Teaching Statistics and Chemometrics Using Large Language Models: The Palmer Penguins Case Study

This study investigates the use of large language models (LLMs) to support the teaching of statistics and chemometrics within a STEAM framework. Thirty-nine questions involving correlation plots, Pearson correlation heatmaps, box plots, histograms with kernel density estimates, hypothesis testing, post hoc analyses, and pairwise scatter matrices were applied to three data sets: the Palmer Penguins data set, used to introduce foundational statistical concepts, and two chemistry-oriented data sets (wine and COVID-19 antibody profiling data; Anal. Chem. 2022, 94, 6529–6539). The activity was structured in three stages. First, students submitted the questions directly to the LLMs using natural-language prompts and evaluated the generated responses. Second, results were independently verified using Jamovi to assess numerical accuracy and appropriate statistical test selection. Third, to ensure computational reproducibility, Gemini was prompted to generate R scripts to reproduce the analyses. When the 39 questions were submitted to Microsoft Copilot (GPT-5) and Google Gemini (3), both models produced consistent statistical results without numerical discrepancies relative to independent software verification. The R scripts generated by Gemini reproduced the analyses successfully; minor graphical rendering issues were resolved by correcting the syntax based on R console feedback. These findings indicate that LLMs can function as reliable pedagogical tools for teaching statistics and chemometrics, facilitating data visualization, hypothesis testing, and multivariate exploration while maintaining reproducibility when outputs are critically verified.