The science behind the score.
Every Grow Progress test result is proven in a randomized controlled trial, the same rigorous testing environment used in medical research. Here's exactly how it works, start to finish.
Scroll through the experiment: one stage at a time, from a representative panel of real people all the way to a weighted, statistically grounded score.
We recruit a large pool of respondents with demographic quotas on age, gender, race, ethnicity, and geography — so the people in our panel reflect the people you're trying to reach.
Need to reach likely voters in Arizona? Union households? Parents of school-age children? We filter to your target audience and size the sample to deliver statistically significant power.
Some respondents are screened out before they ever see your message, but most quality control happens in post-processing, after the survey is complete. Attention checks, speeder traps, and consistency filters catch low-quality responses, while a mix of internal traps and external quality integrations screens out AI bots. Flagged responses are replaced — not averaged in — before you see your results.
The qualified sample is randomly divided into groups — one for each message being tested, plus a control group that sees none. Random assignment is what makes causal inference possible — it's the same logic as a clinical trial. At this point, the groups are statistically identical.
The control group reads neutral, unrelated content. Each test group sees one of your messages — a video, image, ad, or written message. Exposure to these different conditions is how we isolate whether each message has its intended effect.
After exposure, every group completes an identical questionnaire — the same success question, the same response scale, the same order. The only thing that varied was what they read beforehand.
We apply raking weights to align the sample to real-world population benchmarks. Extreme weights are trimmed to prevent any single respondent from distorting the result.
Your Grow Progress Lift is the weighted difference in responses between test and control. We report it with a margin of error and significance threshold — so you always know how much to trust the result.
The same lift calculation is run for each demographic and political subgroup — age, gender, race, education, party ID, ideology, and more. A message that works overall might backfire with persuadables. Now you know.
Causal measurement. No selection bias. No guessing whether your message drove the result — or whether it was something else entirely.