Incumbency: Better the Evil You Know

One of the most secure jobs in the United States is as a Justice on the Supreme Court. But being an incumbent senator comes pretty damn close. Look at this graph:

Sen. Cochran's line actually goes off the chart twice because no Democrat ran in 1990 or in 2002.

There's a consistent pattern here where the guy gets elected pretty narrowly and (maybe) suffers through a nailbiter of a re-election campaign, but after that he's pretty much set for life. (Food for thought: Sen. Patrick Leahy has spent more than half of his life--thirty-nine years of it--in the Senate.) It's known as the "sophomore surge", and it means that no matter how narrowly you win your first election, you're probably going to win the next one by a landslide. And the one after that. And the one after that, should you decide four terms as a U.S. senator is good for your health and happiness.

There are a lot of advantages of incumbency they'll teach you in introductory government classes--casework, name recognition, fundraising power, etc. (The causal relationship with the "hate Congress, love my Congressman" phenomenon is unclear.) So we set out to figure out--just ballparking here--how these advantages manifested themselves in the vote share. To determine exactly how much incumbency helped a candidate, we ran a regression of incumbency and the state's natural partisan lean. We tried three different definitions of "incumbency", labeled, fittingly enough, "Incumbency1", "Incumbency2", and "Incumbency3":

1. "Incumbency1" was the number of times the winner had been elected to the Senate before. A challenger win or an open race meant Incumbency1 = 0.
2. "Incumbency2" was the same as Incumbency1 with one difference: if the winner was a challenger defeating an incumbent running for re-election, it would take the negative value of the number of times the incumbent had been elected. For example, Sen. Mark Begich's Incumbency2 value in 2008 was -7 because he had defeated Sen. Ted Stevens, who had been elected to the Senate 7 times before.
3. "Incumbency3" was a binary variable that took on a value of 1 if the winner was the incumbent and a value of 0 if the winner was not an incumbent.

Interestingly enough, the strength of the correlation didn't improve all that much whether we used Incumbency1, Incumbency2, or Incumbency3, which suggests that incumbency matters--but the amount of incumbency doesn't matter as much as you would think. Whether you're a six-term incumbent or a freshman running for re-election, you still get the same obscenely high vote share. We suspect that the reason for this is that after you start winning by 30, 35 points, there's not much more of the vote you can win.

Because of this we think the definition that best comports with theory is Incumbency3: if only "yes, she is an incumbent" or "no, she is not an incumbent" matters, then the values of the incumbency variable should be binary. We get the following regression from Senate elections in 2004, 2008, and 2012:

*Regression tables with Incumbency1 and Incumbency2 available upon request. Put it in the comments section or just email me.

"DPVI" is a variable we made up: it takes the absolute value of the state's Partisan Voting Index and assigns it a negative sign depending on whether it leans in the direction of the winner of the seat or away from it (the acronym stands for Directional Partisan Voting Index). For example, Maine's PVI in 2008 (which was based on the results of the 2000 and 2004 presidential elections) was D+4, a lean of 4 points toward the Democratic Party. Since the winner, Sen. Susan Collins, was a Republican, Maine's DPVI in her election was -4: that is, four points away from her party. You can think of it as how much the state's partisan lean helped the winner.

What these numbers mean
What's most striking about the regression is the magnitude by which incumbency outweighs a state's natural partisan lean. The confidence interval is quite wide, but it suggests that only in the most unfavorable states does an incumbent get knocked off--which is exactly what has happened, historically. In 2004 former Senate Minority Leader Tom Daschle (D-SD), elected in 1986 to the seat once held by ultra-liberal Sen. George McGovern, didn't get re-elected in a state that had gone much redder since his first election. He went the same way as his colleagues in the graph up there: a narrow first win, and then easy re-elections thereafter, except for that one where he lost. 

This year's elections, however, seem to defy the traditional advantage ascribed to incumbency. In 2014 we have Democratic senators in danger in every state with a Republican-leaning PVI (Republican Sen. Susan Collins of Maine seems to conform to the prior trend). Incumbency appears to be helping Sens. Pryor, Landrieu, Hagan, and Begich only in the sense that their seats aren't sure losses, as the open races in South Dakota, West Virginia, and Montana are. (That's right; John Walsh doesn't count as an incumbent.) Perhaps it reflects a changing trend where, as the houses of Congress become better-sorted by ideology, voters begin to vote less for the person and more for the party.