The experimental population universe consists of registered voters within South Los Angeles. This population was chosen simply because it is the population from which SCOPE draws its volunteers — additionally we knew, from contacts with other grass-
roots organizations and campaigns, that there were no other campaign efforts at- tempting to contact this voter population. Individuals were aggregated to the house- hold level, and, by precinct, each household was randomly assigned to treatment or control. For those households in the treatment group, a canvasser was assigned to go to that address and speak for a few minutes with an individual in that household about the upcoming June 2006 general election. Canvassers began to contact house- holds four weekends prior to the election and continued to do so up until election day.2 Canvassers varied in type — some were volunteers, some were paid employees, and some were residents of the precinct within which they were speaking to the registered voters. Note here that while the contact was randomly assigned, the type of contact is not — and consequently all comparisons about the effectiveness of type of contact will be observational results and not the result of random assignment. Additionally, a canvasser could be both, for example, paid and walk in their home precinct.
This chapter has several significant differences with the previous chapter. In par- ticular, the experimental population consists of all registered voters within twenty-six precincts in South Los Angeles, as opposed to the November population of occasional voters within a larger set of precincts. Additionally, the information about canvassers is at the precinct level. Thus we have recorded, for example, whether or not a precinct was contacted by a resident or by a paid canvasser, but no measurement of the inten- sity or frequency of that contact nor do we know the particular households contacted by the different types of canvassers. As a consequence our analyses of home-turf can- vassing is less precise – here we observe similar results (home-turf canvassing appears to increase turnout) but with larger standard errors.
Table 5.1 describes the breakdown by treatment and control assignments for all 7212 individuals in the experiment. These individuals are randomly assigned at the household level, by precinct, from 26 precincts — a total of 1782 individuals were assigned to the control group while 5430 individuals were assigned to the treatment group.
2During the final weekend prior to the election, some individuals were contacted a second time by the campaign. Unfortunately, no data was recorded during this weekend to verify re-contacts.
Table 5.1 Goes Here
Ideally each individual in the treatment group would be contacted — however, this is simply not possible. The canvassers often make only a single attempt to con- tact a household, so that, for example, it would not be possible to contact someone in the treatment group if they were running an errand when the attempt was made.
Canvassers report all types of problems with making contact with individuals in the treatment group — some individuals had frightening dogs in their yards, other indi- viduals refused to open the door, and most often the canvassers simply did not have time to attempt contact with all of the individuals in the treatment group. Table 5.2 describes the number of individuals who were assigned to control and treatment and then tabulates which individuals were successfully contacted. Approximately 35% of the individuals assigned to the treatment group were contacted. Additionally, ap- proximately 1.5% of the individuals assigned to the control group were contacted accidently.
Table 5.2 also describes the number of individuals who live in a precinct that was canvassed by a paid canvasser as well as the number of individuals who who live in a precinct that was canvassed by someone who lives in that precinct. Almost all of the precincts had some paid canvassing, whereas only six precincts were canvassed by a resident. Finally, we tabulate the number of individuals by control and treatment assignment group who voted in the June 2006 election. SCOPE recorded the indi- viduals who were contacted and the classification of canvasser by precinct. The Los Angeles Registrar-Recorder’s office provided data on which individuals participated in the June 2006 election.
Table 5.2 Goes Here
Prior to analysis, we first determine to what extent the data is balanced across the observable covariates. One of the reasons that it is necessary to conduct a randomized experiment is that there are remarkably few observable characteristics for each indi- vidual voter. This prevents us from using more traditional methods of analysis, such
as regression techniques, and requires the use of randomized experimentation. We check to ensure, however, that for the variables which we can observe that there is not a statistically significant difference between treatment and control groups. Table 5.3 describes the differences. No variable has a statistically significant difference across treatment and control. Note that there is some missing data for the age variable — not every individual in the data-set provided by the Los Angeles County Registrar Recorder had a birth-date, and as such, 462 individuals have missing age variables.
We also present an identical table for the differences in means, by covariate, de- scribing the differences between the individuals who were contacted against those who were not. Table 5.4 is clear evidence of the need to incorporate randomization experimentation in analyzing mobilization efforts. Here we observe statistically sig- nificant differences in the vote history, age, and partisan registration (Democratic and Decline-to-State) between individuals who were successfully contacted vs not.
Table 5.3 and Table 5.4 Go Here