Dylan Poulsen Consulting
Thoughtful statistical consulting for clear, defensible decisions
Thoughtful statistical consulting for clear, defensible decisions
I work with teams on decision-focused statistical consulting: Bayesian causal inference, uncertainty-aware modeling, reproducible reporting, and analytical workflow automation. I am most useful when people face consequential decisions, messy data, and reporting processes that need to become clearer, easier, and more reliable.
You need to know whether an intervention worked, quantify uncertainty before a decision, make a recurring report less painful, or explain technical results to people who need to act on them.
I help teams reason under uncertainty, compare interventions, design analyses, and communicate what a model can and cannot tell us.
I build repeatable workflows with tools such as Quarto, R, Python, Shiny, and AI-assisted document routines. The goal is analysis that can be rerun, inspected, and handed off without unnecessary drudgery.
I review existing models, datasets, reports, and analysis workflows for interpretability, privacy risk, uncertainty, and whether the conclusions are well supported.
My data science work is guided by reproducibility, interpretability, openness, respect for privacy, and human flourishing. I am especially interested in tools that reduce drudgery, improve understanding, and help people make better decisions.
I clarify the decision, assess whether the available data can support a causal claim, choose an analysis strategy, and produce a concise recommendation with assumptions and limitations.
I turn a fragile manual report into a repeatable workflow using Quarto, R, Python, Shiny, or dashboard tooling. The goal is fewer manual steps, fewer hidden assumptions, and a report that can be rerun.
I review an existing analysis, model, or dataset for interpretability, uncertainty, privacy risk, and whether the conclusions match the evidence. This is useful before a model becomes operational or before a report drives a major decision.
Problem: Assess RNA quality quickly using instrument-derived data.
Approach: Developed a machine learning model with Thermo Fisher collaborators.
Deliverable: A model-driven analysis that contributed to a conference poster presentation, which can be found here.
As the John W. Allender Associate Professor of Data Ethics, I direct the Data Science Innovation Lab (DSIL) at Washington College. Through the DSIL, I mentor students through the same workflows I use in consulting on real-world projects with campus and community stakeholders. I apply the same rigor, documentation standards, and delivery discipline in both settings.
Recent DSIL work includes soccer player and ball tracking, data cleaning and visualization for partner organizations, and structured dataset workflows for research teams.
My work on the mathematically optimal way to cut an onion has been cited by The New York Times, The Pudding, Daily Mail, and outlets around the world. The project is a useful example of how I work: take a concrete question, build a model, explain the assumptions, and communicate the result clearly enough that people remember it.