Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Are Americans really jingoistic yahoos? A cautionary statistical tale

April 13th, 2014, 9:36am by Sam Wang

It can be good for a laugh to view U.S. citizens as people who don’t have a clear idea of what’s happening outside their borders…but still want to launch a military attack. This week in The Monkey Cage comes a finding that feeds that view. The headline is impressive: “The less Americans know about Ukraine‚Äôs location, the more they want U.S. to intervene.”

But how big is the reported effect, and what does it really say about American attitudes?

In teaching my students how to evaluate research with a skeptical eye, I usually make a distinction between two ways to look at a discovery:

  1. Is the finding true for large groups, averaged? If so, it is a discovery and publishable in an academic journal.
  2. How large is the effect? The larger it is, the more likely it is to matter in a practical sense.

In many domains of research, ranging from medical research to political science, it is easy to confuse these points of view, especially when all you read is the summarized result. In the case of the Ukraine survey, which point of view informed the headline?

First, let’s consider the result that has led to such merriment. In a sample of 2,066 Americans:

  • About one in six (16 percent) Americans correctly located Ukraine, clicking somewhere within its borders.
  • Only 13 percent of Americans supported using force.
  • The less accurate our participants were [at finding Ukraine on a map], the more they wanted the U.S. to use force…[this effect is] statistically significant at a 95 percent confidence level.

At face value, this sounds pretty bad: there is a statistically significant difference between those who can find Ukraine on a map, and those who can’t. The implication is that ignorance leads to a desire to intervene militarily.

However, is that really true? First, note that there isn’t a massive groundswell for an invasion; 13% is quite small. But here is something that is harder to appreciate: it does not take a large difference between groups to get statistical significance. Assuming there is large difference is a common logical pitfall.

For example, imagine two equal-sized groups: those who can find Ukraine on a map, and a second group thinking it’s somewhere in the vicinity of Nebraska. Given the size of the original sample, 331 people found Ukraine correctly. If attitudes toward military intervention broke down as follows (oppose vs. suport), the following overall pattern would be enough to reach 95% statistical significance (one-tailed Fisher exact test).

#opposing #supporting %supporting intervention
Can find Ukraine 298 33 10.0%
Can’t find 283 48 15.4%

These are just made-up numbers, since I don’t have access to the data. But even if the difference is larger, I think the point is likely to be similar.

The researchers have succeeded in finding that desire-to-intervene in Ukraine is correlated with geographical ignorance of where it is. As they point out, this is consistent with past findings of a similar nature. But it could be driven by a difference as small as that between 10% and 15% – which does not seem all that notable.

Basically what I’m saying is this: nobody really wants to invade Ukraine, and it seems counterproductive to beat up people who can’t find it on a map. After all, if someone thinks Ukraine is next to Kazakhstan, yet shares your policy preference, is it advisable to cast aspersions on that person’s intelligence?

For more about the perils of “95% statistical significance,” check out this excellent Nature article on p-values and effect sizes.

Tags: Politics

4 Comments so far ↓

  • Gibby the Fifth

    My boss taught me another lesson about averages – always look at the range as well as the mean. To illustrate, he pointed out that the average American has one tit and one ball.

  • Adam Eran

    Worth a look for a little bit of context beyond “Hey! Can you find Ukraine on a map?”

  • Amitabh Lath

    It’s 2 sigma. Probably much less if you take into account the “look elsewhere” effect. In looking for a signal you have to take into account all the possible observations you would count as signal.

    So if you are looking for the Higgs boson along a spectrum from 100 to 200 GeV in 2 GeV bins, you have to account for a fluctuation in any bin.

    Or if you are asking the public say, 20 questions, you have to allow for the fact that any pair could show correlation just by fluctuation. For instance, instead of can/can’t find Ukraine on a map, the correlation could have been by young/old or military/civilian, or boxer/briefs, whatever. You have to account for the fact the correlation could have been anywhere. That’s the look elsewhere effect.

  • bks

    You’d have to be awfully stupid to want to attempt a military intervention in Ukraine even if you could sketch a street map of Simferopol blindfolded. –bks

Leave a Reply to Gibby the Fifth (Cancel)