The Limited Explanatory Power of the Median Voter Theorem
Tyler Cowen thinks people should talk more about the median voter theorem when trying to understand what’s happening in politics. I disagree. I mean, admitted the phrase “median voter theorem” is almost never used in newspapers, but I think the conventional political media wisdom tends to drastically overstate voter preferences as an explanatory variable. Andrew Gelman also has his doubts about the importance of the median voter theorem, citing data which indicates that a moderate voting record only modestly boosts your ability to win elections:
The United States Senate also provides a convenient test of this idea, since it offers up a series of pairs of politicians who are accountable to identical groups of voters. But when you look at opposite-party senate pairs, you don’t see a great deal of similarity in their voting records. Or to look at it another way, everyone knows that the voters in Louisiana are more liberal than the voters in Iowa. And this helps explain why Mary Landrieu is more conservative than Tom Harkin. But Mary Landrieu isn’t more conservative than Chuck Grassley. The ordering of those states’ four Senators, from left to right, is Harkin (D-IA) then Landrieu (D-LA) then Grassley (R-IA) then Vitter (R-LA) — the partisan affiliation of the senator tells you more about their voting behavior than does knowledge about the electorate they represent. Indeed, in terms of DW-NOMINATE exactly zero Senate Democrats in the 110th or 109th Senates compiled a voting record more conservative than that of the leftmost Republican (first Chaffee then Olympia Snowe) even though in both cases many Democrats represented states whose median voters are more conservative than the median voter in Maine or Rhode Island.
Part of the answer here is presumably that it’s relatively rare for an incumbent American politician to actually face a competitive re-election bid. Incumbent Senators do frequently form cross-party alliances to defend parochial local interests, for fairly obvious reasons, so it’s not as if people are behaving recklessly with their careers. Helping out local influentials is presumably not only (locally) popular, but actually helps ward off challengers. I would also note that it’s not obvious to me how an incumbent politicians would really go about assessing median voter sentiment—issue polling is legendarily unreliable and subject to massive framing bias.
17 Responses to “The Limited Explanatory Power of the Median Voter Theorem”
Register
RSSLogin
Topic Cloud
ArchivesBlog RollAbout Matt YglesiasContact Me
Donate
June 23rd, 2009 at 1:22 pm
Also you should be aware of Larry Bartels’ research indicating that legislator behavior is only really influenced by the opinions of the wealthiest 20% of their district.
June 23rd, 2009 at 1:31 pm
I agree partisanship seriously complicates this question, but I also think it means we may need to break out different partisan setups and then rerun Gellman’s analysis. In other words, it wouldn’t surprise me if appealing to median voters was more important for candidates who need to attract a lot of independents and crossovers to win, so mixing them up with candidates in “safe” states and districts may obscure this effect.
June 23rd, 2009 at 1:37 pm
everyone knows that the voters in Louisiana are more liberal than the voters in Iowa.
We sure do!
Nevada is a good example of this: the same voters elected Reid and Ensign, who are extremely far apart in their voting records and yet likely both safe, scandal notwithstanding. And in South Dakota, voters replaced the Democratic Majority Leader with the hardline conservative Thune, who they’ll presumably keep electing along with the mainline Democrat Johnson. I think the key factor here is what you said: that Senators rarely face competitive reelections.
June 23rd, 2009 at 1:43 pm
I think the median voter theorem really is saying that moderate policies get enacted, rather than that moderate politicians get elected. I think the evidence largely bears this out…we’ve had some leftward and rightward shifts in America over the years, but overall the policies we end up with have been very moderate.
June 23rd, 2009 at 1:44 pm
Voters have a hard time telling the ideologies of the people they are voting for, but if someone like Jeff Sessions ran for senate in New York and didn’t completely remake himself ideologically, he would loose. The median voter theorem would provide a pretty clear reason why.
June 23rd, 2009 at 1:45 pm
everyone knows that the voters in Louisiana are more liberal than the voters in Iowa
Really? I was born in louisiana and i went to school there and I don’t know this. I think it would be news to most other louisianans as well since they pretty strongly supported republican presidents in the last three elections. I don’t think we can say the same for Iowa.
Not that I dispute the overall gist of the post – but Louisiana is an example of the power of incumbency and the extreme variability of what it means to be a democrat. I suggest you avoid the inner david brooks that’s enticing you to make broad statements about what everybody knows.
June 23rd, 2009 at 1:50 pm
I disagree that the median voter theorem isn’t a huge factor here – only, I think the power of the idea is that SENATORS THINK (erroneously, perhaps, but nevertheless) that being more moderate will help them significantly, and also believe (again, erroneously) that the median voter is somewhat right-leaning when it comes to issues like healthcare and the environment. Or, at the very least, they don’t think the median voter is a straight-up progressive (which, on some issues, he/she probably is).
One thing Matt didn’t note, that I think the chart is telling us: the variance in decades helps explain the beliefs and behavior of Senators elected in the 1980s and 90s as compared to newer Democratic Senators -
In the 80s/90s, there was a small be significant (4-5%) advantage to being a moderate – in a swing state, that might be the difference between winning and losing. In addition, in the 80s/90s (and really until about 2005), what constituted a “median voter” meant something a bit farther right (and white) than what constitutes a “median voter” in the Obama era.
The political instincts of many Senators (especially committee chairs!) were formed during that era, and although they are no longer totally in touch with reality, they still drive a lot of the debate inside the beltway.
June 23rd, 2009 at 2:03 pm
Matt obviously meant that voters in Louisiana are more conservative than voters in Iowa. Read the next sentence!
June 23rd, 2009 at 2:13 pm
All due respect to Tyler, but the MVT is probably less useful when dealing with elected representatives in a bargaining democracy than it might be in any other circumstance. The importance of the swing voter in the US has everything to do with log rolling and building a coalition of factions, but this has more to do with the people electing the representatives than with the representatives themselves. A truism is that the party which can attract the most factions (special interests) and hold them together is the one that holds power. This fact explains why we see the majority of policies enacted poll in at less than 50%, yet find that most incumbents are re-elected time after time.
-AM
June 23rd, 2009 at 2:22 pm
This argument is like pointing to transaction costs as a way of refuting the Coase Theorem. But, of course, the Coase Theorem is just a prediction of what will happen in the absence of transaction costs. Similarly, the median voter theorem is a prediction that election outcomes will reflect the preferences of the median voter if the voters’ preferences can be described by a utility function with a single peak. To get rid of the jargon, that’s just to say that the median voter theorem should work only if voters’ preferences can be put on a spectrum, and a voter will always prefer a candidate whose views are closer on the spectrum to his/her own to one whose views are farther away. That kind-of, sort-of describes the preferences of American voters, but at best imperfectly so. Americans’ ideological preferences can’t be described by a single, simple left-right axis, and, of course, candidates’ charisma and fundraising capabilities will influence, and sometimes determine, outcomes. (NB — Though maybe there’s some considerably more complicated variant of the traditional median voter theorem that accounts for multiple axes, say social and economic policy, along which American voting preferences can be separately mapped. What you’d need, then, is a good theoretical model describing how voters trade off preferences on social and economic policy, etc.)
June 23rd, 2009 at 2:25 pm
(Second NB — It should go without saying that the median voter theorem is relevant only if you’re in a first-past-the-post or simple-majority voting system, and there are no extraneous factors, like fundraising and charisma that determine outcomes. So the theorem can’t help you if the median voter prefers X, but some politician who is five points more conservative, or liberal, than X is a slick-talking slickster with the ability to raise truckloads of cash, and spend it on his campaign.)
June 23rd, 2009 at 2:38 pm
Jesus Christ. Tyler is an economist. Isn’t he supposed to understand game theory?
The Median Voter Theorem only applies to majority elections. We have one national majority elections — the President (and even that is disputable). The power wrangling we are seeing is coming from the Senate. While Senators are themselves subject to majority elections, the governing body as a whole is not. The theorem does not apply.
Or in short: What A Microeconomist said.
June 23rd, 2009 at 2:43 pm
The MVT, which I happen to know a little about, has been one of the least successful theories in the history of the social sciences (OK that’s a little hyperbolic…). I can’t speak of political science (which makes a lot of use of MVT), in econ the MVT is known to be a very rough approximation at best; it’s used, when it is used, bcs its a very simple model. Its a model which allows economists to insert political economic considerations into our other models. The MVT sort of exists along side more complex models of lobbying which are still computationally do-able, but suffer from their own issues with realism.
So, yeah, MVT is known not to work very well. And there are several reasons why.
Off the top of my head (I’m sure the political scientists would note I’m missing some key points): it ignores the endogeneity of voter preferences (voters form opinions on issues to match those of politicians they trust), it ignores the multiple dimensions of “policy” (health care vs national security for example) and the different priorties voters place on these issues, the candidates may not have the same “valence” (to distinguish between the endogeneity point above this could include things like candidate skill) and maybe voters just aren’t rational… I could probably go on.
June 23rd, 2009 at 3:45 pm
The sad fact about our bargaining democracy is that it puts an inordinate amount of power in what is perhaps one of the least palatable sectors of American society, the working class, undereducated white male. While lines are are clearly drawn for most groups, the above is needed to form any governing coalition, so an extremely high percentage of policy promises and policies are directed toward him. Unfortunately, what appeals to this empowered voter tends to be a mixture of xenophobia and homophobia on one side and inefficient economic policy like protectionism and outrageous union law on the other. Whoever can offer more, and whatever appeals during the given voting cycle sets the direction of the country… unfortunately.
-AM
June 23rd, 2009 at 7:21 pm
Even if most of the assumptions of the median voter theorem are met, we shouldn’t expect all senators to hold median voter positions due to the existence of primaries–indeed, senators should all be on either the 25th or 75th percentile if primaries also follow assumptions of the median voter theorem.
June 23rd, 2009 at 7:45 pm
i think you’re ignoring the fact that the median voter’s opinion can change over time, even short amounts of time. Just look at your iowa/louisiana example. isn’t a simpler explanation that chuck grassley and david vitter were last up for election in 2004, a year when, obviously, median voter sentiment was with the republicans, and landrieu and harkin were reelected in 2008, a year when this decidedly wasn’t the case? Obviously there’s incumbency effects and everything, but a random republican would have a better chance of winning either state in 2004, and a random dem would have a better chance in 2008. you could call the relative liberality or conservativeness of a state’s senators endogenous to the stat.
More instructive would be to look at senators from the same state, elected in the same year. This is somewhat rare, obviously, but here’s the last five examples:
Barrosso and Enzi, WY 2008
Cochrane and Wicker, MS 2008
Brownback and Roberts, KS 1996
Frist and Thompson, TN 1994
Conrad and Dorgan, ND 1992
In all five of the most recent cases, both senators were of the same party, and not really that far apart on their politics, either. Perhaps because the median voter they were facing had the same point of view in both elections.
Does this mean MVT is untouchable? of course not. if the sentiment of the median voter is so prone to change over 4 or even 2 years, then it’s hard to draw any hard conclusions from it. Still, it does suggest it isn’t useless.
June 24th, 2009 at 11:55 am
Try applying the MVT to the House and Senate voting on outcomes, not to voters choosing candidates. Lots of new Dems are Blue Dog types: that explains a lot of the Real World. Party control changed, median congressperson’s views not so much.
Yes, the median member of the ruling party has real influence too, as an old Levitt paper among others demonstrates, and as the excellent book Legislative Leviathan argues. But MVT provides Insight.