The common thread is measurement. Most engagements come down to establishing what a measurement can and cannot support, then designing the study or sampling scheme that makes the resulting decision defensible.
Sample size and power determination, design of experiments and response surface methods, monitoring network design, stratified and multistage sampling.
Detection limits and minimum detectable concentration, guard bands, compliance demonstration, decision-error and data quality objectives frameworks.
Dose–response and lethal dose estimation by probit analysis, potency assays, lot release and stability, exposure–response modeling, non-compartmental and compartmental pharmacokinetics.
Statistical analysis plans, Phase 1 dose escalation through Phase 3, time-to-event and longitudinal methods, CDISC SDTM and ADaM datasets, tables listings and figures.
Test-retest reliability, inter-method agreement, covariate-adjusted normative ranges, multi-site trial design, sensitivity and specificity with exact confidence intervals.
Capability and performance indices, statistical process control, gauge R&R and measurement system analysis, method and assay validation, root cause investigation.
Design-based estimation with weights, strata and primary sampling units; replicate-weight variance estimation; SAS SURVEY procedures.
Sampling plan and monitoring network design, trend analysis, geostatistical modeling, power calculations and cost-benefit analysis of sampling effort.
Claims analysis, all-payer claims databases, utilization management, denials and appeals, risk adjustment, comparative effectiveness.
Production R/Shiny applications — the recurring deliverable is a tool the client can rerun, not only a report.
Eight production applications, built so that method reaches the people making the decisions.
Sample size determination, data imputation and outlier detection for animal researchers. Developed under NIH SBIR Phase I award R43DA041758 (NIDA).
Automates raw clinical data through CDISC SDTM and ADaM dataset creation for standardized reporting and regulatory submission.
Covariate-adjusted normative range comparison supporting clinical and regulatory review.
Interactive analysis package placing method selection and execution in the hands of non-statisticians.
Spatial identification and resolution of audit findings for quality and manufacturing users.
Including response-surface tooling and a scan-MDC design harness supporting the NUREG-1507 work.
Most work arrives as one of four situations. A regulator or auditor has asked a question the existing analysis does not answer. A measurement system has been challenged and nobody can say whether the problem is the product or the instrument. A study is being designed and someone needs to know whether it can support the claim before money is committed. Or a model has produced a number that does not look right, and the question is whether to trust it.
All four are judgment problems before they are computation problems, which is why the principal does the work rather than specifying it for someone else. The first conversation is usually about what the data can support, not about method selection.
Written answers to questions that come up repeatedly, with thin coverage elsewhere.
Why classical detection limits fail on a serially correlated stream, and how lag-k differencing restores control of the error rates.
What the agency is actually asking on the statistical side of an active IND, and how to close the question rather than open a new one.
Why raw denial rates are not comparable across plans, and what claims data structurally cannot observe.
Why the confidence interval matters more than the point estimate, and why the case definition does more work than the model.
Study and sampling design; regulatory decision-making under measurement uncertainty including detection limits, guard bands and data quality objectives; preclinical and nonclinical statistics including dose-response and lethal dose estimation, potency assays and lot release; clinical biostatistics including statistical analysis plans, CDISC SDTM and ADaM datasets and TLF production; medical device and diagnostic validation; process capability and measurement systems including gauge R&R and statistical process control; complex survey methodology; environmental and radiological monitoring; health economics and real-world evidence; and production R/Shiny statistical software.
Anything from a half-day review of a design before submission to a multi-year role as named statistician on a funded program. The short engagements are often the highest value: catching an underpowered design or a mis-specified endpoint before data collection costs a fraction of discovering it afterwards.
Frequently — it is the main route for federal work. MRP Group holds GSA Multiple Award Schedule contract 47QRAA25D000P with an awarded Statistician (Ph.D., 10 years) labor category, so a prime can add the firm to a team without a rate crosswalk or separate negotiation. See Federal Contracting.
Yes, including as Co-Investigator. Statistical review of a draft proposal before submission is also available on its own, and is often where the most value is added — a power analysis error found before submission costs nothing to fix.
