Insightology

A question that is of great interest now and the common response tends to be "who's leading in the polls?". You often hear people say that the latest poll from some reputable organization shows one candidate with a 3 point lead. Is that the best measure we have of predicting who will win the Presidential election? I submit that it is not and will walk you through the different measures available and what is most likely the best one. I say most likely because nothing predicts the future with complete certainty. Further, the aim here is not to examine the political strategies used by the candidates or speculate on who has a better ground game. We just want to see what is the best way to predict the winner, using all publicly available information.

As anybody who follows politics is aware, polls tend to have a peculiar quality. They tend to wobble and disagree with each other, but usually provide a good general sense of a race. So they are very useful in races that are clear cut (where one candidate is well ahead of the other), but are more problematic in close races. The reason for this is that a poll is an approximation. Unlike a census where everyone's response is obtained, a poll gets responses from a sample of people. Hence the numbers in a poll are fuzzy (or have a margin of error, statistically speaking). Depending on how many people are included in a poll these numbers are more or less fuzzy. When not enough people are included we can't tell if, say 48% is higher than 45% since statistically they are the same.

Polls also have other assumptions built into them that make the numbers provided by different polling organizations different. One major assumption is who is included in a poll as a likely voter. Another is an assumption about how many Democrats and Republicans will turn up to vote in this cycle. When taken together these and other important assumptions (including actual question wording and method of data collection) will have an impact on the results. Thus it is possible for the results from even the same polling organization to fluctuate over time and certainly provides sufficient opportunity for results from different organizations to significantly differ at any one time. So how do we overcome this?      

One common way of doing this is to take an average of several polls rather than just look at one poll. These averages tend to have more stability and have a better chance of providing us a correct number, largely by cancelling out the errors built into various polls because of the specific assumptions. The RCP poll average is one such nationally followed measure and if you are looking for quick information on what the polls are saying, you are better off looking at that than any single poll. As of this writing it had McCain ahead by 2.5% points.

Of course, Presidential elections are unique in that they are not decided by the popular vote. Individual states decide the ultimate winner in the form of the electoral college. So another way to look for a winner is to look at individual state polls and see how many electoral college votes a candidate has. This can only take you so far because state level polls are even less reliable than national polls. So, safe states are easy to decide but toss-up states have to be either not counted or should be counted on faith. Again the RCP site provides this information and in this case Obama has a slight advantage.

But are polls the only way to go? No. There are at least two other ways of predicting the winner. The first category is generally called quantitative models. The forecasters using this approach generally go with the idea that larger economic conditions, incumbency and external events (such as a war) will have a greater impact than campaign tactics or who the candidate is. Many of them hardly use any poll information. They usually use some form of a regression model (a statistical prediction technique) of these factors to predict the winner. Two of the more popular models are the Fair model developed by the Yale economist Ray Fair and the Lichtman keys developed by the historian Alan Lightman. Both these models as well as almost all others show Obama with an edge, reflecting the generic ballot advantage Democrats hold this year.

A second category of predictions is called prediction markets. This operates much like a stock market where people buy "shares" in McCain or Obama and the prices of the shares provide information on who is going to win. There is no attempt to find a proper mix of people to participate in the bidding. The idea is that like in a stock market people will use all the information at their disposal and pursue their self interest (rather than preference or ideology) and hence the decisions they make will be predictive of the future. The original Presidential election market is the Iowa Electronic Market and as of this writing Obama has a very slim lead which is much lower than what he had a few weeks back. The other popular market is Intrade which is based in Ireland and trades with real money (illegal in the U.S. except for IEM which has a special exemption). The latest information on Intrade gives McCain an advantage.          

So we have all this information, now what do we do? Clearly it is hard to predict a winner because the race is so close even when considering all these diverse methods, not just the national polls. But as we did with the polls, is it possible to combine all these methods to get an even better estimate? That is precisely what the folks at Pollyvote have done to develop what they believe to be the most accurate predictor yet. In their approach equal weight is given to the RCP polling average, several quantitative models, the IEM and a survey of experts. Even with McCain's recent surge in the polls, they give Obama a 3.2% advantage.

However, their expert survey was last done in June and I suspect that when it is updated the predictions will change in McCain's favor, but we don't know by how much. Further they include the IEM but not Intrade which uses real money and recently shifted slightly to favor McCain. So for my prediction I took out the expert survey and included the Intrade prediction market information. Plugging everything in I come up with 49.4% for McCain and 50.6% for Obama. This gap of 1.2% is closer than the 3.2% gap on Pollyvote. So it looks like at this time Obama still has the edge, but it is barely the thickness of lipstick!