30 second Rock the Vote GOTV television ads are found to be particularly effective among their targeted age group of 18 - 24.

Green, Donald P. and Lynn Vavreck. 2006. "Assessing the Turnout Effects of Rock the Vote's 2004 Television Commercials: A Randomized Field Experiment." Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago, IL, April 20-23, 2006.

Rock the Vote TV Ads in General Elections

Donald Green and Lynn Vavreck examine the extent to which television ads produced by Rock the Vote are effective in mobilizing voters. Previous research on the effectiveness of television communications as a method of voter mobilization has consisted primarily of laboratory experimentation, survey research or natural experiments. Past research has indicated that presidential ads have little effect on turnout but other types of do have effects. Hypotheses addressing the failure of presidential ads have pointed to the fact that the ads rarely mention turnout, and they fail to mobilize as a result of their characteristic negative tone. Building on previous work (2005) Green and Vavreck test the effectiveness on turnout of 30-second Rock the Vote ads played on the USA network via single zip code area cable television service providers in several states.

Electoral Context: The 2004 presidential election was highly competitive and included Senate races in all but three of the sites targeted for the experiment. Four of the Senate races were competitive, high-profile contests (Colorado, South Carolina, Oklahoma and Kentucky, Most of the other races exhibited wide margins with incumbents re-claiming their seats. The competitiveness of House races varied ranging from Texas, where the GOP picked up 6 seats, to Tennessee, where every incumbent won.

Subject Population: Prior to the 2004 presidential election, Green and Vavreck assembled a list of cable TV systems that cover a single zip code and eliminated any of the systems that were located in states listed as battlegrounds by the Los Angeles Times. They also excluded systems that had no prime time spots available on USA during the week before the election. A total of 85 cable systems were identified for randomization.

Randomization Procedure: Each single zip code cable system was paired with another according to its past turnout rate in presidential elections. The paired list was then randomly sorted, and the first cable system in each pair was assigned to the treatment, the second to the control.

Treatment: 30 second Rock the Vote ads aired in treatment districts. One ad broached the issue of the military draft, and the other ad addressed education opportunities. Ads were shown as many times as budget constraints would allow. During the last eight days of the election the ads aired four times per night on each of the following channels: USA, Lifetime, TNT and TBS. Measuring the reach (percentage of population with televisions who have seen the ad at least once) and impression (the number of times an ad is viewed) Green and Vavreck conclude that one out of every two people over the age of 18 saw the ad and approximately 32% of the population between the ages of 18 to 24 years saw the ads.

Findings: Green and Vavreck contracted a firm to assemble turnout information for each of the zip codes included in the experiment. For the population as a whole, they estimate the treatment effect to be .55 percentage points with a standard error of 1 percentage point. While the p-value of .27 is higher than the standard of .05 for statistical significance, the authors argue that the output is still meaningful. The average intent-to-treat effect of .56 percentage points multiplied by the number of registered voters in the systems (353,405) indicates that the ads produced an additional 1,979 votes. The cost of the ads was $27,652.20, so the cost per additional vote was $14, which is cost-effective in comparison with other large-scale GOTV methods.

The average intent to treat effect for voters between the ages of 18-24, for which the ads were designed, was 2 percentage points with a standard error of 1.37.