Teaching Stata through real failure: What went wrong, what worked, and what we changed

Analysts and students are often taught Stata through polished examples that assume clean data, stable definitions, and linear analytic paths. In practice, applied research rarely unfolds this way. This presentation uses a real-world analytic project as a teaching case to examine how initial assumptions, data structure decisions, and workflow design can fail and how those failures become powerful instructional tools. We describe an early analytic approach that produced technically valid but misleading results due to hidden data fragmentation and flawed unit-of-analysis decisions. Through iterative revision, we restructured the workflow in Stata to reconcile multiple data sources, correct encounter-level logic, and embed quality checks that aligned analysis with real-world decision-making. Rather than focusing on syntax alone, we emphasize how analytic thinking evolved alongside the code. Presented as a dual-instructor narrative, this session demonstrates how teaching Stata through failure improves methodological rigor, transparency, and learner confidence. Attendees will gain practical strategies for teaching data management, model interpretation, and analytic judgment using imperfect data skills essential for applied work across disciplines.