Included only cities and towns with populations of over 30,000. In 2005, we excluded cities and towns where cost for radio advertisements exceeded $111 per point in the population. In order to increase the statistical power of our experiment, we sought to create a sample of observations that, within experimental strata, were as homogeneous as possible. We gathered detailed information about the institutional and political characteristics of mayoral elections in each of the 151 cities and matched pairs of municipalities based on voter turnout in the previous mayoral election, incumbent vote share in the previous mayoral election, whether mayoral elections are partisan or nonpartisan, and whether the 2005 mayoral election was contested. All of the cities and towns included in the final sample were municipalities in which the local executive is selected by popular vote (as opposed to appointment by the city or town council). Using the criteria described above, we identified 28 closely matched pairs of cities in an effort to make the treatment group as similar as possible in terms of observable characteristics. Once the matching exercise was completed, we randomly assigned one city in each pair to the treatment group and the other to the control group. We restrict our attention to 33 cities—16 in the treatment group and 17 in the control group—in which an incumbent mayor ran opposed. In November 2006, 105 municipalities held mayoral elections. Using the same four matching criteria as in 2005, we repeated the matching exercise to create 11 pairs, half of which were randomly assigned to the treatment group. Of these, 16 cities—seven in the treatment group and nine in the control group—featured elections in which incumbent mayors ran opposed in 2006. The average cost per point for cities in the 2006 sample was $100, raising the average cost per point in the combined sample of 49 observations to $75.
Matching pairs and random assignment
60-second radio advertisement that presented a non-partisan get-out-the-vote message.
Electoral competetiveness (the difference between the vote percentage won by the incumbent in 2005 or 2006 and his or her vote percentage in the previous election)