14493 - AI-Assisted Statistics: Exploring Data with ChatGPT & Co Modulübersicht

• Prof. Dr. Urbig, Diemo

After successfully completing the module, students are able to apply model-based estimation approaches to empirical research questions in business and social science contexts. They can translate substantive theoretical arguments into suitable empirical models, use AI-supported tools for implementation in a controlled and reflective way, and independently assess whether the resulting estimations are appropriate for the research question. Students are able to interpret estimation results beyond a purely technical reading of output tables, evaluate the assumptions and limitations of different modelling choices, and distinguish between descriptive, predictive and causal interpretations. They develop the ability to critically examine potential sources of bias, uncertainty and misspecification, and to communicate empirical findings, including their limitations, in a transparent and responsible manner. The course will also foster critical thinking, enabling students to reflect on the opportunities and limitations of using AI in research, as well as the ethical implications. 

The module focuses on AI-assisted model-based estimation in empirical business and social science research. Students use AI tools to support implementation, coding, debugging, documentation and exploration of model alternatives. However, the central objective is not the delegated implementation itself, but the students’ ability to understand model logic, interpret estimation results, assess assumptions and limitations, and communicate findings responsibly, specifically: 

• Model-based estimation in empirical research: research questions, theory, data-generating processes, model specification and interpretation, incl. structural models versus reduced-form models
• AI-assisted empirical workflow: using AI for coding, debugging, model implementation, documentation and interpretation checks; limits of AI-generated code and explanations
• OLS beyond simple linear regression: transformations, nonlinear terms, interaction effects, moderation, dummy variables and categorical predictors
• Interpretation of OLS results: coefficients, predicted values, simple slopes, confidence intervals, standard errors and model diagnostics
• Generalized linear models: logit/probit models, count models, link functions, predicted probabilities and average marginal effects
• Structural equation models: path models, latent variables, measurement models, confirmatory factor analysis, mediation, model fit and comparison of alternative models
• Endogeneity and identification (omitted variables, reverse causality, simultaneity, selection problems and unobserved heterogeneity) and strategies for dealing with endogeneity: theoretically justified control variables, instrumental variables, control-function logic, instrument-free correction approaches and sensitivity analyses
• Exploring limitations of estimations: assumptions, robustness checks, outliers, missing data, multicollinearity, model dependence, generalizability and ethical risks
• Communication of results (model equations, estimation tables, marginal-effect plots) and transparent reporting of assumptions and limitations
• Hands-on applications with individualized datasets and AI-supported implementation, applications based on student ddata set, Marketing Mix Attribution models, and success factor research in management

• Basic knowledge of statistics, ideally including descriptive statistics, statistical inference and simple linear regression.
• Prior experience with empirical research questions and basic data handling is helpful, but not required.
• Programming knowledge is not required, but students must be willing to work with AI-assisted statistical software workflows, which includes generating programming code.

• Lecture / 4 Hours per Week per Semester
• Self organised studies / 120 Hours

Lecture materials, datasets, and analysis scripts will be provided via Moodle. Literature references will be announced at the beginning of the course.

Assessment Mode for Module Examination:

• Part 0 (data collection): Participation in a survey, 30 minutes; contributes 5 points to the total score.
• Part 1 (take home): Individual take-home homework with individualized data, pass/fail; passing the homework contributes 10 points to the total score.
• Part 2 (bring own laptop): Individual in-class practice session with individualized data and online submission of results and answers to questions, 60 minutes, 35 points.
• Part 3 (bring own brain and pen): Written exam, consisting of multiple-choice and free-text questions, 50 minutes, 50 points.

A maximum of 100 points can be achieved. The final module grade is based on the total number of points.

• Master (research-oriented) / Artificial Intelligence / PO 2022 - 1. SÄ 2024
• Master (research-oriented) / Betriebswirtschaftslehre / PO 2017
• Abschluss im Ausland / Informatik / keine PO
• Abschluss im Ausland / Stadtplanung und Städtebau / keine PO
• Master (research-oriented) / Transformation Studies / PO 2024
• Bachelor (research-oriented) / Wirtschaftsinformatik / PO 2024

Module is continuous assessment, hence, register or deregister in teh university exmaniation system before the deadline announced both via the university semester schedule and the Moodle course pages. You cannot register or deregister and we cannot register or deregister you after that deadline.

Lecture

• no assignment