Why Third-Party Predictions Can Be So Weirdly Optimistic

Doug Mataconis is impressed that third-party candidate Robert Sarvis is polling as high as 10% in the Virginia Governor’s race.

Color me skeptical, not because of anything specific to Sarvis but because of statistics. As the true level of candidate’s support gets farther in either direction from 50%, polls become increasingly less accurate. To quote myself from a post that lays out the math in more detail:

It is simply harder to predict events that are unlikely than events which are likely. If a fair coin is being flipped over and over and you have to guess on which particular flip it will come up heads, you’ve got a 50-50 shot of winning the game. But if the same game is played with an unbalanced coin that comes up heads only 1% of the time, you will almost certainly not guess the right flip, even if you are allowed to play many times. Indeed, any system you might use to predict when the elusive heads result will occur will be less accurate over time than simply predicting that the coin will never come up heads no matter how many times it is flipped.

As I show in the linked post, a 90% accurate poll including a candidate who actually has 1% support will estimate his/her support at 11%. And when your support is really low, most errors in estimation can only go in one direction: Upward.

Sarvis could matter in the tight Virginia Governor’s race even if he only nets a few percent of the vote. But the likelihood that he truly has the support of 10% of Virginia voters is low.

[Cross-posted at The Reality-based Community]

Keith Humphreys

Keith Humphreys is a professor of psychiatry at Stanford University. He served as a senior policy advisor at the White House Office of National Drug Control Policy from 2009 to 2010.