Selection makes false findings from sampling & data processing. e.g. Let happiness be constant from birth, increases prob of marriage. As ppl age, more are married. Conditional on marriage, neg corr btw age & happiness. Below: each point a person aging, red = married Of course happiness is not constant from birth, but the example shows that stratifying by a seemingless innoccent variable like marriage status can create phantom associations of large size, should not be interpreted as causal. This example from Chapter 6 of my book

On another site, someone asked me how I feel about "standardized effect sizes". Like many statisticians, I don't much like them, when used to compare "effect sizes". Some sources in next tweets. First, this article from Sander Greenland and colleagues. (Why is it that whatever the applied stats topic, there is always a useful Greenland article?) doi.org/10.1093/oxford…

Forgive me, for I am about to Bayes. Lesson: Don't trust intuition, for even simple prior+likelihood scenarios defy it. Four examples below, each producing radically different posteriors. Can you guess what each does? Revealed in next tweet >> Huzzah! Posterior distributions in red. The shape of the tails, which isn't so obvious to the eye, can do weird but logical things.

Week 1 of Statistical Rethinking 2023 is done. Here are the memes from this week's lectures The Spidermen: Causal inference, descriptive studies, and research design are alike in that they all depend upon some generative/scientific model of how the sample was produced

Okay so I made a transparent gif of Brandenburger Hasselhoff, in case anyone wants to add him to other historical events. Here he is e.g. at Zeppelinfeld in 1945 Hasselhoff alone for your creative pleasure

Working with a colleague on some household income data, where work is irregular. As usual, I start by writing a synthetic data simulation to talk through with colleague. Helps to ensure I understand the problem right. Also brings up fun (for me) like sources of measure error. In this case, income data are reports and almost certainly suffer rounding and heaping. It's the little things like this that make even simple exercises not so simple.

Many performers of music cannot read it. Okay. There are other, often more intuitive, ways to learn music.

Scientists perform stat models. Most scientists cannot read them. This is less OK, but there are other ways to learn models.

Short thread in which I strain this comparison If you don't read music, the Rzewski excerpt above (left) is meaningless. If you do, it is perfectly clear. You'd read it not by each individual note, but through higher structure like chords & arpeggio patterns & progression.

It's not the notes so much as their relationships.

Cats reacting to bad COVID-19 models, a short thread. [This is beneath me but I really need this right now let me have this]

Model 1 Model 2

In my dept today, I gave a bare minimum proof of why natural selection can favor strategies that do not maximize reproductive rate, provided variance is also reduced. Some papers to start with, if this literature is unfamiliar: > Good place to start is this 2012 paper by Starrfelt & @kokkonutter : doi.org/10.1111/j.1469…