How peer-to-peer texting drove voters to the polls in the midst of a global pandemic
This project was made possible by some of the members of our amazing 16,000-person skilled volunteer community. Thank you to:
Anna Schneider, Cameron Bartok, Connor Kelley, Jocelyn Blumenrose, Lauren Fealey, Leah Nicolich-Henkin, Peter Spiro, Priya Gupta, Stella Mach, and Will Seaton.
We generated findings by combining the results of peer-to-peer texting campaigns with the TargetSmart national voter file. We first explored voting trends through differences-in-differences analyses, in which we analyzed the list of registered voters we texted, using their TargetSmart Turnout Score as a pre-treatment estimate of their turnout and their actual turnout as post-treatment. We could then compare their average Turnout Score with the actual turnout rate, and compare this difference across demographics. This enabled us to calculate the marginal vote increase for populations that received text messages versus those that did not when controlling for their TargetSmart Partisanship Score and other factors.
When we discovered any directional learnings, we then conducted regression analyses to try to uncover statistically suggestive and meaningful findings. With the list of potential voters we texted, we were able to control for various TargetSmart demographic attributes; such as age, gender, ethnicity, urban/suburban/rural environment, and more; and run a regression analysis to identify which behaviors were correlated with higher turnout rates.
Impact of texting on overall turnout -
Controlling for the turnout likeliness of each group, the group of voters TFC texted had a higher voting rate by 0.7 percentage points compared to uncontacted voters in the same districts (Figure 1). Voters with the lowest projected turnout rates (expected to vote about 0-20% of the time) experienced the highest increase in net turnout rates when sent a text. These ‘least likely’ voters voted at a statistically significant higher rate compared to uncontacted voters in the same group (Figure 2). Conversely, texting voters with higher Turnout Scores had a marginal impact on their relative turnout rates.
Impact of message content on turnout by partisanship -
In our regression analysis, we found statistically suggestive evidence that swing voters that received voting information, rather than issue-based messaging, were more likely to vote when controlling for gender, age, and ethnicity. Swing voters were identified as those voters with a TargetSmart Partisanship Score between 30 and 70. This finding occurred for swing voters that had high Turnout Scores (Turnout Score > 70) and medium Turnout Scores (Turnout Score between 30 and 70). We excluded all data after October 22 to limit the impact that a “go vote” text message approaching Election Day would have on our findings. These findings further supported the directional learnings from our difference-in-difference analyses in Figure 3 above.
Impact of texting on turnout by age -
To determine the impact of texting on voting rates in different age groups, we reviewed estimated and actual turnout rates for different populations when controlling for gender, ethnicity, and other demographics. While registered voters ages 18-29 that we texted had a smaller net turnout increase (+6.3 p.p.) compared to their untexted peers (+7.3 p.p.), the underlying regression analyses we conducted suggest that texting these younger voters still positively influences turnout rates. Additionally, our regression analyses suggested other age ranges were significantly impacted by peer-to-peer texting when controlling for multiple variables.
When running our regression analyses, we found that the turnout rates for voters ages 18-64 were positively impacted by receiving a text, especially when focusing on the ‘least likely’ voters within specific age brackets. Across all age groups (18-29, 30-44, 45-64), there was a statistically significant difference in turnout rates between those who were texted and those who were not when focusing on voters in the lowest turnout quintile (defined as voters with a TargetSmart Turnout Score between 0 and 20). Voters ages 30-44 that were sent a text were most correlated with increased voting rates. We even saw evidence that texting positively influenced turnout rates for voters ages 30-44 with a Turnout Score from 20-40 as well as 0-20 (the second-lowest and lowest turnout quintiles). As highlighted above, it appears that texting voters ages 18-29 is valuable regardless of their Turnout Score. We saw no evidence of a change in turnout when texting voters 65 years and older.
Impact of response, response type on turnout -
In our regression analysis, we found statistically suggestive evidence that voters that responded to a text message were more likely to vote when controlling for estimated turnout (TargetSmart Turnout Score), estimated partisanship (TargetSmart Partisanship Score), the interaction between partisanship and response, gender, age, ethnicity, and urbanicity. While a positive response was indicative of higher turnout rates, a negative response was correlated with higher turnout rates for only one demographic group: likely Republicans.
These findings further supported the directional learnings from our diff-in-diff analyses in Figure 4 above.