Elections and the Ecological Fallacy, Part II

Every now and then, people try to learn something about individual voting patterns by looking at regional voting patterns. Given that we have post-election surveys, this approach has always puzzled me. More worryingly, it’s plagued by the ‘ecological fallacy’: aggregating things up doesn’t always give you the same result. I’ve blogged about this before, but here’s a great example of where the ecological fallacy can lead you astray.

Aggregation Reversals and the Social Formation of Beliefs by Edward L. Glaeser, Bruce Sacerdote
In the past two elections, richer people were more likely to vote Republican while richer states were more likely to vote Democratic. This switch is an aggregation reversal, where an individual relationship, like income and Republicanism, is reversed at some level of aggregation.  Aggregation reversals can occur when an independent variable impacts an outcome both directly and indirectly through a correlation with beliefs.  For example, income increases the desire for low taxes but decreases belief in Republican social causes.  If beliefs are learned socially, then aggregation can magnify the connection between the independent variable and beliefs, which can cause an aggregation reversal.  We estimate the model’s parameters for three examples of aggregation reversals, and show with these parameters that the model predicts the observed reversals.

Bottom line: if you want to know how old people vote, ask an old person. If you want to know how rich people vote, ask a rich person.

This entry was posted in Australian Politics, US Politics. Bookmark the permalink.

4 Responses to Elections and the Ecological Fallacy, Part II

  1. Bryan says:

    Andrew, there is a middle ground that takes some data sets and data-mines for relationships using ecological analysis. Where those relationships appear to exist ecologically, test them as you suggest with surveys to validate.

    I am comfortable with ecological analysis in this context. It is not all bad; it is only bad when someone presents an ecological study as conclusive proof of an actual relationship.

  2. Andrew Leigh says:

    Bryan, I’m not clear what ecological analysis adds. Are there any interesting questions for which we wouldn’t go first to the Australian Election Study?

  3. Bryan says:

    Andrew, I am not arguing ecological analysis ahead of individual analysis. Nonetheless, there are sub-population questions where the sample sizes in the AES aren’t large.

  4. Andrew Leigh says:

    Bryan, this is theoretically true. But in practice, I’ve never seen such an ecological analysis. For example, the Stimson study you blogged about recently didn’t seem to ask any questions you couldn’t ask with AES data (and get a better answer).

Comments are closed.