The Ecological Fallacy and the Not So Great Divide

May 6, 2008
Posted by Jay Livingston

Early in the semester, I try to teach the ecological fallacy. Students find correlations of state-level variables and try to come up with explanations. But, I warn them, you can’t infer facts about individuals from facts about states. As an example, I use the fairly strong correlation between the Bush vote in a state and its suicide rate. It can’t be because voting for Bush makes you more likely to commit suicide, I say, nor can it be because those who committed suicide were then more likely to vote for Bush. (Many easy jokes to be had here.)

Many students get it. David Brooks doesn’t. Here’s an excerpt from last Friday’s column.
In the decades since [1958], some social divides, mostly involving ethnicity, have narrowed. But others, mostly involving education, have widened . . . .The college educated and non-college educated are likely to live in different towns. They have radically different divorce rates and starkly different ways of raising their children. The non-college educated not only earn less, they smoke more, grow more obese and die sooner.
In this paragraph, Brooks is talking about differences between individuals — more educated compared with less educated. In the next paragraph, he extends this analysis from smoking and obesity to voting preferences.

This year’s election has revealed a deep cultural gap within the Democratic Party. In state after state (Wisconsin being the outlier), Barack Obama has won densely populated, well-educated areas. Hillary Clinton has won less-populated, less-educated areas. For example, Obama has won roughly 70 percent of the most-educated counties in the primary states. Clinton has won 90 percent of the least-educated counties. In state after state, Obama has won a few urban and inner-ring suburban counties. Clinton has won nearly everywhere else.
Counties with higher levels of education have a higher Obama vote. Brooks explains this county-level correlation in terms of individual differences in education. As John Sides at The Monkey Cage points out, Brooks is committing the ecological fallacy. Exit polls, which survey individuals, show that in Pennsylvania Clinton beat Obama among both the college educated and those without college degrees.

The results give some support to Brooks. Though Sides does not mention it, Clinton’s margin was much greater among the non-college voters (16 points vs. 2 points).

But Sides has other data that show that among Democrats
  • the differences between these two groups are very small
  • the gap between them has not widened

Here, for example, is the graph of Democrats voting for the Democratic presidential candidate. The only year with a big difference was 1972, the McGovern debacle.
If you know someones level of education, you can make a better guess as to their BMI or whether they smoke. But it will not allow you to make a better guess as to whether they prefer Clinton or Obama. If you use information about the average education level of counties to make statements about individuals, you are committing the ecological fallacy.

No comments: