Re: theory 1, as the population of players gets larger, the familiarity with word games and puzzles likely goes down on average. Or put another way, earlier adopters are likely people who more enjoy and are good at Wordle-like games vs. the general public?

That's a really good point. It does seem likely that the first people to play a new puzzle in high volume are naturally more inclined to puzzles. It might have also spread among puzzle-enthusiast circles on twitter and other sites.

Will add a line about that!

oh! to test this: you could look at cohorts based on first tweet. hopefully if the same people are tweeting their scores, you can see if they, individually or based on when they first started playing Wordle, are all getting worse.

Yea that’s definitely testable! Decided not to do any user-level analysis for this in order to keep it shorter, but would like to ask questions about individual behavior over time, consistent player vs inconsistent player, etc in a future piece.

Awesome analysis! One additional thing that came to my mind thinking about theory 1: I guess with it's increasing popularity, the number of non native speakers increases as well (like me 😊). This most likely has an influence on the "word knowledge" property! Regarding the regression analysis, you might be interested in this experimental R-style "mutliple linear regression" framework https://observablehq.com/d/c3ef6d4e928e9ece

That's an interesting theory! In Appendix A you can breakdown each date by tweet language, and the number of non-english tweets definitely rose over time, albeit from a small baseline. Perhaps that did have an effect. Could be interesting to look at breakdown of scores by language, and see if non english tweets had worse scores. And that notebook is really cool, and exactly the sort of thing I had been looking for before I decided on jstats. Might use it for a followup I've been thinking about.