Ebeid and Rodden: Economic geography and economic voting
From WikiSummary, the Free Social Science Summary Database
Ebeid and Rodden. 2006. Economic geography and economic voting: Evidence from the US states. British Journal of Political Science 36: 527-547.
Several studies have asked whether governors are held accountable for state economic performance, or whether they are merely judged based on their party's national performance. Ebeid and Rodden argue that state economic performance does matter, but only if the structure of the state economy is such that the governor can reasonably held accountable for its performance.
For example, the governor of an agricultural state can do little to boost income growth; instead, growth depends on the weather, market movements, federal subsidies, and transportation costs. By contrast, the governor of an industrial state can do quite a bit to attract capital, promote local industries, and so forth. And governors do spend quite a bit of effort promoting their state's products around the country and internationally. Are governors mistaken about how voters evaluate them? As it turns out, they are not.
First, the authors follow the literature's lead by looking for an effect of state economics on the incumbent's vote share. Using a time series of postwar gubernatorial election outcomes, they find that state economics--whether measured as growth or as unemployment, and whether measured in absolute terms or in relation to the national economy--do not matter.
But then they repeat this analysis, interacting the state economic variables with a measure of how much of the state's economy depends on primary products (agriculture, mining, logging, etc). Although absolute measures of state economic performance remain insignificant, the relative measures become statistically significant. Their effect is small--perhaps giving the governor 1 or 2 extra points at election time--but strong enough to make a difference in a close race.
Comments and Criticism
In their enthusiasm at finding a significant result, Ebeid and Rodden fail to notice that the interaction doesn't seem to "fit" the data any better than the non-interactive model. R-squared went up from 0.31 to 0.32 when examining income growth, and from 0.39 to 0.40 when examining unemployment. Clearly, there is still much to explain.