Statistical Rethinking Workthrough Blog

Author

Josef Fruehwald

Causal DAG typology

I get confused thinking about the causality DAGs in terms of models. The arrows and paths in the DAGs don’t directly translate to terms. in the…

Josef Fruehwald

Where I do a lot of work and don’t understand colliders any better.

For me to really get a sense of how coliders work, I’m going to have to simulate a few different datasets, messing around with the…

Josef Fruehwald

Colliders

When A has an effect on Z, and B has an effect on Z, then Z is a “collider” variable between A and B.

Josef Fruehwald

Multicollinearity & Post-treatment bias

Here’s what McElreath says about multicollinearity

Josef Fruehwald

Haunted DAGSs

The first example in the book is about Berkson’s Paradox, which I believe is a kind of selection bias. The question is “Why do so many research results that are newsworthy seem unreliable?” The idea being that funding (or…

Josef Fruehwald

DAGS part 3

Ok, for the milk data, where the neocortex percentage is correlated with bodymass, how would we get the expected effect of increasing body mass?

Josef Fruehwald

DAGs part 2

For part 2, I’m going to try working through this step by step like he does in the book.

Josef Fruehwald

DAGs part 1

So, as a linguist, the only Directed…

Josef Fruehwald

Reporting linear model parameters

Table libraries

Josef Fruehwald

Reporting a linear model

Part of why I’m working through Statistical Rethinking as a blog is so that I can take some time and mess around…

Josef Fruehwald

Linear models, part 2

I think I’ve improved on the model diagram from the last post. Some things I’m still struggling with:

Josef Fruehwald

Predictive Distributions

One last post to work through different predictive distributions given 6W and 3W.

Josef Fruehwald

Linear Models: Part 1

“Suppose you and a thousand of your closest friends line up in the halfway line of a soccer field.”

Josef Fruehwald

Sampling Summaries

This isn’t MCMC sampling, but I’ll plot it as a line just for consistency for how MCMC chains look.

Josef Fruehwald

Starting Sampling

Plot of the base rate vs P(vampire | positive test)

Josef Fruehwald

Garden of Forking paths part 2

I’d like to re-represent the Bayesian Update in a nicer GT table. Some options are

Josef Fruehwald

01 Golem of Prague Chapter

Rejecting Model 1 does not result in a unique identification of a process, or even a hypothesis.

Josef Fruehwald

Setup

I’ve set up the blog using the default quarto blog template in RStudio, also initializing a git repo and renv.

Josef Fruehwald

02 Small Worlds and Large Worlds

In the analogy, models are “Small”, self-contained worlds.

Josef Fruehwald