Forecasting in reverse: Can we use election returns to learn about economic history?

We all know, following the research of Rosenstone, Hibbs, Erikson, and others, that that economic conditions can predict vote swings at state and national levels.

But, what about the reverse? Could we deduce historical economic conditions from election returns? Instead of forecasting elections from the economy, we could hindcast the economy from elections.

Would this make sense as a way of studying local and regional economic conditions in the U.S. in the 1800s, for example? I could imagine that election data are a lot easier to come by than economic data.

P.S. Don’t forget that there have been big changes over time in our impressions of the ability of presidents to intervene successfully in the economy.

11 thoughts on “Forecasting in reverse: Can we use election returns to learn about economic history?

  1. Wow, what a great idea. Some of my stuff tries to explain patterns of state and federal military intervention in America (labor, race and political violence between Reconstruction and WWII), and I had a tough time finding state-level economic data. NBER has some historical state series, but they're mostly yearly national figures. Old PDF versions of the statistical abstract of the census has state-level economic data on railroad track miles and coal production, which I've used as proxies for state economic development, but they only go back to 1878. Creating a consistent measure of state electoral changes as a stand-in for economic data would be a real help.

  2. Let me explain with one example. From 1896 to 1924, the Democratic presidential candidate received over 60% of the vote in Alabama in every election. In fact, except for 1900 (the second William Jennings Bryan match-up with McKinley), the Democratic candidate never dropped below 67%. In 1928, the Democratic candidate dropped to 51%. I defy anyone to explain those results based on economic factors.

  3. Good point. I guess the question is what can be learned about the economy after adjusting for non-economic factors such as ethnicity, incumbency, and regional differences.

  4. Sounds hard. This may be like the paper on search you blogged about a couple of days ago — this indicator may add relatively little to, say, the annual income by state data already available.

  5. Andrew,

    I've been meaning to ask you something. What kind of evidence would it take to convince you that economic conditions don't predict vote swings? I'm really serious about this. After making some careful analysis, I think the correlation is mostly an artifact of other factors. I'm not saying economic conditions don't have any impact, just that it's not very useful in a predictive sense. But I'm not sure how to marshal my arguments in a way that make sense to an open-minded political scientist or statistician.

  6. Andrew: I thought there were longer series on income, but I can't find them in a quick search.

    Maybe the federal government only cared about income once the income tax became a major source of revenue?

    http://www.bea.gov/regional/docs/spi2007cd.cfm
    This provides personal income data by state 1929-2007.

    In "Statistical Abstract Colonial Times to 1970"
    there are also only tables by state back to 1929 (Source: Bureau of Economic Analysis).

    There are price series there back to 1800; I didn't check how far back the labor series went.

  7. William: I'll refer you to the work of Bob Erikson, Steve Rosenstone, and Doug Hibbs, all of which who have written about this. The short answer is that the economy is useful as a statistical predictor, even if it is not perfect.

Comments are closed.