Most organizations running experiments on people — A/B tests, user studies, surveys, model evaluations — are making expensive decisions on designs that don't hold up. Underpowered samples. Broken randomization. Metrics that measure the wrong thing.
I'm a methodologist who teaches this at the graduate level and writes the code to execute it. I review your design, tell you whether the result is real, and show you how to fix what isn't.
I also work at the intersection most teams can't staff: measurement for AI systems. Model evaluations are human-subjects experiments. They need someone who understands both the statistics of valid measurement and the machine learning context. That combination is rare. It's what I do. (Start with the checklist if you want to see the approach.)
Designing valid model evaluations, eliminating bias in human rating protocols, and knowing whether "Model A beat Model B" survives scrutiny. Measurement rigor for ML teams.
A/B tests, pricing experiments, user studies. Power analysis, randomization integrity, construct validity, multiple-comparison correction. I tell you if the result is real.
Qualtrics builds with proper scale construction, randomization, and skip logic. Prolific sampling that gives you clean, usable data the first time.
You have data and need answers. I run the models in STATA or R and write up results you can defend — mediation, moderation, multilevel, longitudinal.
Discovery calls are free. Project rates are quoted up front — no hourly surprises. Retainers available for ongoing work.
Tell me what you're trying to measure.