Social Cohesion Programming Randomized Controlled Trial
From September to December 2025, Galvanize Action ran a field experiment to examine the effectiveness of its nonpartisan, nonprofit affiliate Galvanize USA’s values-based programming designed to foster social cohesion among ideologically-moderate women.
Results indicate that Galvanize USA’s social cohesion programming had a statistically significant effect on treatment group participants’ reported willingness to openly challenge friends or family who express discriminatory views (p = 0.03, +0.09). Additionally, results indicate the program did not show any significant negative effects of programming on social cohesion attitudes, including among ideologically-moderate women of color.
BACKGROUND
In 2025, Galvanize USA and Galvanize Action developed a values-based programming strategy designed to foster social cohesion among ideologically-moderate women. A social cohesion campaign was deployed to audiences in Pennsylvania, Wisconsin, and Michigan through algorithm-driven programs between July and December 2025.
To explore the effectiveness of this strategy, Galvanize Action conducted a randomized controlled trial in conjunction with the deployment of the program. Participants in the treatment group were targeted to receive Galvanize USA programming; participants in the control group were restricted from first party targeting. All participants were surveyed before and after the programming window to assess whether and how the values-based content led to changes in prosocial attitudes among ideologically-moderate women.
DESIGN
A total of 3,472 ideologically-moderate women were recruited for and participated in the field experiment. In the post-test, we received survey responses from 2,121 women (61% of the full sample). See APPENDIX for a table displaying demographic breakdowns of the final sample.
Women who completed a baseline survey were randomly assigned to one of two groups: a treatment group (50%) and a control group (50%). Conditions were balanced based on demographic makeup to control for potential biases. Women in both groups completed the surveys, but only women in the treatment group received Galvanize USA programming. See APPENDIX for a descriptive list of programming targeted to the treatment group.
To measure social cohesion attitudes, both the pre-test and post-test asked participants to what extent they agreed with the following statements on a five-point scale:
- Efforts to increase diversity always come at the expense of people like me.
- The government has gone too far in helping minorities to the disadvantage of people like me.
- I am willing to openly challenge friends or family who express discriminatory views, even at the risk of personal discomfort or disagreement.
RESULTS
It is common for experiments like this to have attrition, which is when participants complete the baseline survey but do not return at the end of the programming window to complete the post-test survey.
When examining the final dataset, Galvanize Action noticed that there was a pattern among attrited respondents. People with higher baseline scores on the third success measure in the treatment group were less likely to drop out compared to control group participants with similar baseline scores on the third success measure. So the treatment group retained people with higher baseline scores, while the control group lost them at similar rates regardless of baseline score. This is an indication that something about the programming turned away people with lower baseline scores on the third success measure.
To account for this, Galvanize Action used Inverse Probability Weighting (IPW), which adjusts for attrition bias by using predictive modelling to assign weights to each participant based on their probability of remaining in the study.
When controlling for demographic and pre-test variables1, we see no significant difference between the treatment group or the control group on the first two success measures. On the third success measure, we see a small statistically significant treatment effect (p = 0.03, +0.09).
Results among the subset of white women in the sample mirror the results of the full sample closely, with the first two success measures showing no effect of treatment and the third success measure having a small treatment effect (p = 0.05, + 0.09). Among women of color, we see no significant treatment effect across any of the three success measures.
Results on the third success measure are promising, as we see treatment effects among both the full sample of ideologically-moderate women and the white women subset. The attrition bias indicates that despite being effective on the third success measure, the programming did turn away participants who were resistant to it (participants with lower baseline scores on the third success measure).
APPENDIX
DEMOGRAPHICS
| Final Sample (n = 2,121) | |||
| Race | Ideology | ||
| White Women | 78.5% | Somewhat conservative | 25.7% |
| Women of Color | 21.5% | Moderate | 51.7% |
| Urbanicity | Somewhat liberal | 22.6% | |
| Rural | 24.0% | Education | |
| Suburban | 26.7% | College degree | 41.4% |
| Urban | 16.7% | No college degree | 58.6% |
| Age | Household Income | ||
| 18 – 24 | 4.7% | $0 – $9k | 3.6% |
| 25 – 34 | 11.1% | $10k – $24k | 12.9% |
| 35 – 44 | 20.2% | $25k – $49k | 25.4% |
| 45 – 54 | 19.5% | $50k – $74k | 24.5% |
| 55 – 64 | 18.6% | $75k – $99k | 16.8% |
| 65 – 74 | 20.0% | $100k – $149k | 10.1% |
| 75 + | 5.9% | $150k + | 6.6% |
TREATMENT
Programming ran from September 17, 2025 to December 3, 2025
Programming ran from October 3, 2025 to December 3, 2025
- Using multiple linear regression (with inverse probability weighting to account for attrition bias) to test if treatment and other covariates (age, education, urbanicity, income, ideology, and baseline social cohesion attitudes) significantly predicted post-test social cohesion attitudes. ↩︎
