Why I Don’t Blindly Follow Numbers; Subtitle: Checking in on the Aaron Dobson Experiment
I have a google alert for Aaron Dobson’s name and I would be lying if I said that I don’t do twitter searches for his name as well. My motivation is ego (who doesn’t like to be right?). When I pulled the trigger on Dobson in the first round of a rookie draft, I didn’t just go against my numbers, I went against my numbers and probably market sentiment as well. It was probably a reach. But it was also a reach that involved logic. To the extent that I think it’s valuable to discuss applying a mix of logic and numbers, I thought I would write about Dobson one more time even though I’ve given him more coverage than I ever thought I would. First, let’s check in on the Aaron Dobson experiment:
Rookie WR Aaron Dobson made the play of the day, adjusting to an underthrown deep sideline pass and catching it over two defenders
— Ben Volin (@BenVolin) May 29, 2013
There’s nothing like giving the “Scoreboard bro” move after a few days of OTAs right? But that’s what I’m left with as I have no idea whether Dobson will be any good. Moving on.
I think there’s a misconception about users of metrics that we put the numbers into our spreadsheets, hit calculate, and then go play World of Warcraft in our parents’ basement while we wait for the almighty spreadsheet to give us an answer. I don’t do that at all and most of the stats nerds that I know don’t do that. All I’m trying to get out of the numbers is an edge, but I also know that they won’t always give me an edge because there are a lot of things they don’t know anything about.
Last year I tried to create a text mining algorithm that would mine through news blurbs and assign a percent likelihood of playing for any player on the injury report. It actually performed ok given that it was based on some crude code that I pasted together from various sources. But I ditched that exercise ultimately because while the algorithm could get most “will play/won’t play” decisions right, it couldn’t outperform my brain. When I compared the results of the predictions versus my own assessments as to who would play, I could see that the algo was getting some easy decisions wrong. It also couldn’t tell the difference between “fully recovered from an ACL tear” and “hasn’t fully recovered from an ACL tear”. The human brain is incredibly efficient at being able to make assessments like that, so I threw out the algo until I could do it better. I didn’t trot out the results and say “see, Marshawn Lynch has a 63% chance of playing this week” because that would have been counterproductive.
Ultimately what we do is try to make good decisions and we lean on the numbers when we feel like the numbers will help us make good decisions. The psychologist Daniel Kahneman has found that human decision making improves even if we’re just presented the results of an algorithmic process but we still have the ability to make the final judgment on our own. That’s an important idea to keep in mind as you read the various articles on the site. We might write about Aldrick Robinson a bunch of times because we believe the market is assigning him a 0% chance of succeeding and we think his chances are at least greater than that by an amount that warrants keeping an eye on him. We’re taking the results of our algorithms (whether they were formally or informally constructed) and we’re then using our human brains to make a judgment. We’re also assuming that you’re doing the same thing.
Aaron Dobson didn’t score the highest on my model out of the remaining receivers when I drafted him. Last year I took Doug Martin in the first round of a keeper draft even though he didn’t have the best speed score. Arian Foster has an awful speed score and yet I’ve owned him in fantasy leagues. I’ll probably draft Mark Ingram this year and he’s freaking awful by any of the measures that I care about. It’s also the case that sometimes you get conflicting signals. I owned Dez Bryant last year even though my sim score projections didn’t show him as being undervalued compared to his draft spot. I just wanted a WR with a 12 TD season in his range of outcomes and his college record told me he probably had one in him.
My point is that as you read this site, always keep it in the front of your mind that the goal is to make a good decision (and to be clear, the jury is still out on some of the decisions I discuss above…Dobson may end up being a bad one, as could Mark Ingram). The goal is not to rigidly apply an algorithm for its own sake.