Bayesian Beginnings helps academic researchers and applied teams design, fit, and validate computational models — and build the reproducible pipelines that make the work last.
Start a conversation →Custom hierarchical and generative models in Stan, PyMC, or brms — from sketch through prior/posterior predictive checks. For decision-making, learning, measurement, or whatever your domain calls for.
Reinforcement learning, drift-diffusion, evidence accumulation, and other process models. Bring me a behavioral paradigm and I'll help you formalize it, fit it, and check that it actually says what you think it says.
End-to-end workflows in R and Python using renv, uv, targets, and Docker. The kind of project structure that survives reviewer rounds, lab handoffs, and your own future memory.
Second opinions on study design, model selection, and validation strategy. Useful before grant submissions, during revisions, or when a project's analytical plan needs a sober outside read.
I've spent the last decade building Bayesian and computational models — first in academic psychology, then across health-tech and insurance. The thread connecting the work has always been the same: take a substantive question seriously enough to formalize it, then build something that can actually be fit, checked, and reused.
Bayesian Beginnings is the consulting side of that work. Most of my clients are researchers who have the domain expertise and the data, but want a collaborator who can move quickly between the modeling theory and the implementation — and who treats reproducibility as a design constraint, not an afterthought.
The hierarchical Bayesian modeling package for decision-making tasks — co-developed with the CCS Lab. Used across computational psychiatry, behavioral economics, and addiction research to fit dozens of established models with sane defaults.
James-Stein, classical true-score theory, empirical Bayes, ridge regression, and hierarchical models all converge on the same idea: shrink toward the group. Here's why that matters.
Read on Haines Lab →Behavioral data deserves models that take its data-generating process seriously. Summary statistics and two-stage analyses produce an impoverished view of the individual differences that actually matter.
Read on Haines Lab →Implicit and explicit attitudes are both latent — we only observe noisy indicators of each. What does it actually take to recover the relationship between them, and what happens when the IAT can't?
Read on Haines Lab →The best starting point is an email with a few sentences on what you're working on, what stage it's at, and what you're hoping to figure out. I usually reply within a couple of days.
modelme@bayesianbeginnings.com