Addonizio, Green, Glaser and Ryan 2006 - Election Day Festivals in NH and CT
Holding community parties at polling sites is found to be effective in mobilizing voter turnout.
Addonizio, Elizabeth, Donald Green, James M. Glaser, Timothy J. Ryan. 2006. "Putting the Party Back into Politics: Results of an Experiment Designed to Increase Voter Turnout through Music, Food, and Entertainment." Paper presented to the Challenges of Participatory Democracy Workshop, University of Southern California, January 24, 2006.
In the 19th Century, Election Day was a festive occasion and polling places were celebratory atmospheres where people often stayed for hours after casting their votes. The turnout rate often exceeded 90% of eligible voters. Elizabeth Addonizio, Donald Green, James Glaser and Timothy Ryan present the results of two pilot studies testing the effect of making Election Day and polling places more festive, community-centered events. The Election Day festivals were held in two distinct settings: a suburban neighborhood in New Hampshire and inner city New Haven, Connecticut. The authors find that the results suggest that Election Day festivals are a promising method for mobilizing turnout in elections.
EXPERIMENT 1: HOOKSETT, NH
Electoral Context: The experiment was conducted during the Spring 2005 local elections. The ballot included low salience issues such as town council members, cemetery and budget committee members and other such offices. The Hanover election was similarly nondescript.
Subject Population: Hooksett, NH was chosen because people are allowed to register at the polls on Election Day. Furthermore, there is only one polling place in town, so every adult citizen who attended the festival would be eligible to vote.
Randomization Procedure: Randomization was carried out at the town level. A coin toss determined that Hooksett would be the treatment group and Hanover would be the control group.
Treatment: The authors employed an experimental design in which the intervention was embedded. Hooksett was chosen as a festival site on the basis of experimental procedure. Prior to selecting Hooksett, the authors compiled voter turnout records for towns throughout New Hampshire and stratified towns by population and past turnout rates. The authors identified Hooksett and Hanover, two similar towns in terms of population and turnout, both which hold elections in single polling places. A coin toss determined that Hooksett would be the treatment group and Hanover would be the control group.
The experimental intervention consisted of a publicity campaign and a festival at the polling site. One week before the election, the authors met with town administrators, election officials and community leaders enlisting their help announcing the festival and handing out flyers. The authors displayed posters throughout the town and a flyer was included in the regional newspaper the Saturday before Election Day. The local Hooksett Neighborhood News also advertised the event and a story about the festival appeared in the Union Leader the Sunday before Election Day. Three dozen lawn signs and 3,000 robo-calls, one that was made on the Saturday before Election Day and another on Election Day further advertised the event. On Election Day, the festival took place immediately outside the polling place. The festival offered food, drinks, raffles, cotton candy, and a DJ. Many in attendance had received information through the various publicity efforts.
The cost effectiveness of the Hooksett festival is determined by dividing the cost of the festival ($3,503) by the approximate number of votes it generated (1,065), which implies a cost of $3.25 per additional vote.
Findings: The authors estimate the effect of their experimental intervention using regression defining the dependent variable in two ways: 1) turnout percentage defined as the ratio of voters to the population of adults, and 2) log-odds of the turnout percentage. Their specifications include various controls for the town and 2005 election (since they follow elections in three towns over six local elections). Their results suggest that the intervention increased turnout by 10.4 percentage points with as standard error of 6.5 percentage points. The estimate borders statistical significance (p=.06).
EXPERIMENT 2: NEW HAVEN, CT
Electoral Context: The experiment was conducted during the fall 2005 local elections. The ballot included no controversial measures, and the contests for alderman, ward mayor and town clerk were not competitive.
Subject Population: New Haven was selected because its demographic composition was markedly different from that of Hooksett in terms of race and affluence. New Haven holds single member legislative elections within districts called wards. The wards studied in this experiment were in a low-income part of the city in which the constituencies are 85% African American.
Randomization Procedure: Randomization was carried out at the ward level. A coin toss determined that Ward 21 would be the treatment group and Ward 20 would be the control group.
Treatment: Using data from local officials, the authors stratified thirty New Haven wards by population, ethnicity, voter turnout history, and degrees of competitiveness in the current and municipal elections. Wards 20 and 21 were identified to be similar in terms of population, turnout and demographics. A coin flip assigned Ward 21 to the treatment group and Ward 20 to the control group. The other New Haven wards served together as a parallel case for statistical purpose.
As in New Hampshire, the intervention consisted of a publicity campaign and an Election Day festival. Local officials were asked to hand out flyers and publicize the event in meetings one week before the election. Three pre-recorded phone calls were directed to Ward 21 registered voters' households on the Friday before Election Day, the Sunday before Election Day, and on Election Day. The calls offered an invitation and gave the times and location of the festival. The festival took place on Election Day immediately outside the polling place. It involved a large tent, food, drinks, cotton candy, and entertainment by a local DJ who doubled as a clown.
Findings: The authors estimate the effect of the festival using regression analysis with a sample consisting of 34 observations (the two most recent elections for wards 20 and 21 as well as 15 similar wards in terms of competitiveness). The dependent variable was the change in turnout from 2003 to 2005. The specification includes controls for wards and elections. The results suggest that the intervention increased turnout by 1.3 percentage points with a large standard error of 6 percentage points. Other measures indicate that the festival did have an effect on turnout. Vote totals taken by party officials indicate that in the control Ward 20, only 8.9% of voters casting ballots that day did so between the hours of 3:00 PM and 6:00 PM (the hours of the festival). By comparison, 35.1% of the votes cast in the treatment Ward 21 were submitted within the time of the festival.
The cost effectiveness of the New Haven festival is determined by dividing the cost of the festival ($4,064) by the approximate number of votes it generated (1,942 registered voters) multiplied by the treatment effect of 1.3 percentage points, which implies a cost of $163 per additional vote. The cost of the New Haven festival was higher than Hooksett as a result of security required by the school which served as the site and an extra round of phone calls to registered voters.