« Why tests are not perfect | Main | Know your data 26: Confidence in numbers dropping »


Feed You can follow this conversation by subscribing to the comment feed for this post.


Thank you, this is really interesting. I am quite often using the similar "one-way confidence bands" in my own professional work, when using sample date when all the sample observations either show a "Yes" or "No"
The "rule of three" introduced at the end could be very helpful for us. Do you have any links to papers or books where this is explained more in detail?


Magnus: Wikipedia has a page on Rule of Three. It literally is the calculation I did above. The only thing is it applies a standard approximation for the log function to simplify the formula. This "zero events" problem has been studied by statisticians a lot - that's because it's a special case where classical statistics lead to an absurd answer.

Antonio Rinaldi

There are two typos: in the following sentence 30 sould be 300:
The chance of 300 negatives in 300 samples is (0.99)^30 = 5%.
Even in the next one.


AR: Thanks for spotting them. Fixed. Yes they should be 300 not 30 in those two formulas. The cutoff for 5% was correctly stated at 99% specificity.

The comments to this entry are closed.

Kaiser Fung. Business analytics and data visualization expert. Author and Speaker.
Visit my website. Follow my Twitter. See my articles at Daily Beast, 538, HBR, Wired.

See my Youtube and Flickr.


  • only in Big Data
Numbers Rule Your World:
Amazon - Barnes&Noble

Amazon - Barnes&Noble

Junk Charts Blog

Link to junkcharts

Graphics design by Amanda Lee

Next Events

Jan: 10 NYPL Data Science Careers Talk, New York, NY

Past Events

Aug: 15 NYPL Analytics Resume Review Workshop, New York, NY

Apr: 2 Data Visualization Seminar, Pasadena, CA

Mar: 30 ASA DataFest, New York, NY

See more here

Principal Analytics Prep

Link to Principal Analytics Prep