|● Project INFLUENCE is a framework for RCT testing interventions to reduce divisive partisanship. |
● First phase is live, testing pro-collaboration messaging on an audience of roughly 400,000 subjects in a defined geo, with a similar number in an untreated control group. After several weeks of treatment, the system will measure any sentiment differences between the groups.
● Behavioral or sentiment impacts can be measured, enabling testing of a diverse array of intervention strategies.
Project INFLUENCE is a system for randomized, controlled testing of the effectiveness of different messages at scalably influencing public attitudes and behaviors toward working with opponents.
Phase 1 is live.
While we believe many politicians foment division, attempting to directly influence them to change behavior is unlikely to succeed because, in the current environment, divisiveness works. “Fear & anger” do get people to the polls. So we believe the only way to change politicians’ behavior is to change the electorate’s support for divisive politicians.
How might we evolve voter sentiment? Project INFLUENCE tests messages, mechanisms for delivering them and tools to measure their effectiveness at influencing citizen sentiment and behavior.
The first phase of INFLUENCE uses social media to target a defined audience with a collection of public service announcements. After a multi-week treatment period, an online survey system measures that group’s sentiment toward collaboration, to assess if it differs from that of an untreated control.
For this initial phase we used 2022 Cook Partisan Voting Index scores to select a geographic region that was meaningfully right-of-center, but not extreme (R+10 to R+19). We will test left-of-center next. We began with pre-polling to confirm an anti-collaboration bias exists in the region. Figure 1 shows one version of that poll. The weighted average response in the selected region was 2.1 out of 5.0, where 1.0 is extremely anti-collaboration. Roughly 70% specified a preference for politicians to be more, or much more principled, versus compromising.
We then structured an advertising campaign targeting this region within Meta’s Facebook platform and are currently delivering “pro-collaboration” messaging to the group via a collection of animated videos.
This first phase of messaging tests variations on the theme, “We ordinary citizens know how to resolve differences with opponents in order to accomplish things. Do our politicians do the same?” Figure 2 shows the first ad in this series.
In our next phase we will test independent academic research-validated interventions.
The initial campaign targets a population of roughly 400,000 subjects, though for efficiency reasons a consistent subset of 25,000 receives the majority of the treatment and we reach about 8,000 of those each day. A separate control group receives no ads. We will treat the test group for several weeks, then invite it and the control to take a collaboration sentiment survey similar to that in Figure 1.
Figure 3 shows one of the ads used to invite participation in the survey. Based on prior experience using social media to invite subjects into surveys about collaboration sentiment, within our current budget we expect to generate responses from roughly 75 test and 75 control participants, for a total response group of 150. We will then compare and identify any differences.
For Phase 2, in addition to testing academic-driven interventions, we intend to measure actual behavioral changes in the treated group, to get beyond just measures of sentiment.
Ultimately, Project INFLUENCE is a flexible “harness” that can be used to test a wide variety of intervention strategies and their effect on both sentiment and behavior.