1998 study: United States-CT; 2002 replication study: United States-MI, IA
Unit of observation:
1998 and 2002 studies: Individuals, within households
1998 study: 29,380; 2002 replication study: 1,905,320 (Individuals, within households)
1998 study: In September 1998 we obtained a complete list of all registered voters, from which we created a dataset of all households with one or two registered voters. To eliminate students from the sample, all names with post office box addresses were excluded, as was one voting ward that encompasses a university and student housing. We were left with 29,380 individuals (22,077 households) whose participation in the 1998 could be determined from public records. 2002 replication study: At the time of this writing, voter turnout results for the November 2002 elections remain unavailable in two small Michigan counties. Eliminating those observations reduces the treatment group to 59,972 and the control group to 1,845,348.
1998 study: Random assignment was done at the household level; the sample was divided into control and experimental groups. 2002 replication study: The congressional districts of each state were divided into “competitive” and “uncompetitive” strata. Within each stratum, households containing one or two registered voters were randomly assigned to treatment and control groups. Only one type of treatment was used: a get-out-the-vote phone call. Just one representative from each household was assigned to treatment or control; the other voter was ignored for purposes of calling and statistical analysis.
1998 study: Personal canvassing, phone call, and direct mail get-out-the-vote messages. 2002 replication study: a get-out-the-vote phone call.
1998 study: Personal canvassing, phone call, and direct mail. 2002 replication study: phone.
1998 and 2002 studies: Voter turnout