For three and a half years I was a statistical verifier at the UNC Odum Institute, where my job was to independently reproduce the analyses behind accepted journal manuscripts before they were published — rebuild every headline result from the raw data in Python, R, and Stata, and document each discrepancy. Specification errors. Silently dropped rows. Weights applied twice. Numbers that could not be regenerated from the files the authors themselves supplied. This was careful work by careful people, and I found problems in it regularly. I am a computational physicist — Ph.D., UNC Chapel Hill; postdoctoral researcher at the University of Maryland in an NSF research hub. My own research is large-scale Monte Carlo simulation and the statistics that turn simulation output into a defensible number with an honest error bar, which means separating statistical uncertainty from systematic uncertainty rather than reporting one and hoping it covers both. Nine peer-reviewed publications. What I do for you -Reproduction audit. You send data and code. I rebuild your results from scratch and report exactly what reproduces, what does not, and why. Specification review. Whether the model you fit answers the question you asked — controls, functional form, weighting, clustering, and the assumptions holding up your standard errors. -Uncertainty review. Whether your stated error bars are credible, what they do and do not cover, and how much of your conclusion survives a reasonable change in the choices behind them. -A written memo. Short, plain, and handable to a co-author, a reviewer, a board, a funder, or counsel: what I checked, what I found, what I could not verify. -How I work. Every engagement is fixed scope and fixed price, agreed before I start. Python, R, and Stata; I also read C and Fortran. English or Spanish. Your data and your findings stay confidential. Typical engagements Second-opinion read — US$300 (2 hours). One analysis, one memo. Is there anything here I would be embarrassed by in review? Reproduction audit — US$900 (6 hours). Full rebuild of a paper, report, or filing from raw data, with a written findings memo. Ten business days. Standing reviewer — by arrangement. A few hours a month, same person each time, for a group that publishes regularly.

Location

USA