Caution: momentum shift in progress

Due to a little discussed problem with national and state polling, it would seem that most election prediction sites are greatly over estimating Sen. Barack Obama's lead.

Now, I stand behind my prediction that Obama wins the election, and I stand by my prediction that he gets 313 electoral college votes. But, due to a lag in polling that runs several days behind, I think some sights are discounting McCain's claims that the race is tightening.

The national polls are usually conducted over the course of two days, sometimes three. Those polls are then published in the following days after the polling organization gets a chance to crunch the data. The state polls are conducted in basically the same way.

The "current" polls therefore, are actually from several days ago - mostly from over the weekend. Polling from the end of last week and the weekend shows that Obama's lead is still stable. But they also show a change in direction. Averaged together, Obama's lead is now 6.5 nationally, a slight down-tick since last week. As I have stated many times, polls are better at showing trends and direction than they are at predicting the outcome of anything - particularly in politics.

Unfortunately, most of the data geeks are not reporting on what seems to be a momentum shift. The frenetic pace of Obama's campaign has slowed considerably over the last week, and I, like some of the conservative analysts, see the race tightening a bit.

I am also concerned a great deal about the tendency of pollsters and data geeks to over-estimate the shift in voter turnout models towards the Democrats. Most pollsters adjust their samples to account for what they think the actual voter turnout will look like. Last year, the models held them roughly even. This year, nearly all have adjusted for increased numbers of registered Democrats - some by silly margins.

Nate Silver, at fivethirtyeight.com, aggregates and averages these polls to reach conclusions about how the electoral college will shake out. Electoral-vote.com does the same. The problem is that they are relying too heavily on models that may be over-estimating Democratic turnout. Add to that, a lag in both the polling data and the effect of those polls on their electoral-college models they use (which again average recent polls per state to compensate for wild swings and outliers), and you get predictions that show Obama in a landslide where he gets more than 350 electoral college votes.

That may very well be the case. But, I for one, am not all that confident in the voter turnout models. Couple that with a shift in momentum, that will become apparent in the next few days, and I think there is sufficient reason to believe that the election could be close.

That said, I do stand by my original prediction of 313 for Obama - meaning he wins NM, CO, VA and FL in addition to the states that John Kerry won in 2004. I am growing less confident, however, about Florida (but I still believe Obama can hold on). Since I made that initial projection, Ohio has actually moved further towards Obama. At this point, those two states could go either way.

In the end, all that matters, really, are New Mexico, Colorado and Virgina. If Obama wins two of those states, plus the Kerry states, he wins. And I think that is exactly what will happen.