Justin Jefferson finished his rookie season with 1,400 receiving yards — the most for a rookie WR this century. His 64.2 receiving FPOE trail only Odell Beckham and A.J. Brown. His 274.2 PPR points trail only Beckham and Anquan Boldin.
Yet going into the 2020 season, Jefferson was just the ninth player picked in FFPC rookie drafts. Likewise, Brown was the 10th player picked in 2019 rookie drafts. Beckham was the eighth player off the board in 2014 MFL rookie drafts? Why do we so often miss on these breakout rookie WRs? By now most in the dynasty community are familiar with the key metrics: market shares of receiving yards, breakout age, and dominator rating are all metrics RotoViz has been relying since the site began, and all are now mainstream prospect evaluation tools. It’s not as though dynasty players are making decisions with bad data or outdated metrics, in other words.
Part of the problem is that Jefferson didn’t have the best numbers in the class. CeeDee Lamb ended with a higher final dominator rating. Jalen Reagor had a higher peak dominator rating. Henry Ruggs was faster. Multiple WRs were drafted earlier. All of this made it hard to project him as the top receiver in the class.
What he did have going for him is that he was the only WR prospect who checked every box — early breakout, early declare, and sufficient production both in terms of market shares and counting stats. But why is checking every box important, and why in particular does it seem more important than having the best projection?
Bounded Rationality and Optimization
The first Wrong Read I ever wrote was an attempt to build what I called a DFS Lineup Satisficer. The terminology is from work by Herbert Simon, who also came up with the idea of bounded rationality. His idea is that when we actually make decisions we operate with limited access to all relevant facts and limited time, and optimization is impossible under such conditions. Human rationality is bounded by various cognitive and environmental limitations. The best we can hope for when making decisions is a satisfactory solution. The process of finding a satisfactory, rather than optimal, solution is called satisficing.
The way we satisfice in practice is by making use of heuristics, or rules of thumb. We count a breakout as a 30% dominator rating because it’s more predictive than many alternatives. But why 30% instead of 29% or 32%? Jefferson was one of only two early-drafted WRs to breakout before the age of 20. But why 20, instead of 19.8? And why think of these metrics as thresholds at all, rather than as continuous variables that have some definable relationship with NFL production? How can these heuristics be useful if they’re not completely accurate?
Cognitive psychologist Gerd Gigerenzer describes the heuristics we use in making decisions as “fast and frugal.” That is to say, these sorts of heuristics probably won’t deliver optimal results, but they also don’t carry the same huge costs that an optimization process would. They enable us to make satisfactory decisions with limited time and resources. And they enable us to come to satisfactory solutions with less information than is required for optimization.
Categorization by Elimination
A useful heuristic Gigerenzer and his colleagues discuss is known as categorization by elimination. While categorization is a task that machine-learning algorithms can do well — classification and regression trees are a good example — they require considerable computational resources and preparation to ensure they run as intended and return accurate and reliable results. The algorithms are usually complex, the results they return are sometimes difficult to interpret, and they often tend toward overfitting. A simpler method for categorization that works almost as well as the best machine learning algorithms is what’s known as an elimination model — you simply eliminate available options one-by-one based on a few cues until you have a satisfactory result.
The main goal in evaluating prospects for fantasy is to determine whether to target them in drafts or not. Using an elimination model is an easy way to build a target list. Here’s how the exercise looks in action with Jefferson: We know that declaring early for the NFL draft is one of the best predictors of NFL success. Jefferson declared early, so he remains a candidate for our target list. We also know that early breakouts outperform late breakouts. Jefferson broke out in college before he turned 20, so he remains on our list. In fact, among the top 2020 WR prospects, he’s one of only two players still on our list, the other being Reagor. We also know that a baseline of raw production — specifically at least 0.45 receiving touchdowns per game in a prospect’s final college season — is a weirdly strong signal of NFL success. Jefferson meets this threshold, and is now the only prospect still on our target list.