STAT RIT, Fall 2026 AI in Statistical Consulting
Fall 2026tentatively Mondays 3pm beginning 9/14/26, Room PHYS 2122
Organizer: Eric Slud, Statistics Program, Math. Dept. Office: Mth 2314, x5-5469, email: slud@umd.edu
Description: This RIT is intended to explore how AI can play a constructive role in Statistical Consulting. We will discuss some background on what statistical consultants do, but mostly will explore in terms of actual datasets and interaction with Large Language Models how the process of turning data and scientific research questions (for clients likely not well versed in statistics) can be speeded up and made useful on the way to constructing informative, theoretically supported data analyses close to publication quality.
Desired background for participants: interest in data analysis, some knowledge of statistical theory, and computing familiarity with some statistical-analysis platform or language (R, Python, whatever) and willingness to experiment with LLMs.
Organizational help in the form of suggested datasets or guest presentations -- anything on this "AI in Consulting" topic -- will be welcome and much appreciated.
Credit: As in RITs generally, MATH, AMSC and STAT students may register for the RIT for 1 credit. The expectation is that registered students will attend regularly and give at least one presentation, possibly shared with another participant. The ideal presentation would describe/exhibit an interactive AI session role-playing a consultation with AI to inform research questions posed about a real or simulated dataset, not necessarily one that has been fully `cleaned up' (i.e., could have missing data, data errors or inconsistencies, etc., just like real life).
Examples of written source materials:
• a `5-step approach' to consultingAdditional AI-related Handouts:
(a) Some examples of interactions with Gemini LLM on templates for submitting consulting requests to AI to achieve desired outcomes: background literature on AI in Stat Consulting, and specifying assumptions, and Presenting data to AI for Stat Consulting"
(b) One rich source of data, with lots of preparation to do in terms of choice of variables and "allocation" flags, is the American Community Survey PUMS ("Public Use Microdata Sample") files. You can extract data from this subsample of the full ACS microdata using the R package "tidycensus", as indicated here.
(c) Here is a rough template, derived from preliminary interactions with Gemini in connection with consulting I have done in the past year with real clients, for how one can effectively submit consulting questions to AI for help in organizaing a consultant's response to a client problem.