#iVoted Festival Increases Voter Turnout
Jonathan Gruber, MIT, May 2022
#iVoted Festival provides a novel approach to increasing voter turnout in the U.S., rewarding those who show a selfie from outside their polling place or at home with their blank and unmarked ballot to access in-person or live webcast concerts. The purpose of this project is to assess the impact of the 2020 virtual #iVoted Festival on voter turnout.
To do so, we collected data from the roughly 12,000 attendees of the 2020 #iVoted Festival on their name, age and zip code of residence. We then worked with Catalist to match information on actual voting of these individuals in 2016 and 2020 based on publicly available records. We obtained matches for 7,564 individuals who had non-missing information on zip code and other data.
The data shows very high turnout in 2020 among #iVoted Festival attendees. Overall, 83% of those who participated in #iVoted Festival were actual voters in 2020. That is an exceptionally high turnout rate, particularly given the relatively young age of the participants.
To illustrate this, we can compare to the turnout rates in the same age groups and areas of the country from which we drew #iVoted Festival attendees. We worked with Catalist to provide the number of voters at each age in each county in 2020. We then collected Census data on the population at each age in each county to determine the population turnout rate. For each #iVoted Festival attendee, we match the turnout rate for their county and age, so that we can create a comparable sample for turnout comparison.
We find that turnout is only 67% in this comparison sample. That is, the turnout rate among #iVoted Festival attendees is 16 percentage points, or 24% higher than those of comparable age and location. This is particularly true for younger voters. In our comparison sample, 55% of those who were below age 30 voted – whereas among #iVoted Festival attendees, the rate was 78%.
Of course, one concern with such a comparison is that those who attended #iVoted Festival are not comparable to other potential voters. They may be more inclined to vote for other reasons, which may be how they found #iVoted Festival. As a result, one might be concerned that the higher voting rate among attendees simply reflects a higher propensity to vote, not a causal effect of #iVoted Festival.
To partially address this problem, we can control for overall differences in turnout behavior between #iVoted Festival attendees and the general population using data from the 2016 election. That is, if the #iVoted Festival attendees would have turned out anyway, they should have been more likely to turn out in 2016 as well. So we can use the experience of 2016 as a sort of “control” to capture any underlying differences in voting propensity among our population.
Of course, in carrying out this exercise, we cannot include those who are 18-21 in 2020, since they didn’t have the opportunity to vote in 2016.
For those #iVoted Festival attendees 22 and over, we do indeed find that in 2016 their turnout was higher than the general population (69% vs. 59%). After controlling for this difference, we estimate that #iVoted Festival attendees increase their voter turnout in 2020 by 5 percentage points more than would be expected given past voting. That is, #iVoted Festival attendees raised their turnout by 14.5 percentage points, while the overall comparable population turnout rose by 9.4 percentage points. So this implies that #iVoted Festival attendance raised turnout by 7.2% relative to 2016.
This estimate is not perfect. In particular, this approach assumes that any differences between #iVoted Festival attendees and the overall population is captured by their 2016 voting behavior. But it may be possible that those who choose to participate in #iVoted Festival in 2020 are those who were particularly likely to increase their voter turnout. This would bias our estimate upwards. On the other hand, this approach excludes 18-21 year olds; #iVoted Festival attendees in that age range were 26 percentage points (50%) more likely to vote than the general population. Excluding this group likely biases our estimate downwards.
Absent a randomized evaluation, it would be hard to successfully address these issues. The 7.2% estimate is our best available given the data and lack of randomization.