Mental Models to Use and Avoid as a Data Scientist
summary
Uses Jupyter Notebooks with runnable Python to show concrete examples and shares slides and code via a bit.ly link. Covers specific pitfalls (p‑hacking, garden of forking paths) with numbers: a GitHub pull‑request effect ~1% and the soccer study where 29 teams 61 analysts produce widely varying results and a 69% chance of a positive finding. Demonstrates mitigation: blinder that masks coefficients with NaN for blind analysis and discusses Thresholdout reusable holdout. Watch if you want code, data, and concrete fixes; skip if you only want high‑level theory.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:03P-hacking and researcher degrees
- 3:09Gender bias in pull requests
- 8:48Fishing expeditions and multiple comparisons
- 9:30Garden of forking paths
- 11:36Avoiding just-so storytelling
- 13:24Consider the opposite model
- 16:34Many analysts one dataset
- 20:25Blinding to address uncertainty
- 25:44Key takeaways
- 27:49Q&A on p-hacking
- 29:57Q&A on holdout sets
- 31:41Q&A on automated analysis
- 32:46Q&A on best practices
- 33:50Q&A on data splits
- 35:16Closing remarks
related talks
41mFootball's data analytics in python
28mUsing Python in a Data Hackathon
04h 02mHands-on Session on Data Science for Beginners
02h 31mTech Tutorial: Data visualization in Python
01h 53mAnalyze your data in Python
28mData Science Design Patterns
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
51mData Literacy