Behavioral Conversion Lab

The Thesis

Behavioral science content is everywhere, and most of it treats every cited study with the same authority — a decades-old finding with an under-powered sample gets repeated with the same confidence as something that’s been replicated across dozens of labs. I wanted to apply the same rigor I use to evaluate a client’s A/B test to the behavioral-science literature itself: what was the sample size, how long did the test or study run, was the experimental design actually built to isolate the effect being claimed, and has it held up under replication.

What I Built

A publication built around sorting the replication crisis’s fallout into something useful — not just reporting that a lot of psychology doesn’t replicate, but going study by study, applying an experimentation-statistics lens to older, foundational behavioral-economics research, and separating what’s genuinely valid and reproducible from what’s popularized but built on a shaky sample or a study design that couldn’t have shown what it claimed.

What Happened

Running it as its own site meant splitting a small, specific body of research from the readers who most needed it: growth and marketing people citing behavioral science to justify a decision. The research and the audience that needed it lived in two different places. I folded the work directly into the replication-crisis research now published on atticusli.com, next to the growth and experimentation content it’s actually meant to inform.

What I Learned

A research property this adjacent to an existing audience compounds faster as a section of that audience’s home than as a new domain competing for its own attention.

• Splitting a small, specific body of research from the readers who most need it dilutes both.
• Rigor is the differentiator — checking whether a study replicates or was adequately powered is a step most “behavioral science” content skips.
• Folding research into an existing platform is often the stronger distribution decision than standing up something new.