Data Science Fact

Data science consulting johndcook.com/blog/applied-s…

'Two words explain the classical preference for parametric models: mathematical tractability.' -- Brad Efron & Trevor Hastie

A data model is not just a likelihood. statmodeling.stat.columbia.edu/2026/03/20/a-d…

Probability that a number is prime johndcook.com/blog/2010/10/0…

'The statistician cannot evade the responsibility for understanding the process he applies or recommends.' -- Sir Ronald A. Fisher

When you hear that a method is optimal, ask 'By what criteria?' and 'Subject to what constraints?'

Naive Bayes is called 'naive' because it assumes things are independent that definitely are not. And in the right setting works well anyway.

Military standard for acceptance sampling johndcook.com/blog/2022/10/2…

In Soviet statistics, a statistician had no degrees of freedom

If hat(theta) is a the MLE of theta, f(hat(theta)) is the MLE of f(theta) for any function f.

A reference prior maximizes the expected Kullback-Leibler divergence of the posterior distribution relative to the prior.

Good statistical practice can increase your chances of being correct while decreasing your chances of being published. Conflict of interest.

Dose finding ≠ dose escalation johndcook.com/blog/2019/02/1…

What every statistician should know about cross-validation robjhyndman.com/researchtips/c…

If X has finite variance σ², the variance of the mean of n samples from X is σ²/n.

Old one, but a classic.

'However beautiful the strategy, you should occasionally look at the results.' -- Winston Churchill (?)

Conjugate models implemented as functional folds johndcook.com/blog/2016/06/0…

mathematical statistics versus applied statistics x.com/nestroytimo/st…

The inverse gamma is a conjugate prior for a normal sampling distribution with known mean and unknown variance.

'Bias is a learner's tendency to consistently learn the same wrong thing.' -- Pedro Domingos