UTK Special 2/7/23
Team Health Report Notes
Every writer knows that doing something team by team sounds good. It’s 30 things you know you’ll write, but man, even if you’ve done it before, you hit a spot at about halfway through where you say “I’m never doing this again.” However, you often do it again.
I’m not. At least not here.
Later this week, I’ll release the spreadsheet with all the ratings and comments for key players. Instead of 30 articles that will make me hate myself by the end, I’m doing one big spreadsheet that is much more usable.
Today, I’m going to just go over the key changes and general notes for what I see in the THR’s. There are some big under-the-hood things that bear some explanation and some trends I’m honestly not sure about. While this has never been more than a black box system and I’m not opening the hood* to you either.
A reminder to you that The System is looking for risk and defines that as a stay on the IL. One of the issues is the varying lengths in the past samples and training set. I ran it with some corrections but decided that it didn’t make much difference. It’s also looking for risk in terms of large numbers. While the risk of Player A hitting the IL as a yellow is not the same as Player B who is also a yellow, in large numbers, yellows will hit the IL the same. I use terms like “low yellow” and “high green” for scores that hit at the bottom or top of the band. 0 is no risk and would be green. 100 would be high risk and glowing red.
Why use bands instead of single numbers? Single numbers aren’t as instructive. A one percent risk difference is almost nothing, but looks like something. It would also make the system really, really easy to reverse engineer and once again, the hood stays closed. Think of this as very close to how insurers look at risk and you’ll have it right, since this is an actuarial instrument at heart.
I’ll have the full spreadsheet soon, but I want to be clear about one thing: while I believe The System is the best it’s ever been, I don’t think it’s perfect. It knows nothing about the game, planned usage, team context, medical staffs, or much else. It’s an actuarial table with a relatively small data set and some very rudimentary machine learning built in. I disagree with it pretty regularly, but it also doesn’t have blind spots or biases like I do.
It’s a tool, like any other, and like any other, you should be using it as a piece of your information diet, not the full thing. Pairing it with your favorite projections - I use STEAMER and Clay Davenport’s EqA - is the best way to use it. That, or just using it for entertainment!
*There’s another metaphor that’s dated. In ten years, frunks will be far more ubiquitous than engines. “Running on empty” might still work in our brave new world, but a lot of phrases are going to change more slowly than the people.