Grantee: Vote.org. The final study report submitted to Arnold Ventures is posted here.
Description of the Intervention: BallotReady, a non-partisan social enterprise that runs a widely-used website providing information about every candidate in upcoming elections, partnered with the non-profit Vote.org to create an online tool called Make a Plan to Vote (MAPTV). The MAPTV tool has three components: (i) a series of specific questions to encourage website visitors to make a plan to vote, (ii) prompts for users to sign up for text or email reminders about when and where to vote, and (iii) a platform where users can easily share their intent to vote with their friends on social network websites.
Study Design: Prior to the 2018 U.S. midterm elections, the study randomly assigned 119,211 users of the BallotReady website in all 50 states to the MAPTV version of the site (treatment group) or the information-only version of the site (control group). Multiple users located at the same street address were randomly assigned to the same group. Roughly 70 – 75% of treatment group members used at least some features of the MAPTV tool prior to the election (e.g., choosing a polling time and location). The study’s pre-specified primary outcome was voter turnout in the 2018 mid-term elections, as measured using publicly available voting records.
Impact on the Primary Outcome: The study found no statistically significant effect on voter turnout in the 2018 midterm elections. The turnout rate was 74% in both the treatment and control group.
Study Quality: Based on careful review, we believe this was a well-conducted RCT that produced valid findings.1
For example, the study had successful random assignment (as evidenced by highly-similar treatment and control groups) and valid analyses that were publicly pre-registered and, among other things, accounted for the fact that users at the same address were assigned to the same study group (treatment or control). One modest study limitation is that voting outcomes could only be obtained (via match to voter records) for 56% of the sample. However, the match rates were virtually identical for the treatment and control groups, as were the baseline characteristics of those treatment and control group members who were matched, making it unlikely that imperfect matching undermined the randomization and led to inaccurate results.↩︎