Regular readers of mine know that I am not at all inclined to write off Barack Obama. This is not to say that I think he is the likely nominee of the Democratic Party. My point has simply been that people are underestimating his chances.

The reason I think this is three-fold:

(a) The guy has a real message that he has honed over the last few months. It could resonate with Democrats, who presumably are the most desirous of a national course correction. As evidence of this, we cannot overlook the fact that he has had donations from more than 300,000 individuals.
(b) He has raised $80 million to date. He will be able to compete as well as any presidential candidate ever has.
(c) He is fully staffed in Iowa and New Hampshire. He'll be blitzing both states with advertisements. He could win one or both states and turn those national numbers upside down.

Most pundits who are favoring Hillary Clinton so heavily (and remember - I am not saying that she is not favored; my objection is that people are favoring her too heavily) are using the polls - most often the national polls - to support their point. I have argued that this line of analysis is problematic because the national polls are too volatile.

The Gallup organization offered a rejoinder to this argument this week. Their basic assertion is that Democratic candidates with leads greater than 20% at this point (or later) in the election cycle have never lost the nomination. This gives them a modicum of stability in the national numbers, which therefore enables them to conclude:

By now, it is obvious that Clinton is extremely well-positioned to win the 2008 Democratic presidential nomination. Her status as the front-runner seems to be strengthening at an opportune time with the Iowa Caucuses less than three months away. It is not beyond the realm of possibility that she could stumble and not win the nomination as did Kennedy and Hart, but those cases occurred under rather extreme circumstances. Also, those candidates held their large leads long before any votes were cast.

I have a lot of respect for Gallup as a polling organization. Furthermore, I have a lot of respect for the analysts they have over there - Frank Newport and Lydia Saad in particular. However, I have to say that this line of analysis is misleading (though I am sure unintentionally so). This is one of those instances where the strength of the conclusion is driven by the cutoff point, which is largely arbitrary.

What do I mean by that? I mean that Gallup has based its analysis on a cutoff point of 20% or more. In other words, they look at candidates who had leads of 20 points or greater at this stage in the campaign. And from this they can argue (correctly) that no candidate has lost if they had that size of a lead at this point in the cycle.

Methodologically, though, this is not as straightforward as it seems. We are in what a salty statistical guy such as myself would call "small n territory." In other words, we do not have very many observations of Democratic primaries - especially open ones. By my count, this is just the eighth open primary since 1972 (when the primary system was implemented more widely). When we have just seven observations to draw inferences about this eighth - we have to be very conscious of observations that do not fit our hypothesis. They factor in a lot. Accordingly, we cannot impose a priori cutoffs that obscure problematic data points.

So - I'll just ask the question: what happens if we switch the margin in the polls to, say, 16%? In that case - we find that Howard Dean did have such a lead in mid-December, 2003 - over Joe Lieberman , the second place candidate at the time! His lead over Kerry was 19%. Kerry, at that point, was in fourth place - behind (in order) Dean, Lieberman, and Wesley Clark. John Edwards, who ultimately finished second, was actually in sixth place - behind every major national candidate. He was just a point above Al Sharpton.

This indicates that anybody who wishes to use the national polls to infer what will happen must advertise loudly and clearly just how wrong they turned out to be in 2003/04. Between mid-October and the end of January, we actually saw three distinct trends in the Democratic contest. At this point in 2003, Clark and Dean were essentially tied for first place. Dean overtook Clark in early December - and surged ahead through Christmas. His lead began to taper off after New Year's. And Kerry surged after the Iowa Caucus.

The problem with Gallup's analysis is that it misses all of this because of the standard that it sets. The problematic nature of 2003/2004 is never brought to bear on its analysis of 2007/08.

So, does this mean that Clinton's lead is as tenuous as Dean's? No. In fact - and here is one reason why I regard Gallup as generally great not only at polling but also at polling analysis - they rightly note that Democratic primary voters this year claim to be much more stable in their candidate preferences than they were in 2003 and early 2004. In the latest Gallup poll, 57% of Democratic primary voters with a preference say they are "certain" to vote for their candidate. This number is at 67% among Clinton supporters. By contrast, only 35% of respondents said they were certain back in early January, 2004.

This favors Clinton. There is no doubt about it. It is one of the reasons I do not think this race is a toss-up. I think that she has an edge in this contest. But I do not yet see this edge as being decisive - or as being great enough to enable us to call the race right now. Return again to that swing to Kerry. At the beginning of January, he was at about 10% in the polls. At the end of January, he was at about 50%. Edwards, for his part, doubled his support from about 7% to about 13%. Where did this support come from?

First, let's make some simplifying assumptions, none of which contradict common sense. Assume that (a) Kerry or Edwards supporters at the beginning of the month were Kerry or Edwards supporters at the end of the month; (b) Voters who initially supported candidates who dropped out, or undecided voters who decided, went to Kerry or Edwards; (c) A candidate's "uncertain" voters will stop supporting a candidate before his "certain" voters; (d) The "certain" supporters of Clark and Dean followed the national trend, i.e. 35% of their respective supporters claimed certitude.

From these assumptions, we can claim the following. First, Kerry or Edwards picked up the support of the Gephardt, Lieberman, and Braun supporters. Second, about two-thirds of the undecided voters decided for Kerry or Edwards. Third, they picked up about 70% of Dean and Clark's "uncertain" supporters. Again, these are all based upon my assumptions. I cannot be certain because the data I have is aggregate, and therefore I run afoul of the ecological fallacy if I claim certitude. But some process akin to the one outlined here most certainly happened. After all, at least 50% of the Democratic electorate switched their preference from the beginning of January to the end. The specifics might have been a little different (my intuition is that Dean and Clark kept more of their "uncertain" voters but lost some of their "certain" voters) - but it would basically have been a process like this.

Now - let's flash-forward to 2008 and run a scenario similar to the one that occurred in 2004. Let's assume that Obama scores a huge win in Iowa that begins a shift akin to the one in 2004. It knocks out all of the candidates except Clinton, Edwards, Kucinich, and Gravel. Let's again assume that Edwards does well enough to benefit, too. But again, like 2004, he is not the prime beneficiary. Further, let's assume that Clinton's "certain" voters are indeed certain - but that she loses 70% of her "uncertain" voters. This gives Obama and Edwards a chance to split amongst themselves 24% of the Democratic electorate - 11.5% of which had previously supported Clinton in an "uncertain" fashion.

In 2004, the split among the new supporters between Kerry and Edwards was such that Kerry accrued 85% of the benefit. Let's assume that this happens again. Obama gets about 85% of this new vote up for grabs, Edwards gets about 15%. This means Obama increases his share of the vote by about 20% to go to 41%. Edwards increases his share by about 5% to go to 17%. Clinton loses her 70% of her "uncertain" supporters - 11.5% of her total support - and thus falls from 50% to 38.5%.

So, we have at the end of January 2008: Obama 41%, Clinton 38.5%, Edwards 17%, Undecided/Other 3.5%. This would be an electoral pivot that is, by my estimate, about as dramatic as the one that occurred in 2004. The difference - an important one - is that Clinton would obviously not be knocked out. Instead, the contest would continue on to Super Tuesday, with Obama being the leader in the national polls.

Do I think that this is what will happen? No. Not really. The point is simply the following. Something like this did happen once, out of just seven trials. This makes it highly problematic to draw inferences about what will happen in January based upon October polling. That "falsifying" observation diminishes our level of certitude to a great degree. This would not be the case if we had, say, 40 or so primary cycles to review. But we don't. We have seven. And one of them was just plain nutty. This limits our ability to infer what will happen.

And remember the initial cause of Kerry's 2004 turnaround. It was essentially because Kerry won the Iowa caucus by 6%. Think about that. That is amazing in retrospect. The actions of 6% of caucus-going Iowa Democrats induced about 50% of the entire nation's Democrats to begin to support John Kerry or John Edwards. That is incredible.

I'll put it another way: the reason that all of these states are jockeying for an early position is exactly the reason why none of us should be putting anything but the most rudimentary of odds on either party's nominating contests. These states are tripping over each other to break party rules because they think these early contests could be pivot points. The reason that we are all following these state actions is because we agree with them. We know that these early contests can be pivot points. So why are we implicitly claiming that the Democratic contest will necessarily not have one?

Like I said - rudimentary analysis is really all that is valid right now. So, here is mine. Clinton and Obama both have a ton of cash. They both have good messages that could appeal to the Democratic primary electorate. Both of them stand a chance at victory. I would estimate that Clinton has an advantage over Obama that is probably due to her greater name recognition as well the fact that she is a known quantity from a family of proven electoral winners.

If you want to argue that Clinton has an advantage because of the size of her national lead - you are simply on unsolid ground. Recent history has demonstrated quite clearly that these national poll numbers are far from stable. They are subject to sudden, dramatic, and decisive changes because of tiny shifts in the early contests.