Articles and Analysis


Morning Status Update for 10/16

Topics: Status Update

Before the debates (and the quick reaction surveys released late last night), the flood of new surveys continued, as we logged 22 new statewide polls yesterday, plus three new standalone national polls and the latest updates from the (now) seven daily national trackers. The pattern at the state level remained consistent with we have seen since mid-September. Of the 18 new surveys that track previous results from the same pollster, and 13 show net movement to Obama, 3 to McCain and 2 show no change. Only three of the statewide polls represent updates since early October, and all three show net shifts to the Obama-Biden ticket.

2008-10-16 daily new

Note for rolling-average tracking polls, that the first table lists the previous non-overlapping sample for each pollster, not the release from the previous day.

The new polls affect our statewide trend estimates in two ways. They nudged the estimates in Obama's direction in six states, including a 1.3 point shift in Virginia that increased the Democrats' margin there to just over five percentage points (51.0% to 45.9%), enough to move Virginia and its 13 electoral votes to the light blue "lean Democrat" category on our map.

Although only three states show movement in McCain's direction, those changes helped shift both Colorado and Florida back from dark blue "strong" to light blue "lean" Democrat status.

2008-10-16 daily

The current Obama margin on our national trend estimate has fluttered around over the last few days, but has remained somewhere between 7 and 9 points over last week. Keep in mind that while the forward "tail" of the trend line (a funny sounding term, but I'm not sure what else to call it) tends to jump around as we add new polls, but regression trend line smooths out as we move forward. Charles Franklin sent an update of his "national forces" chart late yesterday and it shows continuing movement over the last week to the Democrats on both the national trend and his mash-up of all state polls.

2008-10-16 natinal forces.jpg

A note on the "all state polls" trendline. Franklin makes no effort to adjust or weight the state poll numbers to create a nationally representative statistic. He simply calculates the Obama minus McCain margin for each new statewide poll and plots a regression trend line through those points. The closeness of the level of "all state polls" trend to the national trend line is mostly coincidental. Theoretically, the "all state" trend could go up or down if the statewide polls over any particular period happened to be mostly from red or mostly from blue states. For the most part, the polls distribute consistently and the national poll and state poll trends generally mirror each other closely and right now, both trends continue to favor the Democrats.



How about "tongue" instead of "tail"? A tongue is a forward-seeking protuberance. Added benefit of "Taming of the Shrew" reference....


McCain sank again in the INTRADE bets, after the debate. Bettors did not like his performance.

I follow the Bets at INTRADE constantly and make synoptic tables of Toss Up States.

Vicente Duque



Unlike previous two debates, where flash polls' results guided the headlines, today the main-stream media touted the sprite performance by McCain. I don't know why the obsession with McCain's lively attack mode, which was often incoherent and contradictory and filled with malaprops. I suspect it's a desire to appear even-handed, a sop to crazed right wing critics of the media who claim Obama is coddled by them, or some desire by the media to keep the race close. In any case, I don't expect quite the bounce out of the debates that Obama got last time around, partly because he's running up against a ceiling of support, but also because of the how those who didn't see the debate will react based on the more positive picture the headline have portrayed of McCain's performance last night.



Re: 10:00 a.m. link today on politicshome.com feed. I think you mean "wary," not "weary".




I agree that BHO is probably reaching the limit of his support - and we may expect to see the trendline flatten a bit over the next week.

It's remarkable, tho, that while some have commented that the race has achieved some "stability," the trendlines still show a constant upward direction. While this may level off some in the few weeks ahead, BHO is still likely to get a bump from the last debate, and his numbers may stabilize at an even higher plateau.

JMC's problem last night - as it has been during his entire campaign - is that he has played to the base while turning off his natural constituency, the thoughtful independent voter.

My visceral impression of McCain last night was that he was a man about to have a stroke - and as David Brooks commented: "Who wants to look at that for the next 4+ years?"

Independents (and many sober Republicans) are also horrified at the prospect of his successor - if he really were to have a stroke.

Bottom line: JMC may have further solidified his base last night, but did nothing to reverse the trendline. I predict much wailing and gnashing of teeth amongst Republicans as election day draws nearer, and it may take a decade for the party to re-form itself.



How about nose.

People do win by a nose..



Would this be considered a normal sort of statistical paradox? Of the 3 new FL polls, the two that have polled before show an improvement for BHO and the third is as least as good, yet the color went from dark to light blue.



The three new FL polls have the margin closer (5,5,4) than the trendline pre-new polls (6.8), so the regressed margin reduced (to 6.1) - hence the shift, I think.
The change in the individual poll does not matter to the strong/lean/toss-up classification, but the effect of the latest poll on the trendline does... if that makes sense.



Tongue or nose work. 'Tip'?How about 'head of the snake'??




I think you mean JSM as in John Sidney McCain



So in Florida we have two repeat polls. Both showing Obama gains. We have one new poll which is as least as positive as the other two. So why did the state turn light blue? Did I miss something or is this one of those probability/statistics paradoxes?




RS gave a good explanation, but it wasn't nearly long-winded enough, so here's my take...

Pollster's coloring is essentially based on the width of confidence intervals around the current estimates of BO and JM's support. Depending on how much these intervals overlap, Pollster uses different colors for the states. More overlap implies more uncertainty.

The trends of particular pollsters in that state don't come into play at all.

Indeed, an interesting example is the recent Suffolk poll in CO, which (I think) pushed CO back into light blue all by itself. It had BO at 47, which was fairly low compared to the trend line estimate. Although this one poll didn't move the trend line much, it suggested to the model that there is more _variability_ in the estimate than previously thought. Hence the interval on BO's support widened a bit, enough to change the coloring.

Here's an artificial example that might make more sense. Suppose our trend estimates in a state are (45,50) and the state is currently colored dark red (McCain's leading). Now a poll comes in with (35,40) (Obama,McCain). This poll shows the exact same margin for McCain, but it will likely move the coloring to light red, simply because it suggests to the model that there's a huge amount of variation in the polling. (Imagine it saying, holy crap, Obama's support could be as low as 35? I'm suddenly way less _certain_ his support is 45). It might not move the actual trend estimates much, but the intervals widen, and poof, we recolor the state.

Modeling the difference between the candidates solves a lot of the issues here. Also, the current coloring system would be a ton more transparent to people, I think, if we had the ability to graph the CI's for the trend lines, so we could see how the inclusion of various polls alters not just the trend estimate, but the variability around the trend.

As I've been writing this, I've noticed the welcome addition of a candidate margin column in pollster.com's data tables. Is this a portent of graphs/trend lines of the difference? I hope so!



Didn't mean to post twice. But thanks for the tips. I gather its probably a combination of your ideas. I had forgotten about the 'raising the standard deviation' effect.


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.