fladem presented his argument thus:
538 is nuts
If you get a 5 point national shift, this race becomes very close in the electoral college. As I will show below, a 5 point shift is seen reasonably frequently in the last 10 days.
I have never been that impressed with 538 (this makes me a minority of one) and the probability strikes me as absurd.
Here is the simple truth: elections do not happen with enough frequency to permit the type of analysis suggested. Now you CAN compute a probability of winning IF THE ELECTION WERE HELD TODAY - but that is very different than attempting to predict a result.
What is the chance of a 5 point shift between now and election day? Consider:
In '76 Carter lead by 15 in Gallup (Oct 8-11) and by 9 in the NYT (Oct 8-12). A weak later Gallup found the lead cut to 6 (47,41, reported Oct 22). Carter won by 2: this represents a 7 point shift in the last 3 weeks, and 3.5 point shift in the last week.
In 1980 Carter lead in one poll taken Oct 16-20 (NYT, 39-38) and trailed by 3 in the Gallup poll taken by Oct 24-26. Reagan won by 10 (at least a 7 point shift in the last week.
In 1988 Bush lead in two polls taken about 10 days out by 13 points (NYT, Oct 21-24) and by 13 in the Washington Post taken Oct 26-31. Bush won by 7 (a six point shift late, though a Gallup poll had Bush up only 9 on Oct 20-21)
In 1992 there was about an 8 point shift in the race in the last three weeks, and in '96 the polling overstated the extent of Clinton's victory.
First off, fladem is right to call our attention to this history of volatility. Where he errs, I believe is three-fold: first, as I pointed out in a respone to that comment:
you're pointing to a much earler time when there were far fewer polls out there. A couple of polls swinging 6 points in 2 weeks time are nothing, that can be virtually all inside the MOE.
Since we're talking about polling trends, not a few scattered polls, we can't really make a direct comparison. Back then, the apparent volatility was almost certainly overstated, due to the thinness of the data set.
Second, I believe that the core argument is simply mistaken. Even if McCain did get a 5-pont break due to late volatility, it wouldn't get him close in the Electoral College without additional lucky breaks.
Third, I believe that trying to make arguments from recent elections (1976-2004) is a debatable proposition at best.
The first point, I think, has already been adequately made. I will, therefore, focus on expnading the last two.
Is A 5-Point Shift In National Polls Enough For McCain? Not Really
First off, fladem says "If you get a 5 point national shift, this race becomes very close in the electoral college." "Very close," maybe. But no cigar. Here's the relevant sub-section of Chris's table:
Give McCain a 5-point swing and at best he's battling to lose 277-261, at worst to lose 317-221. The former is "very close," while the latter is not. But how likely is the former? Answer: considerably less likely than meets the eye, given the trendlines in the states and the fact that early voting has already begun-with more than 1/4 of Colorado's votes already cast.
Here are the trendlines from Pollster.com:
Bad news all around here for McCain. A margin closer to 6 than 7, trend lines in the wrong direction, and more than a quarter of the votes already counted.
Somewhat better. Pollster.com's margin is only 3.9, but the trendlines look even worse than Colorado.
Pollster.com's margin is the most favorable of all. But, again,the trendline shows nothing to suggest a McCain turnaround.
And, on the other hand, Nevada, North Carolina and Missouri-which appear easier for McCain to pick off on Chris's chart, don't looks quite as good from Pollster.com's point of view, each showing long upward trends for Obama, dating back at least to July:
Of course a sweeping national shift could change all this. But where would that come from? Not only are the historical trands against McCain in all these states, the national favorability trends are all against McCain as well:
These are best seen as leading indicators. They show little possibility of a McCain turnaround.
In Fact, Drawing EVEN In National Polls Probably Wouldn't Be Good Enough
That, of course, is assuming a generally flatly-distributed shift, at least toward the middle of the field of contested states. This, too, seems like a questionable assumption. Obama fell behind McCain briefly in national polling, around September 8-11, according to Pollster.com's methodology, and stetching somewhat longer by Chris's, from September 8 to September 17. Yet, from this entire 9-day period, McCain didn't lead Obama in EVs until September 11, when he eeked out an 8-vote lead based on a national poll lead of 2.5%. This strongly suggests that McCain doesn't pull even in the EC just by pulling even in the national polls-and gaining 5 points doesn't even get him close to pulling even. Of course, this is partly due to the lag in state polling, as can be seen below at the end of the run, when Obama gained a national poll lead while McCain still held the EC lead. Smack in the middle of this period, however-September 15-McCain has a 2.3% national poll lead, but an EC lead of just 4.
Here's the full run of maps and related top-line information from the period when McCain was ahead. And note the states that Obama didn't have in his column then: states like Pennsylvania-for the whole run--and either Michigan or Minesotta:
September 11
Electoral College: McCain 229, Obama 221, Toss-up 83
National popular vote: McCain 47.4%--44.9% Obama
September 15
Electoral College: McCain 221, Obama 217, Toss-up 100 (270 to win)
National popular vote: McCain 47.4%--45.1% Obama
September 17
Electoral College: McCain 241, Obama 229, Toss-up 168 (270 to win)
National popular vote: McCain 46.9%--45.7% Obama
September 18
Electoral College: McCain 241, Obama 229, Toss-up 68 (270 to win)
National popular vote: Obama 46.7%--44.8% McCain
September 19
Electoral College: McCain 234, Obama 219, Toss-up 85 (270 to win)
National popular vote: Obama 47.0%--44.9% McCain
This run ended, finally, on Frodo's Birthday:
September 22
Electoral College: Obama 247, McCain 193 Toss-up 98 (270 to win)
National popular vote: Obama 48.0%--44.3% McCain
These maps show that it's a highly debatable idea that McCain could draw even in the electoral college based on drawing even in the national polls-where he now trails by about 8 points. Gaining just 5 points would leave him substantially short of the mark.
Realigning Elections vs. The 1976-2004 Data Set
fladem makes a very good point in drawing our attention to the recent history of presidential races and closing volatility. But it's one thing to say that it's a good check on tunnel vision, and quite another to argue that it's not a form of tunnel vision in its own right. What do I mean? Simple: I've been publicly calling this as a realigning election since October 2006-and harboring private thoughts for more than a year before that. Where the race stands right now, I appear to have been quite prescient. But really, I was just standing on the shoulders of the right giants. We were clearly due for a realigning election, the GOP's dominance was built on a house of cards, and their foundations were clearly crumbling once Katrina exposed their sheer and utter incompetence and heartlessness. The recent financial meltdown has only served to add a resounding exclamation mark.
If we take realigning elections as our standard, rather than recent elections, then we get a very different logic. fladem expands his logic a bit in a latter comment, saying:
If the odds are about 50-50 of a signficant late swing (based on the elections from 1976 to 2004) you get about a 25% chance of a 5 point or more swing to McCain. We can't predict the size of thde swing, but the range appears to be from about 5 to 10 points.
McCain needs about a 6 point swing to get to about a 50-50 proposition, so my 25% is too high based on this analysis.
However, from the perspective of realigning elections, which I take to be those that define transitions from one party system to another-with the addition of 1800, simply because the era began before the parties were well-defined-there have been six realigning (or dealigning) elections. Two of these have has transitioned between two different eras with the same dominant party-the elections of 1828 and 1896. However, in 1828, it was an intra-party election, as the subdominant opposition party had entirely disappeared, and the victor-Andrew Jackson-defeated the incumbent. Furthermore, in 1896, the incumbent, business Democrat Grover Cleveland, didn't run, but he was doubly refuted, first by the Democrat's swing toward populism, by nominating William Jennings Bryan, and second by the victory of Republican William McKinley.
Indeed, this was the only realigning election in which the subdominant party held office at the time of the realigning election. It's arguably quite possible that if the Republicans had held the Presidency during the Panic of 1893, then the Democrats might well have won the election of 1896, and become the dominant party of the next cycle. There is, thus, no case out of six in which the governing philosophy and the party remained constant across a realigning election.
One can, of course, argue that I am being arbitary in selecting these criteria-but they are precisely what realigning elections are all about. So the question is really whether it makes more sense to adopt fladem's recent-election perspective or my realigning-election perspective.
Of course, the answer is necessarily subjective, since both data sets are simply to small to admit to any sort of statistical argument. But that doesn't mean it is only subjective. There is an entire logic to realignment theory as I see it, which goes beyond a purely numerical argument. By this logic, once Barack Obama emerged as McCain's opponent, McCain was inherently at a deep disadvantage. "Change" and "reform" are the fundamental themes of realigning elections, and this meant that McCain's "experience" was an inherent liability, and Obama's "inexperience" was an inherent assetm quite the opposite of a "normal" election. McCain never really realized the depth of this problem for him, and it is much too late in the game for him to alter that logic, even if he had some idea of what that would mean, and how to do it.
This, I would argue, is the underlying reason driving the sorts of state-level polling and the internals that make it so implausible that McCain could mount a better-than-five-point surge this late in the game. It is the combination of that grimy low-level data and the grand sweep of realignment theory that ultimately lines up against fladem's argument. And it is why I think he is mistaken.
But come 2012, I think his argument will need serious consideration.
In fact, it's worth considering this time, too, on two counts. First, it made me think through all the above, and write it out for your consideration. Second, to the extent it is true, it would be perfectly combatible with a 5-point shift towards Obama, which would lift him well over 10%, which is where I've been saying that he could go all along. |
