SaborData | Stats & ML

Ten stats and ML myths that sound right and aren't.

A free two-page cheat sheet. Each myth comes with what is actually true and one quick check you can run on your own work.

A model that is 99% accurate and finds nothing

100 patients. One of them has the disease.

Accuracy
?
Cases found
?

The model will predict "healthy" for every patient. Press the button to score it.

First page of the cheat sheet, showing the first five myths, each with the reality and a quick check
Page 1 of 2. A4, print-friendly.

What is inside

Ten myths, crossed out here. The PDF explains why each one is wrong and what to do instead.

  • A p-value is the probability you're right
  • Correlation implies causation
  • A 95% interval has a 95% chance of holding the true value
  • Not significant means no effect
  • High accuracy means a good model
  • More data fixes a biased model
  • Your data must be normally distributed
  • Low training error means a good model
  • Deep learning always beats a simple model
  • Preprocess everything, then split

Plus six rules of thumb you can apply to your next project.

Get the cheat sheet

Who makes this

SaborData posts one stats concept or ML myth every day on Instagram. The posts are AI-assisted and reviewed by a human before they go out. This cheat sheet collects the best-known myths in one place.