[ggggtez]

It's a bit confusing. The author starts with the assumption the show got worse, and assumed a causal relation with an random piece of data. This is not proper Bayesian statistics.

The author claims there is no principled way, but there is. Confidence bounds are the way to measure what affect a variable had on an outcome.

[nathcun]

I didn't assume a causal relationship. In the last line I state that a change in writers may not have caused a decline in quality, but the change in the make-up of the writing team around the team most people agree the show went downhill would lend some credibility to the idea.

What about that article suggested I was attempting to do proper Bayesian statistics? It's an exploratory analysis, I'm just looking at the data.

And where did I claim there was no principled way of doing this analysis?

[ggggtez]

To be fair, you said "There’s not really a principled way of doing this" but you were referring to cleaning up the data, not the analysis itself. My bad.

That said, you started with the following prior: Simpson ratings went down over time. Then you looked at 1 variable: The writing staff. You noticed: Aha, if we assign each writer a rating, then we see that later writers have lower ratings than writers that only worked on early episodes!

However that's a tautology. Of course the writers that worked on later episodes have lower scores than those that worked on earlier episodes. The fact the ratings went down over time was the prior we started with! This is the natural result of taking averages. The experimental setup was wrong from the beginning.

As for why I think you would want to do proper statistics, the reason is simple: I assume that people who publish these things are well intentioned, and they want to show off actual statistical correlations.