The Audacity of Daily Tracking

Gallup remarks:

Obama's particularly large leads over McCain in Friday and Saturday's tracking suggest that the massive publicity surrounding Obama's speech at the Victory Tower in Berlin on Friday -- the only major public event of the trip -- and coverage of Obama's meetings with the heads of state in France and Germany may have tilted U.S. voter preferences more in his favor.

Maybe. But look, Obama's been drifting in the 45-48 range and McCain's been drifting in the 41-44 range and there's no reason to think that movements within the familiar bands represent anything other than normal fluctuation in a statistical sample. I think the commentary on the tracking polls looks more and more like the silly commentary on the daily fluctuations of the Dow where first analysts look at numbers, and then second they devise post hoc explanations of the movement. Realistically, I don't think there's anything worth commenting on unless some much more sustained trend develops.

Comments

But, c'mon Matt, with the current conditions, shouldn't Obama be up 35-40? Does Obama have a Jewish problem?

Posted by Stacy | July 26, 2008 1:49 PM

I think it's more plausible to think that there WAS a bounce directly attributable to the trip, but since that's kind of an ephemeral event (like nomination wrap up bounce, or looking ahead to convention bounce), it's real, but not particularly relevant for November. So just keep looking at the trend.

Posted by JMS | July 26, 2008 2:02 PM

Was the bounce attributable to the trip--or was it due to the ridiculously whiney way McPouty conducted himself outside "der Wiener Haus?"

Posted by Ella in NM | July 26, 2008 2:12 PM

That's funny, CBS network radio news at the top of the hour has been saying all day that Obama's trip hasn't helped him in the U.S. polling results, and that McCain still holds a small (2%) lead over the presumptive Democratic nominee.

Posted by Ben | July 26, 2008 2:12 PM

Agreed--we are still within the range of random fluctuations.

Of course if the gap starts averaging 6-7 points, not just periodically reaching 6-7 points, then I think that would count as a sign of a legitimate change.

Posted by DTM | July 26, 2008 2:16 PM

Ben--I can't find a "news" report that doesn't seem to infer that. It seems to be the week's conventional lack of wisdom they are imparting to the ignorant masses.

I feel the McCain campaign taking on more and more of the exact same Karma as Hilary's. He's actually morphed into her same tactics, and it seems like it's having the same long term effect on the public, even if (just like with her) the media can't see it.

Posted by Ella in NM | July 26, 2008 2:18 PM

and that McCain still holds a small (2%) lead over the presumptive Democratic nominee

Did I read that right? Are they saying McCain has a 2% lead? Or did you get that backwards? What polling has ever shown McCain ahead?

Posted by ResumeMan | July 26, 2008 2:22 PM

Although I guess it is notable the gap has been trending up in the Rasmussen tracking poll as well. I still want to see more, but that does represent a bigger data set.

Posted by DTM | July 26, 2008 2:24 PM

Yes Matt, you're right that a one point change from one day to another is really nothing worth commenting about.

However, the real richness of the Gallup data is to combine the last 3 days of data, and to see if the difference is statistically significant at a 95% confidence level with a sample of n=3000 (the margin of error being +/-1.8%). And while earlier this week the data indicated a statistical tie, the last couple of days now indicate a significant difference, meaning that there truly has been a bump in Obama's numbers.

Of course, the weakness of the Gallup poll is that it doesn't include Barr and Nader.

Posted by Phil | July 26, 2008 2:25 PM

McCain has been at 41 only once before on that chart, on July 20. Obama hit 48 twice in early July. It's the 1st time there has been such a gap, 7%, following 6%, which means the last two days are even higher, because the days dropping off were closer. I'm inclined to beleive it is the juxtaposition of McCain whining in front of the the cheese counter and Obama speaking to 200,000 in Berlin. Let's see if Obama can capitalize on it--see if it lasts.

Posted by Mimikatz | July 26, 2008 2:26 PM

Just like imputing causes to minor changes in the Dow Jones Industrial Average has always struck me as silly, so too does imputing causes to a candidate's daily poll numbers. So I agree with Matthew. But I do find it interesting that this is one of the largest Obama leads the poll has ever captured. It could be just statistical noise, but it will be worth watching other polls to see if they match up.

Posted by Cello | July 26, 2008 2:29 PM

McCain whined overtime this week to get into the news-cycle, only to fuck up so badly he had to cancel appearances and start over.

If he weren't senile, he'd have know to lie low and wait for Obama to screw up. But the more McCain puts himself in front of the public, the more they're going to wonder what the point is.
.

Posted by Grand Moff Texan | July 26, 2008 2:35 PM

Doesn't it look like the tracking poll gets close about once every 7 days?

It seems like the day of the week on which people are polled may have some significant effect, considering the massive age disparity between Obama and McCain voters. If the pollster calls on Friday after elderly diners have returned from the early-bird special at the local Luby's, they may get a pro-McCain bias.

Posted by Ban Johnson | July 26, 2008 2:40 PM

That Gallup analysis is not precisely worded, as it has been in the past, but I think what they're saying is, Friday's numbers (which have been folded into the average) and Saturday's numbers (so far, which are not reflected in the average) are significantly higher than they were at any point in the past few weeks. They're trying to preview the bump for us. We'll see tomorrow and Monday if it pans out. If Obama reaches 50, I think the CW will be that he got a good bump from his globetrotting.

Posted by along | July 26, 2008 2:41 PM

Quinnipiac: McCain Gains in Key Battleground States

Oops!

Quote:

Most interesting finding: "Voters in each state say energy policy is more important than the war in Iraq. And by margins of 22 to 31 percentage points, voters in each state support offshore oil drilling, and by seven to 12-point margins, drilling in the Alaskan National Wildlife Refuge."

Posted by Mixner | July 26, 2008 2:43 PM

Matt, the thing here is, the explanation is post-hoc. We had an null hypothesis, it was tested, and we saw a change.

The real hope w/ these numbers is this: McCain was never popular enough to get over 45%, but Obama's support has softened and weakened over time given voters' various concerns over him. Obama may finally have assuaged a lot of those doubts and he'll stay solidly higher.

Posted by Joel W | July 26, 2008 2:48 PM

I think the commentary on the tracking polls looks more and more like the silly commentary on the daily fluctuations of the Dow where first analysts look at numbers, and then second they devise post hoc explanations of the movement.

LOL! Matt do you *really* think that that is how the daily movements of the Dow is reported - that analysts pay no attention to the internal workings of the market and instead just make stuff up post hoc?

Maybe if you actually understood the internal workings of the market like the analysts who watch the market for a living you might understand the reporting instead of making idiotic assumptions.

Posted by Ethel-To-Tilly | July 26, 2008 3:11 PM

That's why it's called bounce.

You're up for a while then you return to some sort of equilibrium.

What might be interesting, i.e. meaningful, would be to devise a way to track whether or not the collective effect of several bounces over a period of time move the equilibrium in any measurable way.

And as I've argued elsewhere just this morning, it's still too early to worry about this stuff. Maybe after the conventions and Labor Day.

Regards

Monica

Posted by Monica Wolf | July 26, 2008 3:13 PM

Also, I would think that the statistics of the -difference- of two random variables (i.e., Obama preference minus McCain preference) is more uncertain than either the Obama preference or the McCain preference alone. You can't just do arithmetic with random variables and expect the uncertainties to stay the same.

Posted by MattF | July 26, 2008 3:24 PM

A short thought experiment...

Let's assume that both McCain have a real-world floor of 40%. (Even Mondale received 40%!)

100% - (40%+40%) = 20%

In any given poll this 20% of voters must lean McCain or lean Obama in the polls, others are reported as Undecideds.

Most experts seem to think these Leaners and Undecideds are of two general types:

i) Higher information voters who genuinely cannot decide,

ii) Ultra-low information (these are the folks we shake our heads at when x% of Americans cannot identify the president or find the U.S. on a map).

For argument's sake lets say that that both groups are about equal in number at about 10% and that the second group rarely influenced by events like Obama's trip because they are impervious to information.

That means that any change resulting from events comes only from the 10% who honestly cannot decide.

When a swing in the polls occurs it represents either people moving from McCain directly to Obama or intermediate movement McCain to Undecided to Obama.

Imagine a 3 point swing. If the entire change is McCain to Obama then 1/3 of those who could be changed have changed. If the change is entirely intermediated 3% McCain to Undecided, 3% Undecided to Obama then roughly 2/3 of those who could change their mind have done so. The real effect would obviously be somewhere in-between.

This means that for events to explain even a small observed change between 1/3 and 2/3 of all the people who could be moved actually move. Is this a reasonable result? I don't know but I can see why it gives Matt pause...

My numbers are very rough (Obama's floor is probably higher than McCain's for example) but play with each group and see what results you get.

Posted by sven | July 26, 2008 4:06 PM

And as I've argued elsewhere just this morning, it's still too early to worry about this stuff. Maybe after the conventions and Labor Day.

At which time whatever polls are taken will pretty much remain constant until the November election. By Labor Day, the die has already been cast, and there won't be much either candidate can do to change anything. We're already in late July. What happens between now and September 1st is what is going to matter.

Posted by Tyro | July 26, 2008 4:21 PM

I think the McCain campaign is a little worried about an Obama bounce and a lot worried about whether he's closing the "commander-in-chief" gap. Evidence: I watched Fox for about an hour at the gym, and there were three blocks about Obama's last-minute cancellation of his trip to the hospital at Landstuhl. Bill Hemmer pushed Obama at least four times, refusing to accept that Obama decided against politicizing a visit to war amputees. HE SNUBBED TEH TROOOOOOOPS!

McCain hasn't actually said anything about this - it's been left to conservative media and a couple McCain surrogates. We'll know how panicked McCain is if he actually mentions it himself when the new cycle starts on Monday.

Posted by Dave Weigel | July 26, 2008 4:52 PM

Tyro,

At which time whatever polls are taken will pretty much remain constant until the November election. By Labor Day, the die has already been cast, and there won't be much either candidate can do to change anything.

Gallup's daily tracking poll for the 2000 election showed Gore with a 5+ point lead over Bush through most of September. And just two days before the election, the poll showed Bush with a 5-point lead over Gore. The final vote was of course much, much closer. So close, in fact, that to this day it is still disputed which candidate received a larger share of the popular vote.

The idea that polls are any kind of reliable predictor of presidential election outcomes even by Labor Day, let alone more than a month earlier, is simply not supported by historical experience. If Obama had had a persistent 10+ point lead over McCain, or vice versa, then there might be some reasonable basis for a confident prediction, but as things stand now the race is wide open.

That's one reason why Matthew and his buddies are desperately flinging poo at McCain in the hope that something will stick.

Posted by Mixner | July 26, 2008 4:58 PM

I watched Fox for about an hour at the gym, and there were three blocks about Obama's last-minute cancellation of his trip to the hospital at Landstuhl. Bill Hemmer pushed Obama at least four times, refusing to accept that Obama decided against politicizing a visit to war amputees. HE SNUBBED TEH TROOOOOOOPS!

That's going to be an interesting dynamic because the Bush administration is elbow-deep in the Landstuhl story. It's a transparently political gambit by a Pentagon staffer, Bryan Whitman, who has a history of launching transparently political gambits. So if McCain pushes this hard, he may end up with a raft of stories tying him even closer to the Bush administration.

Posted by southpaw | July 26, 2008 5:03 PM

Mixner, you provided a counterexample for an elections whose circumstances hadn't been seen in the US since the 1800s. So what?

I remind you that the polls taken after labor day are invariably consistent with the final outcome of the election. If I were making an incorrect claim, you would have had more examples. All you could come up with was a poll from an anomalous election. I think you're just trying to pick a fight. Don't do that; it's asinine.

Posted by Tyro | July 26, 2008 5:05 PM

One thing worth pointing out is that Rasmussen's daily tracker has also shown a considerable bump for Obama, going from a dead heat to a 6-point Obama lead in only four days. So there's some corroboration to the idea that Obama has in fact gotten a "bounce" from his trip abroad.

Posted by Darius | July 26, 2008 5:11 PM

Tyro,

I remind you that the polls taken after labor day are invariably consistent with the final outcome of the election.

You'll have to produce some evidence that this assertion is true before you can "remind" me of it. I know you have a hard time distinguishing between evidence and your wishful thinking, but try.

The "invariably" part is clearly nonsense, as the example of Bush vs. Gore shows.

Posted by Mixner | July 26, 2008 5:14 PM

Mixner, the facts are on my side. The winner of the first polls after labor day always goes on to win. Exceptions, other than 2000 were 1960 and 1948. And polling was less sophisticated in 1948 than it is now. (src)

There's little either candidate can do to overcome a post-labor-day trailing in the polls. What goes on in August matters much more than September.

Why did you not marshal any evidence to back up your ridiculous claim, anyway? All you could come up with was something from an anomalous election. Are you really that ignorant, or are you just lazy? Or are you just stubborn and ornery and enjoy being defiant when you have to deal with people, like myself, who know what they're talking about?

Posted by Tyro | July 26, 2008 5:24 PM

So close, in fact, that to this day it is still disputed which candidate received a larger share of the popular vote.

Disputed by whom? Who won the popular vote in Florida is disputed, but I don't think there are any sane people claiming Bush won the national popular vote in 2000.

Posted by KCinDC | July 26, 2008 5:26 PM

What happens between now and September 1st is what is going to matter.

No, the die is already cast. It was pretty much cast in January, at least to the extent of which party's nominee was going to win (even if it wasn't decided who the nominees would be). This is true most years, but it's especially true this year. (See Lichtman's Keys theory for more discussion of why this is so.

You can watch tracking polls for their entertainment value if you like, but it's a bit like reading a story that you already know the ending to.

Posted by jimBOB | July 26, 2008 5:29 PM

Who disputes that Gore won the popular vote in 2000?

Posted by Jeffrey Davis | July 26, 2008 5:43 PM

Realistically, I don't think there's anything worth commenting on...

And, yet, here's a post and 30+ fairly empty-content comments. I guess you have a quota to fill each day for your soon-to-be-ex-overlords? Otherwise, I fail to see the purpose of a "this isn't worth the time or effort, but here's the time and effort" exercise.

Posted by David | July 26, 2008 5:48 PM

Tyro,

Mixner, the facts are on my side. The winner of the first polls after labor day always goes on to win. Exceptions, other than 2000 were 1960 and 1948. And polling was less sophisticated in 1948 than it is now. (src)

Er, your source is an article from early 1992, four general elections ago. It gives no information about the size of the leads of the final winners at Labor Day. Labor Day polls may usually predict the winner in elections where one candidate has a large and persistent lead by Labor Day, but that doesn't mean it is a reliable predictor for closer elections, like Gore vs. Bush in 2000 and Obama vs. McCain in 2008. Many pundits in September 2000 were confidently predicting a clear win for Gore on the basis of polling data that turned out to be wildly wrong. William Saletan's Slate piece Why Bush Is Toast, from September 14 2000, is typical.

If by Labor Day Obama has managed to acquire a persistent 10+ point lead over McCain, then he'll probably go on to win. But if his lead remains only at its current level of around 2-4 points and is still fluctuating significantly, then it's anybody's guess as to who the winner will be in November. Of course, Obama's lead may also have vanished completely by Labor Day.

Posted by Mixner | July 26, 2008 5:59 PM

If giving a speech in Berlin gives Obama a few points bounce, I'd say a war in Iran should give "war hero" McCain the ten points he needs to knock Obama off if Obama is still only leading by 4-7 points in October.

Especially if the MSM entirely focus on McCain's response to the war, while spinning Obama's no doubt equal support for a disaster.

Still have yet to hear ONE Democrat address this possibility.

Posted by Richard Steven Hack | July 26, 2008 6:35 PM

What's the thing flying over my head called "the point"?

Posted by Mixner | July 26, 2008 9:36 PM

Actually, I'm coming round to Mixner's view on this issue.

Posted by Jeffrey Davis | July 26, 2008 10:44 PM

Me too. He's making a persuasive case.

Posted by Tyro | July 26, 2008 10:47 PM

Is the increase in Obama's lead from 2% to 7% consistent with fluctuations due to sampling ?

I just calculated. When calculating the standard deviation of the change due to sampling I assumed that there are no undecided voters (so I overestimate that standard error). Nonetheless I calculate that the change was 2.21 times the standard error.

Calc goes. The two polls overlap on one day, so the change from the (july 21 + july 22)/2 to (July 24+July 25)/2 is (3/2)5 = 7.5%. The sample sizes were about (2/3)2600. with the no undecideds assumption the variance of Obama-McCain is 1 for each respondent. The variance in the fraction obama-fraction McCain is 3/(5200). The variance of the change in this difference from 21-22 to 24-25 is the sum of the variances of each difference (they must be independent if change due to sampling) so is 3/(2600)
The standard deviation is 3.40% so change/sd is 2.21 rejecting your null with a p-level of 1.36%

Now If I hadn't picked the points ex ante, this would refute your null hypothesis. The probability that there would be such a large ratio of the change to it's standard error over any of N intervals is N(1.36%) so your null would survive my efforts only if we are convinced that I searched over at least 4 different intervals to find the one I liked. Since I eyeballed the graph, this is very possible.

But Rasmussen also showed a big bounce for Obama. I'm going to use the same time interval as for Gallup. Rasmussen has an even larger sample but a smaller bounce. from 7/21 report to 7/26 Rasmussen bounced up 5 which rejects the null (if one considers the undecided when calculating standard deviations as I have numbers not including leaners ignoring that change in diff over overestimated sd is 1.94) . Those are really really the only polls I know about. The chance that 2 out of 2 reject at the 5% level due to samping is 25/10,000 that is really low. Now I have made another choice (both of two tests not say either of two and so on) so that the amount of insta-data-dredging is increased. Still I just don't believe that the apparent bounce is due to sampling

So what is going on ? If anyone is still reading, I am wasting your time. My point (if any) is that two things are confused. Normal fluctuations which tell us nothing about who will win the election and fluctuations that may well be due to sampling.

Pollsters generally choose sample sizes so that the two are similar. Gallup has a very large sample. If one averages over many polls one reduces variance due to sampling. The large Gallup sample and the large pooled samples are confusing to people who identify statistically insignificant and politically unimportant. Ordinary boring unimportant fluctuations and fluctuations due to sampling are not similar at all. I'd say Matt Yglesias meant to say "tells us little about November" when he wrote "normal fluctuation in a statistical sample."

Posted by Robert Waldmann | July 26, 2008 10:49 PM

"And just two days before the election, the poll showed Bush with a 5-point lead over Gore. The final vote was of course much, much closer. So close, in fact, that to this day it is still disputed which candidate received a larger share of the popular vote."

No, that's not true at all. Gore won the popular vote by about half a million votes. No one disputes that.

Posted by mad6798 | July 26, 2008 10:59 PM

"Also, I would think that the statistics of the -difference- of two random variables (i.e., Obama preference minus McCain preference) is more uncertain than either the Obama preference or the McCain preference alone. "...

Posted by MattF

Absolutely MattF. The standard devition due to sampling of McCain-Obama is almost exactly twice the standard deviation due to sampling of support for Obama which is almost exactly the same as the standard error due to sampling of support for McCain.

If there were no undecided voters, the s.d. of support for McCain would be exactly equal to the s.d. of support for Obama and the s.d. of the difference would be exactly twice that number.

Also the s.d. of the change from one poll to another in the difference Obama - McCain is greater than the s.d. of the difference in each poll (if they have the same sample size it is greater by a factor of the square root of 2). This is how I made my calculations.

Posted by Robert Waldmann | July 26, 2008 11:06 PM

I haven't read the polling methodology, but I would think if it is based on land-line surveys, Obama may have a larger lead. I think it's mostly 20-somethings (generally Obama supporters) that have cell phones only. Still, it appears to be a close race with Obama in the lead. I think the debates have the best chance to affect the race either way.

Posted by S.P. Gass | July 26, 2008 11:24 PM

in fact, that to this day it is still disputed which candidate received a larger share of the popular vote.

Now that is some world-class hackery. By who? On what basis?

Posted by Scott Lemieux | July 27, 2008 2:42 AM

Re: Many pundits in September 2000 were confidently predicting a clear win for Gore on the basis of polling data that turned out to be wildly wrong.

Um, Gore did have a clear win, as far as the popular vote goes.

re: But if his lead remains only at its current level of around 2-4 points and is still fluctuating significantly, then it's anybody's guess as to who the winner will be in November.

He will probably win the popular vote, but could lose in the electoral college as Bush did. A second such result in so few years will probably be the death knell of the electoral college.

Posted by JonF | July 27, 2008 8:18 AM

Why does eveybody call the Victory Tower? Tower would be Turm. A Säule is a pillar, or when it's a Roman phallic symbol not propping anything up, a column. As in: Me Trajan, you Jane!

Posted by James Wimberley | July 27, 2008 8:55 AM

Why does everybody call it the Victory Tower? Tower would be Turm. A Säule is a pillar, or when it's a Roman phallic symbol not propping anything up, a column.
As in: Me Trajan, you Jane!

Posted by James Wimberley | July 27, 2008 8:57 AM

I've arrived: someone else has posted under my name! Just think what this means: women, money, cars, health care, dental!

Posted by Jeffrey Davis | July 27, 2008 12:47 PM

So with the Gallup poll now showing a further increase to a 9-point lead, I am now ready to jump sides--this looks like a real shift. It could of course be a transient shift, but I no longer think it is likely just the product of sampling variation.

http://www.gallup.com/poll/109102/Gallup-Daily-Obama-49-McCain-40.aspx

Posted by DTM | July 27, 2008 1:25 PM

Um, Gore did have a clear win, as far as the popular vote goes.

Gore won the popular vote a tiny margin, 0.5%. Labor Day polls had put him in the lead by an average of 5%. Some had him ahead by 10%. That is why many pundits were confidently predicting a clear win for Gore. The Labor Day polls massively overstated Gore's lead in the popular vote compared to the actual outcome of the election.

Posted by Mixner | July 27, 2008 3:22 PM

"Obama's been drifting in the 45-48 range and McCain's been drifting in the 41-44 range"

until today, when Obama has gone up to 49 and McCain down to 40.

get ready for the landslide...

Posted by charles | July 27, 2008 6:16 PM

get ready for the landslide...

The degree to which liberals are blinded by wishful thinking is truly wondrous to behold. When a daily change shows McCain gaining on Obama, it's dismissed as random variation. When a daily change shows Obama increasing his lead, it's the start of a "landslide." You people really need to reign in your tendency to jump to conclusions.

The latest Rasmussen daily tracking poll, by the way, shows a contraction in Obama's lead, from 6 points in yesterday's poll to 5 points in today's.

Posted by Mixner | July 27, 2008 6:48 PM

Mixner: none of the cited polls in the last three weeks shows McCain leading Obama. The Intrade betting contract for Obama to win is at 64.1, for McCain at 31.7. Don't you begin to see a pattern?

Posted by James Wimberley | July 28, 2008 9:04 AM

Mixner: none of the cited polls in the last three weeks shows McCain leading Obama. The Intrade betting contract for Obama to win is at 64.1, for McCain at 31.7. Don't you begin to see a pattern?

I see a persistent but small and volatile lead for Obama. That suggests the most likely outcome in November, on the basis of current information, is a modest win for Obama. But it doesn't provide any basis for having much confidence in that result. It's a close race and could easily go either way.

By the way, Rasmussen's daily tracking poll for today has McCain gaining another point and Obama losing a point. So Obama's lead is back down to 3 points. Gallup also has Obama losing a point. So much for the "landslide."

Posted by Mixner | July 28, 2008 2:37 PM