Defending the Median Voter Theorem

Tyler Cowen recently wondered why the chattering classes don't view politics through the lens of the Median Voter Theorem. Matt Yglesias and Andrew Gelman respond by suggesting that disinterest in the MVT reflects the fact that the approach is empirically inaccurate.

This dismissal is both uninformed and uninformative. First, the MVT is trivially true. Given its assumptions (a single policy dimension, voters with single-peaked preferences (I most prefer zero troops in Iraq, second favorite is 500,000 in Iraq, least favorite is 1 million troops troops in Iraq), and simple majority voting, the outcome will embody the median voter's ideal point. Hence, if Congress votes on issues one at a time using simple majority, and all members have single-peaked preferences, legislation will embody the median voter's ideal point. Q.E.D. and all that.

Like any theorem, the MVT's expectations fail to hold when its assumptions don't hold. When members vote on multiple dimensions simultaneously or have non-single peaked preferences (I most prefer zero troops in Iraq, second favorite is 1 million troops in Iraq, least favorite is 250,000 troops in Iraq), then the median voter no longer dictates outcomes. This is why the evidence that Gelman and Yglesias present does not support the conclusions they reach. Gelman suggests the MVT is inaccurate because he sees no relationship between a moderate voting record in Congress and vote share at the state level. Yglesias asserts that the fact that Senators from the same state have different voting records in Congress refutes the MVT. Yet, neither Gelman nor Yglesias demonstrates that the MVT's assumptions hold in the cases they examine. Consequently, their evidence does not support any conclusions about the MVT.

Not only can we not dismiss the MVT for the reasons Gelman and Yglesias suggest, but in fact we might suggest that the MVT provides a strong first cut at understanding Congress. It's central expectations are consistent with what we observe. Legislation typically embodies centrist preferences and rarely embodies extreme outcomes and is generally quite stable. One might also suggest that the MVT helps us understand why change in the majority party has less impact on legislative outcomes than one might expect. The MVT certaily isn't the last word on Congress, but it's not a bad place to start. I challenge Yglesias or Gelman to propose a better one.

The MVT also helps us resolve puzzles such as, why did Congress include IMF funding (and "Cash for Clunkers" and money for H1N1 pandemic) in legislation to fund action in Iraq and Afghanistan rather than vote on all of these issues one at a time? Clearly some members wanted IMF funding. However, it is probably also true that the median voter on this issue did not. Members that want IMF funding can escape this "tyranny of the median" by linking their issue to others also likely to lose if voted on singly. IMF funders agree to support cash for clunkers, or larger military expenditures, or promise future consideration in exchange for votes for IMF funds. By engaging in heresthetics, one who is not the median voter can secure legislation that embodies her ideal point. And although the resulting legislation never embodies the MV's ideal point, the MVT nevertheless exerts a powerful influence on congresssional activity.

In short, the chattering classes should use the MV framework as a lens through which to view politics. Not because it is unconditionally true, but because it's useful. Their failure to do so, I suspect, reflects a failure to understand the framework beyond its most obvious (and possibly least interesting) implication. I think this state might be the fault of an academy which fails to teach rather than of a media that fails to wish to learn.