who's gonna win?

As a veteran election observer, the number one question my less political friends have been asking me (besides, "why are you so damn good looking?") is "Who's gonna win in November?" For a good part of the summer my answer was certain. Maybe it was out of excessive optimism or a need for something to push me to work harder. But I was convinced that John Kerry would come out the victor.

Today, I am not so certain. But this is not because I think Bush is going to win. When it comes down to it, I have absolutely no idea what is going to happen in November and I have a hard time believing anyone really does. Here's why - For two reasons, predicting the outcome based on polling (for this election in particular) is futile: 1) The race is really really close; and 2) There are faults in the methodology behind the polling. What makes this election different is that because it's so close (#1), the faults in the methods (#2) make predictions impossible.

1) No one would dispute the fact that this race is neck and neck. According to polling compiled by The Hotline, in only 28 states is one of the candidates leading outside of the margin of error of the poll. In other words, in 22 states, making up 256 of the 538 electoral votes, no polls can statistically predict who is leading. In the states where we can be very certain about who is leading, Bush has 153 electoral votes to Kerry's 129 (a lead of only 24 electoral votes). Think 2000 was close? That was nothing.

2) Because so many of the polls are too close to give a statistical winner, the faults in polling methods, which always skew the results by a few points, make a real difference this time. Basically there are two main problems with polling: a) accurate samples; and b) the likely voter vs. registered voter issue. I will just quickly touch on these two issues, but for a very thorough discussion of this topic see "Why the Race is Closer than People Think" on Ruy Teixeira's blog Donkey Rising.

a) The point of polling is to accurately sample the population using a very small sample size relative to the total population size. Unfortunately, there is error inherent in this process, although most polls reduce this error to + or - 3/4%. What this means is that if you have a poll with +- 3% error, even if the sample population shows Bush with 50% and Kerry with 45%, the actual numbers could very well be Bush with 47% and Kerry with 48%.

When taking polls, pollster try to accurately sample the total population. But this is not a perfect process. If polls are done over the phone, then people without phones are left out. If polls are done during the evening, then people that work the night shift are left out. The point is that polls are not perfect at sampling the true population and so they are inherently flawed. As Teixeira points out, the recent national polls have been showing a 9-10 point Republican affiliation advantage when, in fact, Democrats probably have a 3-4 point advantage (based on data from 2000 and taking into account realistic fluctuations in this number). If these polls are sampling a higher percentage of Republicans than actually exist, they are most certainly overstating Bush's lead.

b) In trying to make their polls more accurate, polling groups have adopted the method of calculating if an individual is a likely voter (LV) or a registered voter (RV). Polling people that aren't going to vote (even if they're registered) skews the results of polls. The problem is, there is no way determine if someone is going to vote in an election that is several weeks away (note: the likely voter approach is much more useful for polls in the days before an election). The result is that many individuals who are not determined to be likely voters are left out of the data. In this election, these people (we know how they vote because some pollsters release both LV and RV polling) tend to be Kerry supporters.

The point is, it could still go either way - so if you care strongly about what happens, then get out there and help!