5 comments

[ 3.2 ms ] story [ 26.1 ms ] thread
This is probably the most often forgotten rule among engineers, software developers etc:

> Rule 5: Statistical Analysis Is More Than a Set of Computations

I think I get asked once a week, "can't we just run it through a neural network and be done?"
Oh my. As if an algorithm can answer all problems. The algorithm is only a tool to analyze the data. The real science is understanding what the underlying data is telling you about the hypothesized behavior.
I thought this was sort of funny given the authors published a set of rules.

Conjecture 1: Statistical practice is more than a set of rules.

I am not complaining about the quality of the list, but I fear the authors have conflated several senses of the word "rule" in this list.

I have an intuition that "signal comes with noise", isn't the same kind of rule as "keep is simple."

It might be useful to break it out into a checklist of things to look at, philosophical principles to keep in mind, and useful tips or something.