Taxes, Transfers, and Voter Behavior in U.S. Midterm Elections

Article excerpt

Abstract

The effects of changes in per capita real GDP, real taxes and real government transfer payments on midterm congressional election outcomes during the 1946-2002 period are examined. Voters are found to take all of these, except taxes and transfers at the state and local government levels, into account in casting their ballots. However, the weights they place on each are found not to be the same. Consequently, the common practice of summarizing the economic conditions faced by voters through disposable income seems to be inappropriate. Also, omission of tax and transfer variables from the vote equation, and using vote swing rather than vote share as the dependent variable is found to result in underestimation of the coefficient of per capita GDP growth. (JEL D70).

Keywords: congressional elections, economic voting, taxes, government transfers

Introduction

Whether economic conditions affect the US midterm congressional election outcomes is a highly controversial issue. Conclusions in this regard vary from study to study and depend on the way the vote equation is specified, in particular, how its dependent variable is defined. Tufte [1975] in his seminal study used, as the dependent variable, the standardized vote share of the incumbent party, defined as the deviation of the presidential party's vote share from the average of previous eight House elections. Regressing this variable on the growth rate of per capita real disposable income and the presidential approval rate, using data covering the 1938-1970 period (excluding 1942), he found both of these independent variables to be important determinants of midterm election outcomes. Later Tufte [1978] reached the same conclusion, analyzing the 1946-1974 period and getting substantially larger parameter estimates for his two explanatory variables. For many studies that followed Tufte [1975, 1978], his specification became the norm. It was adopted with only minor modifications. Hibbs [1982], for example, dropped the presidential popularity from the equation and used geometrically weighted average income to represent economic conditions in examining the 1946-1978 period, while keeping Tufte's standardized vote share variable as the left-hand side variable. He too found income to be an important factor in explaining midterm election results.

Erikson [1990], however, challenged the Tufte specification. He pointed out that using Tufte's standardized vote as the dependent variable in a regression is equivalent to running it with the presidential party's vote as the dependent variable, and adding eight lagged values of this variable on the right-hand side of the equation, but with the coefficients of the latter restricted to be 1/8. He argued in favor of lifting these implied restrictions and letting the values of these coefficients to be determined empirically. When he did that, using the data for the 1946-1986 period, the parameters of all lagged vote share variables, except the one for the previous election, turned out to be insignificant. Furthermore, with only one lagged dependent variable in the equation, the coefficients of per capita real disposable income growth and the presidential approval rating were not significantly different from zero. Thus, he concluded that, while the political inertia (or voting history as he referred to it) exerts a strong effect on midterm election outcomes, the effects of economic performance and presidential coattails do not matter. However, according to Erikson's results, the coefficient of the lagged vote term was not significantly different than unity. Consequently, Jacobson [1990] argued that one can assume it to be one. Then the lagged vote term can be moved to the left-hand side, making the dependent variable the vote swing (change from the previous vote). Regressing this dependent variable on the two variables in question produced statistically significant coefficients.

Alesina et al. …