Green, Donald P., Lynn Vavreck (2008). Replication Materials for ‘Analysis of Cluster-Randomized Experiments: A Comparison of Alternative Estimation Approaches.’ http://hdl.handle.net/10079/92deee27-37dc-467b-a7f6-0e96d76370f0. ISPS Data Archive.
Data source information:
November 1, 2004
Unit of observation:
23,869 voters, 18 years of age, first federal election; 85 cable systems (in 12 states)
We removed all cable systems in 16 states that the Los Angeles Times classified as presidential battlegrounds (closely contested states). We then excluded any systems that had no time available in prime time during the week before the election or that cost more than $15 per 30-second advertisement on the USA television network. We excluded all systems in Mississippi because its voter file is very difficult to obtain. This left 85 cable systems for randomization.
Each cable system was matched with one or two other systems in the same state according to its past turnout rate in presidential elections. This procedure resulted in 40 strata containing the 85 cable systems. After sorting the list of 85 cable systems by strata and then by a random number, the first cable system in each stratum was assigned to the treatment condition, the others to control.
People living within the treatment systems saw two different 30-second advertisements produced by Rock the Vote. Both advertisements used the same format. The first dealt with the draft and the second, with education. In our experiment, both advertisements were shown as many times as budget constraints would allow.
Turnout rates in 2004 general elections
Donald P. Green, Lynn Vavreck
Area of study: