Winning is Losing Unless You Win

Barack Obama's lead in the polls is good news for his prospects, right? Wrong, says Gallup, who points out that the July polling leader lost in six of the last nine competitive US presidential elections. It seems to me that Gallup is generating a spurious level of counterintuitiveness through use of the "competitive" qualifier. If you look at the most recent fifteen presidential elections, the July polling leader has won big six times, won narrowly three times, and lost narrowly six times. That gives you the totally intuitive result that leading in July is better news that losing in July, but that it doesn't guarantee anything.

Comments

It seems to me that Gallup is generating a spurious level of counterintuitiveness through use of the "competitive" qualifier.

It seems to me that Gallup is generating a strong case that July polls don't mean s***, and any analysis performed with them is doomed to fail under the weight of their inaccuracy.

Posted by LFC | July 9, 2008 10:26 AM

The number of other mistakes are also obviously quite glaring: in 2004, we had a much wider picture of the race, and Kerry was not up 7.

I think the biggest thing this whole "look at past polling" idea misses is the fact that people know a lot more about candidates now than they did in any other July past. 24 hour cable news, a long primary for one party, and a famous candidate in the other.

I wish Gallup would tell us what the name recognition and "strong view" numbers were in July in the past elections.

Posted by Joel W | July 9, 2008 10:28 AM

This election is taking place in an environment with more widespread political communication than any of the elections listed above. The model has changed, and so it's spurious to claim anything from these charts at all.

Even if that weren't the case, though, you're right - it's nonsense for Gallup to select data points to try and make a curve that fits their point. Of course, we can look forward to plenty of "if you look at it right, it's a close race!" this fall. Nothing's harder than getting somebody to understand what they're paid not to understand, and that's as true for our political media as any lobbyist.

Posted by Jon O. | July 9, 2008 10:41 AM

No one is ever going to say he's really ahead, period. We all know this. Blow-outs are no fun for the media...they have to do everything they can to keep the narrative as a tight-tight-tight race?

Did everyone see that AP poll Pollster ripped apart analyzing results by PET OWNERSHIP (apparently McCain's people are dog people)? I guess the lows will never end. What's next? Spending 5 figures in polling to figure out the political leanings of PB&J eaters?

Posted by evie | July 9, 2008 10:42 AM

What a wretched case of cherry-picking the data. I wonder what kind of look they'd give you if you asked them for their mathematical definition of "competitive." 1996 ("not competitive") was about as close, in terms of percentages, as was 1988 ("competitive"), for instance. This thing is a joke.

Posted by Craig | July 9, 2008 10:51 AM

well hopefully we can take this "it's close" stuff to heart and work our asses off to make it a true blowout in november. and wouldn't it be an extra-sweet blowout if "no one" saw it coming?

Posted by john b | July 9, 2008 10:51 AM

"Competitive" also seems to be defined by the final result. 1976 wasn't competitive in July when Carter led by 33, but was in November when he won by 2. Conversely, Reagan was up 10 in July 1984 so at this point in the race 1984 was more competitive than 1976. So by this definition we don't know yet whether 2008 is a "competitive" election. Conveniently, if Obama wins in a landslide it doesn't count against Gallup's theory because the race will then be categorized as non-competitive. It would be more useful to see if there was a change in results depending on whether the race was competitive in July since that is all we currently know about the 2008 race, but that might contradict Gallup's clever little claim that leading in July is actually bad news for a candidate.

Posted by ibid | July 9, 2008 10:53 AM

You also have to take into account the election fundamentals. How many underdogs came from behind in July to win? It seems to me that if the heavy favorite is up by 10 at halftime, that is a very different situation than the underdog being up by 10 at halftime.

Posted by jc | July 9, 2008 10:58 AM

Maybe the real conclusion to draw from that data is that Gallup is a sucky pollster.

Posted by Glenn | July 9, 2008 11:05 AM

Bush and Gore were statistically tied in July, and statistically tied on Election Day. In '92 there was Perot.

So this very interesting in-house statistic from Gallup has, in the last twenty years, been relevant, twice.

Fascinating.

Posted by Jim | July 9, 2008 11:08 AM

Or, maybe it means that if Obama is up about five now, in November he'll win in a landslide, which hasn't happened in a while.

Posted by cnnr | July 9, 2008 11:11 AM

Ford looks like the real comeback kid. Up 31%! Wow, hats off to him. Hope people who claim Obama is Carter II are wrong, though this is one more indicator why Republicans hope so.

Posted by stefan | July 9, 2008 11:15 AM

Craig is right, Clinton beat Dole by 8.5% which is more competitive than 1980 and about the same as 1988 (and was much more competitive than either in terms of electoral votes). So the criterion for "competitive" is apparently completely subjective. I can understand writing off 1964, 1972, and 1984 as landslide years, but beyond that there has to be some sort of objective measurement. I also like Glenn's point: it seems odd that Gallup is cherry-picking data in order to suggest that we shouldn't trust their polls.

Posted by ibid | July 9, 2008 11:26 AM

A significant part of your appeal is being a MOTO.

Posted by mickslam | July 9, 2008 11:46 AM

There's still a considerable regression to a close election. Nate covered it here.

Posted by Alex | July 9, 2008 11:48 AM

Matt is correct. And I have a pretty scatter plot and regression to prove it!

http://elvaliente.wordpress.com/2008/07/09/its-better-to-be-ahead/

Posted by Julio Gonzalez Altamirano | July 9, 2008 11:52 AM

Where are the '64 numbers?

Posted by Nasara | July 9, 2008 12:27 PM

The only thing more ridiculous than assuming the July poll numbers are a true representation of the November vote is the idea that the July poll numbers are anti-correlated with the final vote.

9 data points is a pathetically small sample size from which to draw any conclusions, especially when the scatter is so high. Gallup presumably has some mathematicians on staff who could explain this.

Posted by Jinchi | July 9, 2008 2:38 PM

Different elections have different results.

Correlation is not causation.

Most Presidents' names end in -n, ergo McCain is a shoe-in.

Poll-takers should know better than to propagate such nonsense.

Posted by AdamK | July 9, 2008 4:31 PM

"It seems to me that Gallup is generating a strong case that July polls don't mean s***, and any analysis performed with them is doomed to fail under the weight of their inaccuracy"

AS per Glenn, it seems to me that Gallup is generating a strong case that Gallop polls don't mean shit.

Posted by Ginger Yellow | July 10, 2008 7:46 AM