Economics Thesis : The Effect of Early Voting on Turnout Alex Mourant Economics Department April 6 2011
Mourant 1 Introduction Early voting is an increasingly common option for voters in the United States With early voting voters can cast a vote in person days before the actual election. Since potential voters are given a number of days to vote rather than solely the first Tuesday in November the eas e of voting should be improved with the option to vote early Election re formers have addressed early voting as a simple way to increase participation because the available window for voting is much larger. Early voting is getting increasingly popular in the United States. By the 2008 election t hirty two states plus the Distric t of Columbia ha d adopted early voting to increase the ease of participating in elections. While one would reasonably expect turnout to increase with the implementation of an early vote system, scholarly research shows early voting to have a minimal effec t on turnout in elections. Research on the topic suggests that early voting serves to make voting easier for groups who would otherwise vote on Election Day and does not address the underlying reasons why many citizens abstain from elections in the United States However, the research is not particularly broad and was largely conducted in incipient early voting programs which were not well established. Past research does not cover general elections in the United States, comparing the turnout effects between states. I decided to perform a multivariate regression to explore if early voting has a statistically significant positive relationship with turnout, as proponents suggest and scholars deny. Since early v oting is still a relatively new creation across the nation, few general elections have been cond ucted where it has been a factor in a wide range of states. This study will use data from 2004 and 2008 general elections to analyze the relationship between early voting and turnout.
Mourant 2 Literature Review The Unit ed States started implementing early voting widely in the 1990s. Thus, scholarly work on the subject does not have a substantial history and most of it is focused on the intra state level. There appears to be a substantial amount of room for nationwide re search on the topic. Much of the research on the topic indicates no statistically significant effect of early voting on turnout rates. Neeley and Richardson conducted an individual level survey of voters in a Tennessee county to determine the impact of ear ly voting on turnout and found no support for an increased turnout among group s that were less likely to vote (Neeley and Richardson) More general research has similar findings. Gronke, Galanes Rosenbaum, and Miller ran a multivariate regression of data f rom a broader group of states which had adopted early voting before it became common throughout the country. Their study indicated no relationship existed between early voting and increased turnout (Gronke, Galanes Rosenbaum, and Miller). While broader tha n the other research on the topic, this study was still limited by the small number of states that were the first to adopt early voting dominating the data set. Stein and Garcia Monet researched the effects of early voting in Texas counties during the 1992 presidential election and found a modest effect on turnout. This is the only study incorporating data analysis where early voting was found to have a statistically significant positive impact on turnout. The authors acknowledge that the increase in turnou t as a result of early voting may also be related to partisan mobilization efforts meant to increase participation of the base voter groups of the parties. A
Mourant 3 relationship between turnout and early voting in general election s may be p artly a result of the p artisan mobilization efforts. Oliver analyzed removing barriers for absentee voting and found that those reforms which reduced an individual's "costs" of voting had a positive effect on turnout. Oliver's thesis would therefore support the argument in favo r of early voting, because early voting makes voting easier. Oliver's work does not provide a compelling case for the supposed turnout boost provided by early voting since it is focused on different barriers to v oting, but the analysis may be valuable to c onsider for the case of early voting (Oliver) Similar to Oliver's work, Kousser and Mullin wrote on voting by mail a s an alternative to voting on Election Day which like early voting reduces the "costs" of voting. With a higher ease of voting, voting by mail would be expected to have a positive effect on turnout. Kousser and Mullin propose that voting by mail may counterintuitively have a negative effect on turnout in general elections (Kousser and Mullin). A possibility exists that early voting h as a neg ative relationship in general elections. While the past scholarly research is valuable, it does not broadly compare states employing early voting. A regression incorporating all of the states will be important to better understand the effects of early vot ing on turnout. This information can help further the understanding of the role of early voting in increasing participation in elections.
Mourant 4 Hypothesis H 1 : States with an option for citizens to vote early will have higher turnout rates for general electio n s in the United States than states without an early voting option The r esearch arguing that early voting does not have a statistically significant impact on turnout is limited and was conducted b efore early voting became an established program in many s tates. Early voting does decrease the opportunity cost of voting and as such one would expect potential voters, acting in their self interest, to be more apt to vote With that theoretical framework, the hypothesis is that states with early voting have hig her turnout rates than states without early voting H 0 : States with an option for citizens to vote early will not ha ve higher turnout rates for general election s in the United States than states without an early voting option. Most scholarly works argue th at early voting does not have an effect on turnout. As the research suggests, early voting may not have had a substantial influence on increasing turnout by lowering barriers to voting. Thus, the null hypothesis for this study is that states with early vot ing will not have a statistically significant turnout rate than states without early voting.
Mourant 5 Data and Method The regression us es a sample of all 50 states in the United States and the District of Columbia for the 2004 and 2008 general elections In a ll, there are 102 observations. The dependent variable is the turnout rate ( VAP Turnout ) Change in the rate of turnout will be important for determining the efficacy of early voting programs VAP turnout is measured as the percentage of the voting age pop ulation that votes in the general election. Data on turnout is available from the Federal Election Commission. The regression incorporate s a number of independent variables. A dummy variable for early voting is important for determining the effect of early voting on turnout. The dummy variable is 1 for states with early voting and 0 for states without early voting ( EarlyVote ) This measure should explain the direction and size of the relationship between early voting and turnout. The relationship is expecte d to be positive as an early voting option increases the ease of voting The United States Election Assistance Commission provides data on which states offer the option of early voting. Another independent variable use d is the closeness of the Presidentia l race in the state ( P residentClose ) PresidentClose is a dummy variable indicating whether the presidential race in that state was close for that election. Any race where the margin between the two major candidates is less than 5% is attributed a 1 and ra ces where the difference between the two major candidates is 5% or greater is given a 0. PresidentClose is expected to be positive as states with very competitive races are expected to have higher turnout rates. The data for this variable come from the Fe deral Election Commission.
Mourant 6 Demographics of the states are likely important determinants of turnout. I incorporat e two demog raphic variables in the regression : percent of population with a bachelor's degree or higher ( Education ) and the percent of a state' s population that is non hispanic white ( %White ) Demographics of each state may play an important role in determining turnout rates. A more educated population can be expected to be more engaged in the political process, so that variable would be expected to be positive. N on H ispanic whites may feel more represented by government and thus have more interest in participating in federal elections. %White is expected to be positive, where states with a population composed of more non Hispanic whites have high er turnout. The demographic data for these variables come from the U.S. Census Bureau Another dummy variable is incorporated in the regression which accounts for turnout effects of Senate races ( Senate Close ) While some Senate races can be expected to ha ve a positive effect on turnout, many Senate races are blowouts, where the favored candidate has little chance of losing. Those races are unlikely to have much of a role in turnout. Thus, this variable only counts close Senate races. A value of 1 is attrib uted to states with a Senate election where no candidate received above 55% of the vote and a value of 0 is attributed to states without a Senate election or with a Senate election where the winner received over 55% of the vote. This variable is predicted to be positively correlated with turnout rates as more voters will participate in a general election with an additional important state wide race. The data come from the Federal Election Commission. The final variable is a dummy variable accounting for th e unique status of the 2008 election ( 2008 ) While not as pronounced as some expected, turnout was higher in
Mourant 7 the 2008 general election than in other recent general elections. Many new voters who had not been involved in past elections participated in 2008 as enthusiasm over the Obama candidacy was higher than for other recent candidates. This variable accounts for the higher turnout in 2008 by assigning a value of 1 to results from 2008 and 0 to results from 2004. This va riable is expected to be positive, a s turnout rates were higher in 2008 than in 2004. Findings One multivariate linear regression w as run, providing interesting results in regards to the hypothesis. The regression incorporates all of the variables explained in the data and methods section. With an adjusted r squared of 571 for the regression, a substantial amount of variability in turnout is explained by the independent variables The variable PresidentClose was positive as predicted. This variable was strongly significant, with a t stati stic of 3.64 and significance at 1%. The states with a close presidential race had higher turnout than states where the presidential race was not competitive. This result correspon ds to intuition, as people are more likely to feel the desire to vote in a r ace that is in question An individual vote is more valuable when it appears that it can influence the outcome of an election. Additionally, more campaign dollars are focused in states with closer races Money spent by campaigns in those states could be us ed in advertising and mobilization efforts which increase turnout. The variable %White was strongly significant, with a t st atistic of 10.18. It was significant at 1%. The coefficient was positive. As predicted, a population that composed of more non Hisp anic white residents should have a higher turnout rate, all else equal,
Mourant 8 according to the model. White voters are often depicted as being more engaged in the electoral process and this result indicates as such. SenateClose as predicted, exhibited a posit ive coefficient that was statistically significant The variable had a t statistic of 2.15 and was significant at 5%. Close Senate races were shown to increase turnout as voter interest and motivation to vote were higher when the Senate race was in questio n. As with close presidential races, the perceived value of voting is higher when an individual vote could potentially affect the outcome of an important election. Campaign dollars will be more focused on close races, so more advertising and turnout effort s will occur in states with close Senate races. The percentage of the population with a bachelor's degree or higher ( Education ) was found to be statistically significant at 1% with a t statistic of 3.99 in the r egressio n The coefficient for Education was positive, as predicted, supporting the as sertion that a more educated population sho uld have a higher turnout rate. States with a more educated population have more voter engagement than states with a less educated population. The variable for the 2008 e lection ( 2008 ) was positive, as expected. The t statistic was 2.35 and it was significant at 5%. States had a higher turnout in the 2008 election than the 2004 election. This variable indicates that the 2008 general election was u niquely high in engagement and turnout. Most importantly for the study, the EarlyVote variable defied expectations. EarlyVote was negative The t statistic was 1.44 and it was not significant at 5%. The null hypothesis could not be rejected, indicating that the data cannot identif y an effect from early voting on voter turnout The inclusion of an early voting option in a state was not
Mourant 9 fo und to increase turnout rates. According to the regression t urnout rates were no higher with early votin g. One potential reason for this finding c ould be that the implementation of an early voting system takes away the importance of Election Day. Rather than likely voters impressing upon their peers to vote on Election Day, with early voting, they are likely to vote another time. The impact of peopl e walking around with "I Voted" stickers and talking about Election Day is reduced with an early voting system. This may conflict with the decrease in the opportunity cost for voters and the increased ease of voting on their own time. Conclusion The regre ssion results support past scholarly research on the topic of early voting and its effect on turnout rates. E arly voting was not found to have a statistically significant effect on turnout in the 2004 and 2008 presidential election s The null hypothesis is not rejected and states with an early voting option showed no statistically significant increases in turnout over states without early voting options. The sample consisted of the United States during the 2004 and 2008 Presidential Election s, which had a w ide variety of states offering votes the option of early voting. Since the regression only uses data from two elections, the findings are limited. Future research should be conducted to analyze the effects of early voting on turnout in subsequent elections As more elections are conducted with an option of early voting, more robust research will be possible. The goal of this paper wa s to better understand the relationship between early voting and turnout in presidential election year s It showed that t he re lationship was not
Mourant 10 statistically significant and this paper provide s another level of support for scholars questioning the efficacy of early voting as a tool for increasing turnout While early voting decreases the opportunity cost of voting, it does not s eem to be effective in increasing turnout. Other tactics should be considered if grea ter participation is the goal.
Mourant 11 Bibliography Federal Election Commission. Federal Elections 2004: Election Results for the U.S. President, the U.S. Senat e, and the U.S. House of Representatives ." Web. . Federal Election Commission. Federal Elections 2008: Election Results for the U.S. President, the U.S. Senate, and the U.S. House of Representa tives ." Web. . Gronke, Paul, Eva Galanes Rosenbaum, Peter A. Miller, and Daniel Toffey. "Convenience Voting." Annual Review of Political Science 11 (2008):437 55 Kousser, Thad, and Megan Mullin. "Does Voting by Mail Increase Participation? Using Matching to Analyze a Natural Experiment." Political Analysis 15 (2007): 428 45. Neeley, Grant W., and Lilliard E. Richardson, Jr. "Who Is Early Voting? An Individual Level Examination." The Social Scien ce Journal 38 (2001): 381 92. Oliver, J. E. "The Effects of Eligibility Restrictions and Party Activity on Absentee Voting and OverallTurnout." American Journal of Political Science, 40.2 (1996): 498 513. Stein, Robert M., and Patricia A. Garcia Monet. Voting Early but Not Often." The Social Science Quarterly 78.3 (1997): 657 71. U.S. Census Bureau. Current Population Survey Data on Educational Attainment ." Web. < http://www.census.gov/hhes/socdemo/education/data/cps/index.html>. U.S. Census Bureau. "S tate Population by Race and Hispanic Origin." Population Estimates Program 2000s Archive Web. . U.S. Election Assistance Commission. Election Administration and Voting Survey ." Web. .
Mourant 12 Descriptive Statistics VAP Turnout PresidentClo s e %White EarlyVote Mean 58.48627 Mean 0.166667 Mean 72.97003 Mean 0.588235 Standard Error 0.667978 Standard Error 0.037083 Standard Error 1.592077 Standard Error 0.048971 Median 58.7 Median 0 Median 77.16521 Median 1 Mode 58.7 Mode 0 Mode #N/A Mode 1 Standard Deviation 6.74625 Standard Deviation 0.374518 Standard Deviation 16.07919 Standard Deviation 0.494583 Sample Variance 45.511 89 Sample Variance 0.140264 Sample Variance 258.5404 Sample Variance 0.244613 Kurtosis 0.4467 Kurtosis 1.322182 Kurtosis 0.795059 Kurtosis 1.9053 Skewness 0.005272 Skewness 1.815665 Skewness 0.95087 Skewness 0.36394 Range 29.8 Range 1 Range 72.78922 Range 1 Minimum 44.1 Minimum 0 Minimum 23.32505 Minimum 0 Maximum 73.9 Maximum 1 Maximum 96.11426 Maximum 1 Sum 5965.6 Sum 17 Sum 7442.943 Sum 60 Count 102 Count 102 Count 102 Count 102 Education SenateClose 2008 Mean 27.16078 Mean 0.22 549 Mean 0.5 Standard Error 0.535832 Standard Error 0.041583 Standard Error 0.049752 Median 25.75 Median 0 Median 0.5 Mode 24.5 Mode 0 Mode 0 Standard Deviation 5.411633 Standard Deviation 0.419968 Standard Deviation 0.502469 Sample Variance 29.28577 Sample Variance 0.176374 Sample Variance 0.252475 Kurtosis 1.981694 Kurtosis 0.22679 Kurtosis 2.0404 Skewness 0.989595 Skewness 1.333433 Skewness 3.59E 17 Range 32.2 Range 1 Range 1 Minimum 15.3 Minimum 0 Minimum 0 Maximum 47.5 Maximum 1 Maximum 1 Sum 2770.4 Sum 23 Sum 51 Count 102 Count 102 Count 102
Mourant 13 Regression SUMMARY OUTPUT Regression Statistics Multiple R 0.772715 R Square 0.597089 Adjusted R Square 0.571642 Standard Error 4.415359 Observations 102 ANOVA df SS MS F Significance F Regression 6 2744.638 457.4397 23.46399 7.78E 17 Residual 95 1852.063 19.49539 Total 101 4596.701 Coefficients Standard Error t Stat P valu e Lower 95% Upper 95% Intercept 26.18489 3.819485 6.855607 7.11E 10 18.60225 33.76752 PresidentClose 4.359327 1.197463 3.64047 0.000443 1.982063 6.736591 %White 0.291618 0.028646 10.18002 6.82E 17 0.234748 0.348487 EarlyVote 1.33665 0.927376 1.44133 0.152781 3.17772 0.504422 Education 0.350385 0.087799 3.990776 0.00013 0.176082 0.524687 SenateClose 2.294278 1.064247 2.155776 0.033627 0.181481 4.407074 2008 2.095404 0.888364 2.358722 0.020386 0.331778 3.85903