Nate Silver wrote a very good post yesterday about the ability of economic models to forecast election results. I think he does a good job of debunking the idea that real economic fundamentals are the unquestioned major factor predicting elections. However, I have several major disagreements with his post. They are very theoretical so I apologize if this post gets overly wonky. But some serious clarification about political science theory and findings is necessary to truly understand how the economy affects elections; and, how it is still very relevant despite Silver's post. Before I get into the theory behind economic voting, I want to make a quick methodological point since Silver’s post is primarily empirical (feel free to skip this paragraph if you hate statistics). Silver uses margin of victory as the dependent variable to analyze most of his hypotheses. That’s a fine DV but it’s an odd assessment for picking a winning candidate. If you are trying to predict the winner, why care about how much they won by? I think leaning on the old coaching motto, “A win is a win is a win” is probably more appropriate. Political science dedicated to forecasting elections follows this mantra (though I doubt they derive their theory from sports). Most models use individual vote choice, not share of two party vote (as Silver does later in his post) or incumbents margin of victory. So the best political science on this subject uses actual individual vote choice and then predicts electoral outcomes based on economic conditions. Silver, by shrinking the DV down to margin of victory reduces its variance, making it more difficult for other variables to explain. Similarly, using two-party vote share gives you a different estimate for electoral fortunes, but it lacks a real cut-point for victory. Lincoln won with 37% of the vote. More recently, Clinton won with 43% of the vote in ’92. Presidential elections are often won by a plurality of popular vote, not a majority. So it’s not cut and dry that these are the best variables to use to predict winners.
Despite this, my bigger problem with the post is its lack of theory. Voters don’t often base their vote on actual information or facts, but on their perceptions' of that information. For example, this means is that voters will not specifically vote based on the nation’s GDP, but will base their voting decision on whether they believe GDP growth is strong or weak. A standard survey question for economic voting is something along the lines: “Looking back over the past year, would you say the national economic situation has gotten worse, better, or stayed the same?” Because these questions are based on voter perceptions and not real economic indicators, there is a level of subjectivity that plays into these forecasts. However, actual economic factors fail to predict election results as well as “fake” economic indicators (i.e. voter perceptions). For example, real disposable income does not explain as much as voter perceptions of the economy (Nadeau and Lewis-Beck 2001). This may seem ridiculous but it’s what we should expect. We know voters, on the whole, lack a lot of information and knowledge. It’s safe to assume that most take their cues from opinion leaders like politicians, the media, etc. So it makes sense that their vote leans toward their perception of economic conditions and not how real economic indicators are actually performing. Pulling this back into Silver’s analysis, real GDP, actual unemployment figures, and exact inflation levels are relevant only to the extent that voters perceive these figures as good or bad. For example, if unemployment numbers are bad but have improved from the previous month or year, voter’s may perceive the numbers to be better, which would influence their vote choice in favor of the incumbent (or vice versa).
This all points toward something that I, at various points on this blog, have tried to highlight. That is, the effect of politics and positioning on issues that affect vote choice and outcomes. The actual economic numbers may be bad, but if politicians can persuade voters that they are better than before or gaining momentum, they can leverage the issue in their favor. Lynn Vavreck (2009) has a great book on how presidential candidates are able to influence elections by highlighting good economic conditions or downplaying bad economic conditions (The Message Matters). While presidents cannot overcome all environmental conditions or external factors, to an extent they can influence how the public perceives those factors. Taking positions or making a case for their policies can affect the election.
So on the whole, Silver’s models take real economic numbers too seriously. As Jon Bernstein points out, that is in part the fault of political scientists, this blog included. In trying to illustrate the importance of the economy we may have simplified the issue too much. This runs the risk of making it easier to forget the nuances and caveats that make the issue rich but more difficult to understand. While economic conditions do play an important role in predicting elections, they are limited because economic conditions don’t have a vote. So when looking at these variables, we have to estimate how voters believe these conditions are performing rather than their actual performance. That’s a lot trickier, requires much more money (i.e. surveys/polls), and something that just isn’t done enough outside pundits’ random guesses on consumer confidence.
Silver is right though. Other factors besides the economy play a role in elections; and a significant role at that. But should we beware that economic models don’t explain that much? Hardly. We (as a community of public contributors) need to articulate a better understanding of the theory behind the statistics.