When I started this substack, I thought it’d be cool to take you through some of the more interesting modeling challenges I’ve gone through recently. Like, how rt.live works, or a fancy new media mix model, or something something.
Turns out, though, that I don’t have that many models to write about. Most of the work I do is grinding out small improvements on existing models. For example, if you work in Stan, you may realize that sometimes your models run quickly, and sometimes slowly. For Recast, these variations mean differences on the order of days of runtime. Diagnosing what was causing this led to a rabbithole of investigations into now-defunct Google Groups threads and figuring out how to rescale our parameters to be unit-scaled in the posterior distribution. This may sound like good reading to some of you, but… I have my doubts about its broad appeal.
Instead, I guess I’ll write about what it’s like to run a small business that does consulting work informed by statistics, and to start a SaaS business that is driven by a statistical model.
What would you like to hear about? What would you like to know?
Nice! I'm hyped to read whatever you write, broski.