"Data Morph: A Cautionary Tale of Summary Statistics"
summary
Demonstrates that identical means medians and Pearson correlations can hide wildly different distributions and introduces Data Morph, an installable Python package that preserves mean standard deviation and correlation to multiple decimal points while transforming arbitrary point sets into target shapes using simulated annealing. Shows a command-line entry point Data morph, animation output, and practical limits: bald spots, slow convergence and example runtimes like 50,000 iterations for a dinosaur-to-circle morph.
AI-written summary · about
chapters (14)
- 0:00Data Morph project
- 1:06Summary statistics limits
- 3:34Anscam quartet pitfalls
- 4:59Hypothesis bias
- 6:45Datasaurus limits
- 7:49Data Morph tool
- 9:56How Data Morph works
- 12:43Designing shapes
- 14:27Morphing algorithm
- 16:33Limitations and future work
- 21:12Research to package
- 25:46Open-source Data Morph
- 26:48Key lessons
- 28:14Q&A on misleading stats
related talks
28mData Morph: A Cautionary Tale of Summary Statistics
01h 38mEffective Data Visualization
08mIt might look normal but this distribution will ruin your stats
43mWhat data and design can learn from each other
34mDimensionally reducing data squeezing out the good stuff
02h 23mEffective Data Visualization
01h 53mAnalyze your data in Python
50mThe Shape of Data: Machine Learning and Topology