Gerber, Grebner, Green and Larimer 2006 - Effect of Publicizing Individual Turnout on Voter Mobilization

Pre-election and post-election mailings that publicize neighbors' turnout records in an election are found to be effective in mobilizing voters. Social pressure to adhere to civic norms may resolve "paradox of participation."

Gerber, Alan S,, Mark Grebner, Donald P. Green and Christopher Larimer. 2006. "Does Voter Turnout Increase When Neighbors' Voter Turnout Records are publicized?" Paper prepared for presentation at the Annual Meeting of the Midwest Political Science Association, Chicago, IL, April 20-23, Palmer House Hotel.

Alan Gerber, Mark Grebner, Donald Green and Christopher Larimer present experimental evidence that social pressure to adhere to the civic norm of participation may help resolve the "paradox of participation." Based upon past psychological research which demonstrates that people are more likely to follow social norms when their behavior is observed by others, the authors conduct two experiments to test whether publication of voting behavior compels people to vote. Prior to a 2004 primary and a 2005 municipal election, treatment group members received mailings indicating that after the election in question, all registered voters in the neighborhood would receive a flyer indicating the turnout of the residents in the neighborhood including the subject. The treatment areas exhibited a two percentage point increase in voter turnout suggesting that heightened awareness that others may observe one's voting behavior has an effect on whether a citizen chooses to vote. The experiment is the first empirical test of whether publicity increases the pressure to conform to civic norms.

Experiment 1: August 2004 Michigan Primary

Electoral Context: The first of two experiments was conducted prior to the August 2004 primary election. The primary was a statewide election for which the overall turnout was 20%. The ballot included primaries for Michigan's 15 members of the U.S. House of Representatives as well as county and state level offices (including the Michigan legislature) and several judgeships.

Subject Population: The sample for the experiment consisted of 38,066 registered voters in 188 precincts in Michigan selected from the 4000 precincts statewide. The precincts were chosen to be broadly representative of Michigan. Some precincts were eliminated prior to selection because 1) data were unavailable, 2) the precincts had low rates of home ownership or 3) the precincts had low population density.

Randomization Procedure: Precincts were divided into address blocks; then two address blocks were randomly chosen from each precinct. For each precinct one of the chosen blocks was randomly assigned to the treatment group and the other to control.

Treatment: Subjects in the treatment group were sent one mailing 10 days prior to the primary election. Approximately 2% of the mailings were returned by USPS, The mailings were distributed according to blocks. Each mailing contained the names of all registered voters on the block with a note describing the purpose of the mailings. The first mailing included the names with blanks beside them for the August primary and November general elections. The first mailing served the purpose of informing subjects who was a registered voter on their block and to let them know that a second mailing would be sent showing who actually voted in the primary. The post-primary mailing was sent as promised, displaying the word, "Voted" beside names of registered voters from the block who voted with blanks remaining for people who did not vote.

Findings: Using a simple OLS specification, the authors estimate a 1.8 percentage point increase in turnout among the treatment group (standard error: 1.38 percentage points). When covariates are included in the estimate to control for past turnout, race, gender, and home ownership, the treatment effect is estimated to be 2.4 percentage points (standard error: 1.04 percentage points).

Experiment 2: November 2005 Detroit Mayoral Election

Electoral Context: The second experiment was carried out during the November 2005 Detroit Mayoral election.

Subject Population: The sample consisted of voters who voted in the November 2001 mayoral election or the November 2004 presidential election, but voted neither in the August 2005 mayoral primary nor voted more than once by absentee ballot. The criteria were selected to avoid targeting particularly high or low probability voters.

Randomization Procedure: For this experiment, blocks with fewer than 15 voters or more than 25 voters as well as blocks with apartment buildings were excluded prior to assignment to treatment and control groups. The treatment group assignment was performed at the individual level. Within each block, samples consisted of voters who voted in the November 2001 mayoral election or the November 2004 presidential election, but voted neither in the August 2005 mayoral primary nor voted more than once by absentee ballot. The criteria were selected to avoid targeting particularly high or low probability voters. Voters meeting the above criteria were ordered by precinct and address and every fifth voter was assigned to the treatment group. Because the number of voters in each precinct varies, the selection method approximates random assignment. 9,328 individuals were assigned to the control group and 2,332 were assigned to the treatment group.

Treatment: Voters in the treatment group were assigned to receive either one or two mailings. Subjects assigned to receive one mailing received it 19 days prior to the general election. Subjects assigned to receive two mailings received one 19 days prior to the general election and another mailing nine days prior to the general election. 1,533 members of the treatment group received one mailing and 799 members received two mailings. Each mailing displayed a list of names of registered voters on the block who voted in the November 2004 as well as the individuals who voted in the August 20058 primary. The mailing emphasized the importance of the November election and asked, "Even if you didn't vote in August, can we count on you in November?

Findings: The results indicate a 2 percentage point increase in the turnout rate of voters in the treatment group (standard error: 1.03 percentage points) but no significant difference in the effect for one versus two mailings. Including covariates controlling for past turnout, gender, race and home ownership, the authors estimate a 2.16 percentage point increase in turnout among the treatment group (standard error: 1.01 percentage points).

Tiffany Davenport – Fri, 2006 – 09 – 22 11:19