Walker Harrison

NFL teams make better 4th down decisions in preseason games that *don't even matter*. In other words: when coaches are freed from the objective of winning, they sometimes end up making more optimal winning decisions! Thread below on the data, methodology, and results ⬇️

all jokes aside, I did do some analysis on my text messages. since 2022, my dad has sent me 156 thumbs ups. no one else even has 20.

in honor of Father’s Day yesterday, I built dad-gpt.com, an ambitious addition to the LLM landscape that replicates conversations you might have with your dad. check it out:

did some quick, fun research on personal spending: I spend about $50 a week at the local bodega for simple groceries, snacks, etc. For years there was a good one just a single block away... and then another bodega opened up literally at the base of my apartment building. somehow I'd been gifted an even closer convenience store! I quickly felt my allegiances shifting... how quickly? after less than a month I completely abandon the old spot, according to my credit card data ⬇️

on the left is an actual topographic map of Brooklyn on the right is one I estimated using only @CitiBikeNYC ride data and the principle that it's harder to bike uphill pretty close! methodology and #rstats code in my linked post and the thread below 🧵 walker-harrison.com/posts/citibike…

The cool thing here is that this data isn't "supposed" to have a map hidden in it. But gravity is such a powerful, ubiquitous force that it seeps into CitiBike rider data without being explicitly defined or programmed.

Plugging these 700+ estimated elevations into a smoother creates a full contour map of Brooklyn. Its peaks and valleys are a little more gradual than the real map, which is what we'd expect from a noisier, coarser dataset. But the main features are recovered, which is pretty neat

@jakebruemmer @straightenerbot Excitement Index adds up all the incremental changes in win probability over the game. You can try to normalize it by game length but I just take the raw sum. So the translation would be 1.0 on the x-scale = 100% cumulative change in win prob

@WalkWearsCrocs @straightenerbot What is the translation here? Each unit +/- 50% win prob equals 1 unit on the horizontal axis?

Week 1 is in the books — time for a @straightenerbot recap 🧵 A somewhat underwhelming week: 16 games with an average Excitement Index of 3.7, good for 31st percentile among the 429 regular season weeks going back to 1999. x.com/straightenerbo…

more straightening next week 👍

Not only did teams have byes in Week 1 for those three years, but the 2001 Cardinals got further delayed by the 9/11 attacks and didn’t play their first game until September 23.