Introduction
In the lead up to the Virginia elections, the State Navigate team noticed that voter registrations on college campuses this year were skyrocketing, beating voter registration additions during the 2017 gubernatorial cycle. On Election Day, for example, there were more student voters in Blacksburg at Virginia Tech than there were in 2024. In anticipating this surge of a growing demographic that swung hard to the right in 2024, the State Navigate polling team made it a priority to make sure their raw sample of their statewide polls was close to their weighted targets based on previous Virginia gubernatorial election analysis from Catalist.
The 18-29 age demographic is one of the hardest demographics to poll, as their response rates fall well below a poll’s overall response rate. We knew that in order to create a sample that would be reflective of the upcoming electorate, we would not only need to oversample these voters in our raw sample, but also determine what could be an effective message to send them to pique their interest in filling out our survey.
Executive Director Chaz Nuttycombe, age 26, came up with the idea of sending them customized text messages that suited their day-to-day language. In other words, we ourselves had to “text like a Zoomer.” Two customized text messages with matter-of-fact, familiar language seemed to be an effective way of reaching this demographic. Nuttycombe and Elections Coordinator Michael Foley crafted the language, and we found that this tactic was extremely effective and a key part of State Navigate’s success compared to virtually every other pollster in Virginia.
Experiment and Results
State Navigate’s text vendor, Alliance Forge, sent about 72,000 texts to potential poll respondents for our first statewide poll, and just under 69,700 for the second. Texts for all of our statewide polls were sent in the early afternoon over a 3-day period.
Below is a chart showing when the texts for the first poll were sent, the message used, the number of completed responses received as a result of each text, and the overall response rate:
When conducting the first poll, we noticed that we needed younger respondents in particular in order to make the sample more representative. On the second day of the poll, we targeted 20,000 voters who were from underrepresented populations in our sample— generally younger voters, minority voters and women. On the last day of the poll, we sent out an experimental text to 18-35 year old voters using vernacular that is more firm in their zeitgeist. This greatly increased the 18-35 response rate from 0.26% for the regular message to 0.43%— despite it being a recontact message sent to people who had already received a text poll invite.
As the chart for the second statewide poll below shows, recontact messages usually get lower response rates than initial messages, which is why we understand that some pollsters will just contact an entirely new sample the next day instead of trying to recontact harder-to-reach voters. We think that approach is often incorrect because attempting to recontact voters theoretically removes some nonresponse bias.
Learning from this experiment, we sent a slightly different experimental message to a new sample of 18-29 year old voters on day 2 of the second poll. This message was lowercase (except for the potential respondent’s first name) and included an emoticon. Sending this message increased 18-29 response rates to 0.63%, under a tenth of a point less than that for the generic message sent to the non-18-29 sample that day.
Conclusion
What we learned from this experiment is simple: Targeting hard-to-reach younger voters in their language works— and makes the sample more representative. Cultural competence goes a long way.
