Voter turnout in US presidential elections: does Carville’s law explain the time series?

Abstract

We estimate a time series model of voter turnout for 34 US presidential elections, 1880–2012. Employing a variety of econometric techniques, our major results are as follows. (1) A negative and significant structural shift in voter turnout occurs in 1972 and is too large to be explained by the lowering of the voting age. (2) The 1972 shift is the only statistically significant structural shift to occur since the first decade of the twentieth century. (3) Short-term macroeconomic conditions significantly impact turnout, with unemployment having a positive effect. (4) Turnout in recent presidential elections has not deviated significantly from the post-1972 norm. (5) Turnout is positively related to the expected closeness of the election outcome, but contrary to some theoretical predictions, closeness exhibits no trend over time.

Keywords:

• turnout
• voting
• macroeconomics
• institutions
• time series

JEL Classification:

• D72
• D70

Notes

¹ An exception is Radcliff (1992), but his time series stops in 1988 and his analysis does not consider as many explanatory variables as does the present study.

² An exception is an early study by Arcelus and Meltzer (1975) that employs a time series of US Congressional elections.

³ An alternative series compiled by Walter Dean Burnham (Stanley and Niemi 2001) is essentially equivalent to Rusk’s.

⁴ We follow Fair’s practice of setting GoodNews equal to zero for 1920, 1944 and 1948, since economic performance in the years prior to these elections was distorted by war.

⁵ We obtained unemployment rates from the following sources: 1948−2012: Bureau of Labor Statistics; 1932−1944: Historical Statistics of the United States; 1892−1928: Romer (1986); 1880: Fairlie and Sundstrom (1997). We backcasted estimates for 1884 and 1888 by regressing Romer’s series on detrended real GDP for 1890−1914.

⁶ Most researchers define their closeness variable similarly to ours. Cox and Munger (1989) and a few other studies take an alternative approach by using the absolute (not percentage) vote differential. Using absolute votes, however, introduces a problem of endogeneity, because higher turnout will tend to inflate the absolute differentials. Some cross-sectional studies use differentials in party registration, which offers the advantage of measuring closeness ex ante. Party registration, however, correlates much less closely with voting results at the time series level, and in the early part of our particular time series, registration does not even exist.

⁷ Use of the logit is necessary to obtain unbiased estimates because otherwise we cannot plausibly assume that the conditional expectation of the error term, E[u_t|X], equals zero for all values of the explanatory variables.

⁸ The models that do include Southern turnout nevertheless offer the advantage of aligning nationwide turnout with our macroeconomic variables, which are also nationwide.

⁹ Caporale and Grier (2005) argue that the Bai and Perron statistical methodology is a powerful tool for analysing political economy models using time series data.

¹⁰ For modern evidence of the relevance of registration rules, see Timpone (1998).

¹¹ The marginal effects for our logistic regression are given by β_iTurn_t(1−Turn_t). The dependence on Turn_t implies that the marginal effects vary over time with the values of the explanatory variables, but given the range of sample values, this variation is not very large. At the median of each turnout series, the marginal effects can readily be obtained by multiplying the coefficients in Table 1 by .24 for national turnout and .22 for non-South.

¹² Another possibility is that inflation could influence turnout through redistributive effects. In particular, inflation favours those with fixed-rate debt such as homeowners, a group that might also have a relatively elastic turnout rate. A prominent historical example can be found in the late nineteenth century, when a significant political movement arose to agitate in support of free coinage of silver. The free silver policy was designed intentionally to redistribute wealth towards certain interest groups by increasing the rate of inflation.

¹³ See, for example, Trende (2013).