Insightology

To read the first part of this post click here. In that we discussed how to use the available information to predict a winner in the current presidential race. While the polls (including the poll average) gave a small edge to McCain at that time, the other measures gave a mixed picture with the combined measure giving Obama a slight edge. Since that post a month ago the race has changed quite a bit. In this post I will revisit those numbers to see where they stand, how they have changed and what that means when you track such numbers. In the next one we will look at the impact of race on the race, or the so called Bradley Effect.

After discussing the general unreliability of a single poll we discussed why a combined poll where the rating from different polls is averaged is better. The RCP poll average provided as a good example. That number is currently made up of nine polls and gives Obama a lead of 7.6% points.  A similar number is provided by CNN called poll of polls that is based on five polls and it currently gives Obama a lead of 8% points. It is important to understand here that these numbers are not to be interpreted as regular poll numbers. Because they are a combination of separate polls each with several hundred to a few thousand respondents, the normal margin of error calculations will not apply. Hence the difference of 7% points here is more real than when obtained from a single poll. These numbers also tend to be a lot more stable than individual poll numbers.

For example, you can track the RCP poll average from before the Democratic convention to now in this chart. Obama had a maximum lead of about 4.5% points right after the convention that steadily dropped and turned into a 2.9% point McCain lead, which in turn has gradually turned into a 7.6% point Obama lead. If you looked at individual polls at these times you will see a lot of jumping around in the numbers and not this sort of stable pattern.

We also discussed how the election is really based on electoral college votes and not the national popular vote. While they tend to be strongly related, in close races they can go in opposite directions as it happened in the 2000 election. Electoral college predictions are based on state level polls which tend to be less reliable than national polls because less number of polls are conducted and the sample sizes are often smaller. However, they are relatively more reliable in battleground states where polling organizations focus more (just like the candidates do) and spend their resources doing more polls. Using the same combination forecasting method, Pollyvote at this writing predicts that Obama will win about 348 electoral college votes (with 270 needed to clinch). RCP provides a more detailed breakdown listing the state numbers separately but the basic  result is the same.        

What do some of the other numbers we looked at say now? The quantitative models which mainly use economic conditions to predict the election had shown a lead for Obama (winning 52.5% of the two party vote) and that has remained essentially unchanged. A good question is why they haven't moved in spite of the very bad economic news in the last few weeks. One answer is that most of them haven't been updated in the last three weeks. When they are these numbers are likely to shift at least a bit. The other answer is that many of them are not based on the perceptions of the economy (which is very negative now), or on volatile numbers such as stock prices (which have been tumbling down), but on more fundamental factors that really don't change that quickly. That is one reason why the prediction from the quantitative models tends to be the most stable of all of these predictions.        

Speaking of prediction, where do the prediction markets stand? In the last post they were very close to even with the Iowa Electronic Markets slightly favoring Obama and Intrade slightly favoring McCain. IEM provides both a prediction of the vote share a candidate will win as well as the odds of winning, while Intrade provides the odds of winning. The vote share will not swing much while the odds will show more extreme values. Currently IEM predicts that Obama will win about 54.5% of the vote share, which is higher than what the quantitative models predict. Its odds of him winning the election are about 84% while Intrade places the odds of an Obama victory at about 77%.

Finally, the best prediction method Pollyvote which is a combination of all of these methods and includes expert surveys, predicts that Obama will take 52.7% ( a lead of 5.4% points) of the vote share. So it is clear that by every measure that we looked at last time, Obama is clearly ahead.

A last question peculiar to this Presidential election is the issue of race and the so called "Bradley Effect" whereby a black candidate may actually get fewer votes than he polls. That will be addressed in the next post.