Partisan mass GOTV email messages sent by the Democratic National Committee are not found to be effective in mobilizing turnout among registered Democrats.

Stollwerk, Alissa F. 2006. "Does E-mail Affect Voter Turnout? An Experimental Study of the New York City 2005 Election." Unpublished Manuscript. Institution for Social and Policy Studies, Yale University.

Alissa Stollwerk reports the findings of a randomized field experiment examining the effect of partisan get-out-the-vote, or GOTV, e-mails on voter turnout. The experiment was conducted in New York City in conjunction with the Democratic National Committee during the 2005 municipal elections. Stollwerk finds that partisan e-mail targeted at party activists does not have a positive effect on voter turnout; on the contrary, there is a slight negative, but statistically insignificant, finding that is consistent with previous experiments testing e-mail and other impersonal forms of voter turnout communication.

Electoral Context: In November 2005, New York City held municipal elections for mayor, public advocate, and city councilmen, among other races. While most of these races were contested, very few of them were actually competitive. In the race for mayor, the highest office up for election, Republican incumbent Michael Bloomberg was expected to defeat easily Democratic challenger and former Bronx Borough President Fernando Ferrer. Without a more competitive race at the top of the ticket, the New York City races were considered rather low salience and turnout reached a low 34%.

Subject Population: The Democratic National Committee targeted 52,413 e-mail addresses for which there was a corresponding zip code in its database and that matched NYC voter files.

Randomization Procedure: Approximately 20 percent of the e-mail addresses were randomly assigned to a control group that would receive no e-mails. The remaining 80 percent were assigned to a single treatment group that would receive three e-mails encouraging them to vote. The sample was divided into treatment and control groups through a stratified randomization process that used zip code information as a proxy to control for a number of demographic factors associated with home location, including race and economic status. Stollwerk first assigned each subject a random number; sorted the data file of subjects first by zip code, and then by random number within each zip code and assigned every fifth address to the control group. The remaining 79.9% were assigned to the treatment group.

Treatment: In the thirty-six hours leading up to the election, three e-mails were sent to the treatment group encouraging them to vote. The control group received no e-mails during this period from the DNC. The e-mails were sent from the DNC server in the standard design DNC e-mail and signed by Tom McMahon, the Executive Director of the DNC, who often sends e-mails to the membership list. The e-mails were structured around a partisan appeal that urged Democrats to vote for Democrats on Election Day. The messages repeatedly asked the subjects in the treatment group to vote and provided a link for them to look up their polling place location in the second and third GOTV e-mails.

Findings: Stollwerk's analysis compares the extent to which partisan GOTV e-mails increased turnout among Democratic partisans by comparing the percentage of people in the control group who voted to the percentage of people in the treatment group who voted. A basic cross-tabulation finds that the e-mails did not have a positive effect on turnout. The treatment group voted at a rate one percentage point lower than the rate of the control group. However, the difference is not statistically significant. OLS regression including controls for demographic information, Democratic Party registration and past turnout estimates a 0.4 percentage point decrease in the likelihood of voting although the results are not statistically significant. Two stage least squares regression including the same covariates as the OLS regression, incorporating the contact rate measured by the open rate of HTML formatted e-mails, estimates a treatment-on-treated treatment effect is -3.4 percentage points with a standard error of 3.5 percentage points.