Skip to main content


Here we go. It's going to be a long day...

I'm refreshing my knowledge of probability distributions. Every field of knowledge based on data uses them, but few stop to think about them.

Chapter 2 of Thom Baguley's Serious Stars

#science #mathematics #statistics
Yeah. I probably won't be reading this.
What are the chances of that?
I agree it's absurd to talk probability without a sense of what probability distributions are. And so is to imply a definite probability distribution to exist, when nothing supports it.
Indeed, probability distributions are theoretical. But they are pretty good at predicting future events
Any speech of probabilities implies probability distributions. Many people tend to over-generalize from the elementary case of throwing fair dices or coins, where the symmetry determines the probability distribution.
Last month I learned the term "Improper prior" - it's when your probability distribution for your bayesian evaluation starts out with a total sum greater than 1.
And I guess the risk of doing that might be greater if one doesn't know what probability distributions are? Maybe?

What are the odds?
If the sum is >1 that is a distribution but technically not a probability distribution. It only crossed my mind during our recent conversation that, despite their psychological unlikeliness, bayesian subjective probabilities could charitably be imagined to hide behind people's on the whole manifest abuse of the assumption of the existence of probability distributions in frequentist-form arguments. It works all the better if this bayesian framework of charitable interpretation doesn't require the secretly subjective probability distributions to actually be probability distributions at all.
We have compartmentalized cognition, and poor coordination between competing processes, so if the naughty sub-processes output a distribution with a sum over 1, we don't really seem to have a way to correct that.
Our thinking is very meta, but it's missing the meta-meta-supervision.