close× Call Us
Title Intensive Latinx Youth GOTV: Voter Mobilization prior to the 2018 Midterm Elections
Post date 11/08/2018
C1 Background and Explanation of Rationale see pre-analysis plan
C2 What are the hypotheses to be tested?

The principal hypothesis is that those randomly targeted for intensive mobilization will vote at higher rates than their control group counterparts.

We also will follow-up on the groups that were randomized in previous elections to see if there are enduring effects.

In order to assess spillover effects, we will assess the voter turnout of others living at targeted addresses relative to the control group.

C3 How will these hypotheses be tested? *


Experiments are conducted in the following locations.
• NYC (Washington Heights, Harlem, other neighborhoods in northern Manhattan and South Bronx)
• Redwood City, CA
• Gainesville, FL
• Dallas, TX
• Houston, TX (maybe)
In each location, the coordinator of that site obtained a voter file, identified likely Latinx voters using a model based on last names and Census information, and then restricted the sample to those 18-23 years of age. Treatment groups of 10 to 20 people (depending on the site coordinator’s capacity for outreach) were randomly selected as the treatment group.


The outreach effort varied somewhat by site but basically consists of the following elements: (1) an initial handwritten letter/postcard that introduced the canvasser (and gave contact information) and emphasized the importance of the election, (2) a visit to the target voter’s address, culminating in either a conversation about the importance of the election, communication with friends or relatives, or a note saying that the canvasser had visited, and (3) follow-up text messages or calls.

Data and Outcome Measures

Voter turnout will be assessed by obtaining updated voter files and calculating turnout rates for the randomly assigned treatment and control groups.

Method for Estimating Average Treatment Effects

Since the experiment is blocked by location with somewhat different probabilities of assignment, our regression models will include indicator variables for each block in the hopes of reducing disturbance variability.

We will report 95% confidence intervals for the intent-to-treat effect, using a margin of error equal to the estimated standard error multiplied by the appropriate critical value from the t-distribution. This estimate will be generated by regressing turnout on treatment assignment, using the covariates listed below to generate more precise estimates. We will also estimate the CACE using instrumental variables regression of turnout on (any form of) contact, with random assignment to treatment as an instrument.

To assess robustness, we will also report a simple regression result with only block and treatment indicator, omitting prior turnout statistics. We expect these results to be similar but less precisely estimated due to the exclusion of prognostic covariates. When interpreting the results, we will rely primarily on the covariate-adjusted estimates.

C4 Country United States
C5 Scale (# of Units) approximately 500 per site
C6 Was a power analysis conducted prior to data collection? Yes
C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval? Yes
C8 IRB Number IRB-AAAS1321 at Columbia University
C9 Date of IRB Approval October 18, 2018
C10 Will the intervention be implemented by the researcher or a third party? Researchers
C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research? No
C12 If relevant, is there an advance agreement with the implementation group that all results can be published? not provided by authors
C13 JEL Classification(s) not provided by authors