I never really took statistics seriously as a pure mathematician, taking courses more just to remain employable than out of actual interest. But the more I learn and the more I use it, especially Bayesian statistics, the more I appreciate it as an elegant and efficient way to model and describe pretty much anything.
Links and resources #
- The German tank problem, an interesting example of a simple but ingenious statistical idea being used to determine the production rate of German tanks in the second world war.
- A very interesting paper on a Bayesian alternative to traditional $t$-tests. Implemented in Python here.
- A new correlation coefficient that seems to address many of the shortcomings of many commonly used coefficients. Seems to be significantly more involved to compute though, but theoretically a very clean idea.
- Anscombe’s quartet of four very different datasets with the same correlation coefficient (and many other common descriptive statistics).