The Ecological Fallacy

May 10, 2010
Posted by Jay Livingston

The ecological fallacy is alive and well. Ross Douthat, the New York Times’s other conservative (the one that isn’t David Brooks), breathes life into it in his op-ed today on Red Families v. Blue Families, the new book by Naomi Cahn and June Carbone.

First Douthat gives props to the “blue family” model:

couples with college and (especially) graduate degrees tend to cohabit early and marry late, delaying childbirth and raising smaller families than their parents, while enjoying low divorce rates and bearing relatively few children out of wedlock.

Then there’s the “red family” for whom the stable, two-parent family is more a hope than a reality:

early marriages coexist with frequent divorces, and the out-of-wedlock birth rate keeps inching upward.

Blue looks good – good for the couples, good for the kids, good for society. But Douthat finds a moral thorn among the blue roses – abortion.

The teen pregnancy rate in blue Connecticut, for instance, is roughly identical to the teen pregnancy rate in red Montana. But in Connecticut, those pregnancies are half as likely to be carried to term.

So it isn’t just contraception that delays childbearing in liberal states, and it isn’t just a foolish devotion to abstinence education that leads to teen births and hasty marriages in conservative America. It’s also a matter of how plausible an option abortion seems, both morally and practically, depending on who and where you are.

Douthat is channeling Balzac: Behind every great fortune lies a great crime. Behind every more modest fortune – say, enough to live in Danbury if not Greenwich – is a more modest crime, i.e., an abortion or two.

But here’s the fallacy: Douthat makes it appear that the Connecticut residents who are getting those abortions are the same “couples with college and (especially) graduate degrees” we met in the paragraph on blue families. The illogic goes like this:

Blue states with higher levels of income and education also have higher levels of abortion than do Red states.
Therefore more Blue chip people have more abortions than do Red necks.

No, no, no (I hear myself repeating to my students). You cannot assume that a correlation at the state level also exists at the individual level. Just because wealthier states have higher rates of abortion, you cannot assume that wealthier individuals have higher rates of abortion. To make that assumption is to commit the ecological fallacy.

In fact, the Connecticut women who are getting abortions may also be relatively poor and uneducated. The difference is that abortion may give them access to further education or employment – not a graduate degree and a 6-figure job, but something better than what they could expect were they in Alabama. Or Montana.

The Uses and Abuses of Surveys

May 10, 2010
Posted by Jay Livingston

Ask a silly question, you get a silly answer. Ask a politically loaded question, you get a political answer – even if the literal meaning of your question seems to be asking about matters of fact and not opinion..

Here are eight questions from a Zogby poll. Respondents were given a Likert scale from Strongly Agree to Strongly Disagree, but the authors treat answers as either correct or incorrect according to basic economic principles.

1. Restrictions on housing development make housing less affordable.
2. Mandatory licensing of professional services increases the prices of those services.
3. Overall, the standard of living is higher today than it was 30 years ago.
4. Rent control leads to housing shortages.
5. A company with the largest market share is a monopoly.
6. Third-world workers working for American companies overseas are being exploited.
7. Free trade leads to unemployment.
8. Minimum wage laws raise unemployment.

Respondents were also asked to classify themselves on a political spectrum – Progressive, Liberal, Moderate, Conservative, Very Conservative, Libertarian.

This survey wasn’t designed to discover what people think. It was designed to prove a political point: “The Further Left You Are the Less You Know About Economics.” That’s the title of a post about it at Volokh Conspiracy. A paper by Zeljka Buturovic and Dan Klein, who designed the survey, gives the results.

(Click on the image for a view large enough to actually read)

The results were similar for the other questions.

To be sure, the liberals view of economic cause-effect relationships reflects the way they would like the world to be rather than the way the world actually is. But the bias of the poll is obvious. As monkeyesq says in his comment at Volokh,

1. Pick 8 liberal positions that have a questionable economic basis;
2. Ask people whether they “agree” or “disagree” with the statements;
3. Find that liberals are more likely to support liberal positions;
4. Claim that liberals don’t understand economics.

There’s an even larger problem here – a problem that affects not just polls that have an obvious ax to grind,* but a basic problem of all survey research: the question the survey asks may not be the question the respondent hears or answers.

These eight questions have a literal meaning. As Todd Zywicki, who wrote the Volokh post, says, “Note that the questions here are not whether the benefits of these policies might outweigh the costs, but the basic economic effects of these policies.”

True, the questions do not ask about costs and benefits, although I don’t think that the survey included an explicit caveat like the one Zywicki adds after the fact. Still, we have to wonder about how people really heard these questions.

“Mandatory licensing of professional services increases the prices of those services” – Agree or Disagree? Maybe some people hear a different question, a question about policy implications: “Would you like cheaper, but unlicensed, doctors.”

“A company with the largest market share is a monopoly.” Maybe the what the person hears is: “Can companies with large market share – though less than the share required for it to be a monopoly (100%?) – still exercise monopolistic powers?”

As for the “exploitation” of third-world workers, the word may have a precise economic definition (e.g., it’s exploitation only if the worker has no choice) – I don’t know. But even if such an economic definition exists, to most people the word evokes moral judgment, not economics.

The other items also have flaws, as some of the comments at Volokh (now 200 and counting) point out. (I confess that I’m still puzzled by the responses to Standard of Living. Nearly a third of all the respondents think that the standard of living today is no better than it was 30 years ago – 55% on the left, 12% on the right 21% of libertarians.)

The survey may tell us that “epistemic closure” is a disease that can infect the left as well the right. But it also tells us to be cautious about interpreting survey questions literally. Even innocuous questions may mean different things to survey respondents. Until a question has been tested several times, we can’t be sure what respondents hear when they are asked that question.

*A Kos poll that set out to show that quite a few Republicans were extremist nuts suffers from a similar problem. I blogged it here.

How Genetics Works (borrowed post)

May 8, 2010
Posted by Jay Livingston

Flaneuse at Graphic Sociology reposted this from some place. I couldn't resist reposting it as well.

(Click on the image for a slightly larger view.)

I know nothing about the photo. It reminds me of Elliot Erwitt, though I’m sure it’s not.
It looks too old to have been photoshopped.

Lies, Damn Lies, and Statistics About Lies

May 7, 2010
Posted by Jay Livingston

We all know that Barack Obama is a Muslim who was born in Kenya and that his campaign was funded by Hugo Chavez.

Yes, I know, presidents going back to Washington have been the subject of rumors. But it seems that with the arrival of Obama, presidential rumors have become something of a growth industry (maybe our only growth industry these days.)

J.L. Bell blogs mostly about comic books and fantasy literature at his Oz and Ends. But two weeks ago he posted some numbers about presidential rumors on the Internet.* For his data, he went where most of us would go – Snopes.com. Here’s what he found.

(Click on the graph for a larger view.)

In less than two years, Obama rumor-mongers have had nearly twice the output that their Bush counterparts managed in eight years – 87 to 47. And while the Bush rumors split almost evenly true-false, false Obama rumors dwarfed the true ones.** The false rumors about Obama outnumbered the total number of rumors about Bush. And while the lies about Obama are almost all negative, some of the false rumors about Bush are quite flattering, along the lines of the George Washington cheery tree rumor – like the rumor that had Bush paying for the funeral of a boy who had drowned near the Crawford ranch.

Is there really a right-left difference? If the “epistemic closure” hypothesis is accurate – if conservatives, even the chattering intellectuals, live and write in a bubble that keeps out any realities that might conflict with their ideology – then conservatives of all sorts might also welcome into the bubble even the most preposterous and unfounded rumors.

Surely, there must be a sociology of rumor. What are the demographic correlates, if any? What are the conditions under which rumors are more likely to arise and spread? I would imagine that lack of trust is important. The less we trust others who are outside our relatively small circle, and the less we interact with them, the more likely we will be to rely on rumor.

Trust, at least trust in government, has been decreasing generally, but conservatives, when they are not in control of the government, are especially mistrusting. Under any circumstances, false beliefs are frustratingly resistant to facts. It probably doesn’t improve conservative’s grasp of reality when they have a major TV network giving airtime to these rumors and when their leaders tell them about death panels.


* His post was picked up by Salon.com and then by Brendan Nyhan, who has had some interesting posts on the epistemic closure discussion.

** The other categories were “mixed,” “undetermined,” and “unclassifiable.” In the graph, I collapsed the latter two categories into “other.”