Seven ways numbers fool people. Each one is a short post, a real case from the news, a small dataset analysed with pandas, a figure, and an animated figure. Everything runs in a notebook you can download and run.
Open the animated web page · Start with the notebooks · Read the posts
| # | Mistake | Real case | Notebook | Open in Colab |
|---|---|---|---|---|
| 1 | Survivorship bias | The bombers that came back, 1943 | 01_survivorship_bias |  |
| 2 | Mean instead of median | US wealth: mean 551,000, median 108,000 | 02_mean_vs_median | |
| 3 | Relative risk without absolute risk | Bacon and the 18 percent, WHO 2015 | 03_relative_vs_absolute_risk | |
| 4 | A chart axis that does not start at zero | Fox tax chart 2012, Georgia chart 2020 | 04_truncated_axis | |
| 5 | Correlation is not causation | Chocolate and Nobel prizes, 2012 | 05_correlation_vs_causation | |
| 6 | Base rate neglect | The 99 percent accurate test, Cardiff 2017 | 06_base_rate_neglect | |
| 7 | Publication bias | 74 antidepressant trials, 2008 | 07_publication_bias | |
Every notebook is saved with its outputs, so you can read it on GitHub without running anything.
 |
 |
 |
 |
 |
 |
On your computer:
git clone https://github.com/CS42org/data-make-sins.git
cd data-make-sins
pip install -r requirements.txt
jupyter labOpen notebooks/00_start_here.ipynb and follow the links.
On Google Colab: click a badge in the table. Run the first cell, it downloads the repo.
notebooks/ one notebook per mistake, plus 00_start_here
data/ small CSV files used by the notebooks
figures/ every figure and GIF the notebooks produce
posts/ the seven posts as plain text, and the master list of 50 mistakes
docs/ the animated web page, published with GitHub Pages
src/dms.py shared colours and plot style
The seven posts are the ones people fall for most often. The full list of 50 mistakes, grouped in six families, is in posts/data-mistakes-list.md.
Every real case links to the original reporting inside the post and the notebook. The main ones:
- Levitt, Freakonomics, 2008: Good to Great companies after the book
- UBS Global Wealth Report 2023: mean and median wealth per adult
- The Guardian and Cancer Research UK, 26 Oct 2015: processed meat and bowel cancer
- Fox Business, Cavuto, 31 Jul 2012, via FlowingData: the tax chart
- Business Insider, May 2020: the Georgia chart with dates out of order
- Messerli, New England Journal of Medicine, 10 Oct 2012, and Reuters: chocolate and Nobel prizes
- BBC News, 4 May 2018: face recognition at the Champions League final
- Turner et al, New England Journal of Medicine, 17 Jan 2008: selective publication of antidepressant trials
- Cochrane and BMJ, 10 Apr 2014: the Tamiflu data
Inspired by How to Lie with Statistics (Huff), The Art of Thinking Clearly (Dobelli) and Thinking, Fast and Slow (Kahneman).
Two datasets are not real measurements. chocolate_nobel.csv holds values read from the published figure, so they are approximate. georgia_cases_illustrative.csv is invented to show the trick, not the real Georgia counts. Both are marked in the notebooks.
MIT. Use it in class, in a talk, in a blog. A link back is welcome.