but I'd wager a lot more people knew who they were in their home ridings than many obscure backbenchers (I'm looking at you Brian Jean). Quite simply, not all incumbents are created equal. But a 5% swing is nothing to sneeze at - 41 MPs were elected by smaller margins than that last time.

So, once MPs start announcing their retirement, we should be paying attention. Because if they're running in ridings with a slim margin of victory, it certainly could put those seats into play.

Labels: , ,

6 Comments:

  • Andrew's blogpost stimulated Professor Stanbury and I to examine the benefits of incumbency further, and calculate incumbent reelection rates in Canada and the United States between 1968 and 2008.

    We published it in an article in the Hill Times. The table with all the calculations can be found here. And we looked at inter-party differences in Canada here.

    Your work today actually answered the question commenter the Jurist posed on my April blogpost, which I never did get around to answering myself because I didn't have the complete dataset available in my relational database yet.

    By Blogger The Pundits' Guide, at 4:37 PM  

  • Cool post! I also wonder if the incumbency effect might be bigger for first-time incumbents. If my party does equally well in my region in 2008 as in 2006, and if I was first elected in 2006, I may well do better in 2008 since my name recognition increased a lot. But if I've been an MP for 20 years, I would probably not get that extra boost. Does your data show such an effect?

    By the way, I've just discovered your blog recently, and find it very enlightening. Keep up the good work!

    By Anonymous Election Watcher, at 10:27 PM  

  • "What's also interesting, as an aside, is that MPs were more likely to call in quits in regions where their party was heading for a drop in support (think of the Tory Newfoundland MPs before last election). So "spending more time with my family" loosely translates to "I can see the writing on the wall" quite often."

    Does this not suggest the possibility of simultaneous causality. Does incumbency drive better election results, or do better (prospective) election results encourage a candidate to stay on longer.

    Interesting to see that incumbency still matters a bit in Canada (relative to the US, I would tend to think that individual candidate traits will matter less in Canada).

    By Blogger hosertohoosier, at 11:05 PM  

  • I would expect that if you ran the numbers based on how many elections the incumbent has won, or how long they've been around, you'd find that matters too.

    By Blogger calgarygrit, at 3:04 PM  

  • Here are some controls that might get a more nuanced picture of how much incumbency matters.

    1. Experience vs. incumbency.
    Obviously these two are going to be very similar, but not the same. For instance, when Joe Clark ran in a Kings-Hants by-election in the 90's he was not an incumbent, but was a very experienced politician. Similarly, Lorne Nystrom was very experienced when he lost in 2006, but he was not an incumbent (he lost in 2004 as well).

    I can think of other "candidate quality" variables like "are they a good constituency MP" (hard to measure), or candidate education. Maybe it is the case that voters value high quality candidates and re-elect them. In other words, incumbency may be a stand-in for all the qualities that tend to make voters re-elect incumbents.

    2. As you suggest above you can examine incumbency as a binary variable, or look at the number of years. This would give a more nuanced picture.

    3. I think you may need to run an instrumental or two-stage least squares model to account for simultaneous causality. Switching away from a dichotomous variable for incumbency still wouldn't solve the problem, it would only change it a bit. Does incumbency help in re-election, or are candidates likely to be re-elected less likely to quit politics (or both)?

    4. If there was a clearer mechanism linking incumbency and election results it would also be helpful. For instance I could see the case being made that incumbency helps with name recognition. Opposing parties may be less likely to target incumbents because they believe (true or not) it will be more difficult.

    This is all just food for thought...

    By Blogger hosertohoosier, at 6:48 PM  

  • Have you had a chance to explore the less statistically derived aspect of incumbency: the situations where a (strong) incumbent loses the election to a newcomer?

    By Blogger paul.obeda@, at 5:36 PM  

Post a Comment

Links to this post:

Create a Link

<< Home