Wednesday, March 20, 2013

Brad DeLong Versus Political Science: Grasping Narrative with Both Hands

. Wednesday, March 20, 2013

DeLong, in Democracy:

If there was a single moment when Mitt Romney lost the 2012 presidential election, it was in May when he stood in front of the $50,000-a-plate audience at Sun Capital honcho Marc Leder’s home in Boca Raton and spoke his soon-to-be-infamous words:
"There are 47 percent of the people who will vote for the President no matter what…. There are 47 percent who are with him, who are dependent upon government…who believe that government has a responsibility to care for them, who believe that they’re entitled to health care, to food, to housing, you name it…. These are people who pay no income tax…. My job is not to worry about those people—I’ll never convince them that they should take personal responsibility and care for their lives…"
John Sides, at
To commentators used to thinking of campaigns like a boxing match, [the 47 percent video] seemed like the knockout blow.

In reality, the impact of the video was much more muted. This is the argument UCLA political scientist Lynn Vavreck and I make in our forthcoming book on the election, “The Gamble.” ... 
It is always tempting for those following a presidential election closely — pundits, reporters and political scientists alike — to assume that every new twist is the proverbial game-changer. But in retrospect, the 47 percent video did not live up to the hype.
Models predicting Romney's loss well before the video emerged -- those based on economic fundamentals -- performed very well. The 47% video had almost nothing to do with the electoral outcome.

The rest of DeLong's essay -- essentially a defense of the welfare state against the entreaties of Nicholas Eberstadt -- is fine enough if a bit boilerplate. But if he's going to deliver lectures on how folks should understand the basic tenants of macroeconomics if they are to comment on them, then he should understand the basic tenants of political science if he is to comment on it.


brad said...

The residuals in political scientists' regressions matter. That all of the residuals wind up summing to near zero doesn't mean that each of the components of the residuals is zero. The 47% was a big deal for elite opinion and grass-roots cadre mobilization.

As the Obama nowcast probability in the Silver model swung from 62% to 80% to 62% to 99.5% to 55% to 93% on election day, I remain unconvinced by your claim that each of the components of the residual was zero--and you should not be convinced either:

Kindred Winecoff said...

"The 47% was a big deal for elite opinion and grass-roots cadre mobilization."

No it didn't. At least, there's not much evidence that it did. Good time series analyses look at interventions into the series to see what the effect was: to see whether the residuals are white noise or whether there are effects which are durable: which have an impact on the whole future series. The was no such effect from the Romney video, at least according to the best evidence now.

The video was released on Sept 17. Eyeballing the graph you linked, on Sept 17 85-90% of the probability density had Obama winning. Shortly after, his odds *fell* (although I think almost all of that was noise... almost the entire series fluctuates around 75%). Are you saying that the effect of the Romney video was first to increase Obama's chances of winning by 5 or so percentage points, then to decrease them by 30 percentage points, and then to increase them again by 30 percentage points?

The association you believe is there is not there.

brad said...

You do know that this isn't an efficient market event study? That the direct and indirect effects of things like "47%" or debate #1 diffuse gradually through the voter population? Proprietary materials I have seen do show that as many people think they voted for Obama over Romney because of 47% as think they voted for Romney over Obama because of debate #1. But you seem to know better...

Brad DeLong

Kindred Winecoff said...


I don't think debate #1 mattered either! It couldn't've mattered much since Obama lost that debate badly but won the election handily.

But it's not that I know better. I just take the word of my fellow political scientists who study elections, public opinion, and voter behavior systematically and carefully. Folks like John Sides, but also folks in my department (which might be the best in the country at analyzing these questions) like Jim Stimson. And to a person they tell me that electoral ephemera like this does not have a big impact on outcomes. In an election as lop-sided as this one, the outcome in no way hinged on Romney's 47% speech. Or the debates.

Perhaps ironically, the reason for this is that not much of the electorate is in play; Romney was right about that -- there is 45-47% of the country that would never vote for him, and roughly the same percentage that would never vote for Obama -- even if he was wrong about the reason why. The 6-10% of the electorate which is truly independent tends to vote on big-ticket issues like the economy and wars. The economy was improving last year, and Obama was extracting the US from wars. So Obama won.

Latinamericanist said...
This comment has been removed by the author.
Latinamericanist said...

"Proprietary materials I have seen do show that as many people think they voted for Obama over Romney because of 47% as think they voted for Romney over Obama because of debate #1."

wow. Because behavioral research has shown that people are really good at identifying the causes of their actions. not.

(and the "proprietary data" - is that supposed to make it better? Like the proprietary data that led the Romney folks believe they were winning until the end?)

On the plus side, how bad de Long does once he moves from economics into polisci does drive home the relevance of his point on the division of labor from the Democracy piece.

Anonymous said...

instead of you and brad exchanging snarky comments, can you two agree on a decisive test or set of data ?

I work at a small biotech in boston; based on how my coworkers view the election, I would say that there were ~10% undecided, and they didn't pay much attention to the 47% remark (i know anecdotal small data set..)

Kindred Winecoff said...

Can we agree on a decisive test? The discipline of political science has. We peer review it and everything.

Brad's gut just knows better. Whatever.

It's really very simple: the only people who aren't partisans don't pay attention to politics. That's about 10% of the electorate (same as your office). They might have been affected by the 47% remark (who knows in what direction), but they probably never even heard it because they don't watch Fox News or MSNBC or whatever.

This is all well-established by reams of literature.

Robert said...

I note the first word in Brad's article "If".

There are also reams of peer reviewed literature on theoretical macroeconomics (some of it written by me). Peer reviewed reams do not demonstrate scientific legitimacy.

Frankly, I think the authors and editors of the literature you note better respond effectively to Nate Silver here or apologise and shut up (as should most economists not including Paul Krugman, Brad DeLong and lots of guys but definitely including Robert Waldmann who will now stop typing)

Kindred Winecoff said...

Political scientists love Nate Silver! Some might be a little jealous of him, considering that they've been doing similar things for awhile without getting as large of an audience, but on balance Silver is pretty popular among political scientists. You do know that political science isn't like Politico, right?

Of course I noticed the "If". It's a nice little hedge. But DeLong didn't start his article with that anecdote because he thought the video did NOT matter. It's a Politico-type move: hype something up way beyond proportion but then leave yourself a little out.

Regarding the question at hand, on the day the tape was released Silver downplayed its importance (because he's read the political science literature, note):

When the polls come out a week later, as Obama's NowCast went from winning very easily to winning slightly more very easily, Silver wrote this:

"There’s a case to be made that [the video] did damage Mr. Romney’s standing some."

This is not only an absurdly weak claim concerning the existence of any effect -- Silver's not saying there's any direct evidence that it *did* have such an effect, just that if you squint hard enough you can see a tiny bit of movement which could be attributable to the video (or maybe something else) -- it says nothing at all about its impact on the outcome of the election. Which was probably statistically indistinguishable from zero. It certainly was not the moment that Romney lost the election, since he was already losing the election, and had been losing the election since before the primaries began.

So I'll put Silver in the camp of those who believe the impact of the video on actual outcomes that matter was essentially nil. And I'll put DeLong and Waldmann in the anti-Silver/anti-political science camp on this issue.

Anonymous said...

You would be well advised, if you want your arguments to be taken seriously, to learn the difference between a "tenant" and a "tenet."

Ronan said...

Why exactly is knowing the difference between ‘tenant’ and ‘tenet’ necessary to having an argument on US electoral politics taken seriously? Especially when that argument, as stated above, is backed up by considerable evidence developed by a number of people who do (I would imagine) know the difference between ‘tenant’ and ‘tenet’.
Think logically dude

Dogen said...

How do poliscientists measure the opportunity cost of Romney's gaffe?

It seems to me that Romney spent a whole lot of time and effort trying to undo damage and respond to questions and it kept him off message and on the defensive frequently.

Whereas Obama's reportedly poor debate simply made him get more aggressive about his message.

So intuitively,anyway, it seems like these two events had very different costs to the candidates.

Of course, if Romney hadn't had to deal with the gaffe he might have spent his time/money/effort poorly, so it's kind of impossible to know.

but my point is that I have a hard time with the confident statement that research shows the gaffe had no effect on the outcome-- if you aren't able to measure the lost opportunity,how do you make that statement?

Kindred Winecoff said...

The baseline counterfactual should be twofold:

1. Was the outcome of the election at all different from what we would expect from models based on fundamentals?

2. What has been the effect on electoral outcomes of "gaffes" in the past (noting that this wasn't a gaffe)?

The answer to #1 is "not at all, this was a completely typical electoral outcome given the fundamentals". The answer to #2 is "almost nothing". Both have been found in dozens if not hundreds of studies, using a variety of methodologies, samples, etc. The results are remarkably robust, given the inherent contingencies in social systems.

So DeLong has a really really big mountain to climb here.

Brad DeLong Versus Political Science: Grasping Narrative with Both Hands




Add to Technorati Favorites