well put. I'm not sure I completely understand the way his early season stats work. I was under the impression that before any games were played, he did his best to rank every team based on some kind of system. can you explain?
Sure. He calls it his "Crazy Uncle." He just started it last year, and had some pretty solid success (his preseason top 4 all went on to become #1 seeds, for example).
It's got 3 factors:
1) Prior years' efficiency
2) Returning players
3) Top 50 recruits
To expand a little bit:
1) He factors in the past 3 years of efficiency data. I would assume that the years are weighted so that the most recent year is the most meaningful. Last year he used the past 2 years, but has since found that using 3 is more accurate. This accounts for the fact that teams that are good tend to stay good (and vice versa).
2) This is the area that is most unknown, I think. He isn't exactly using returning minutes. I think he's essentially using returning usage rate combined in some way with respective efficiencies to estimate offensive efficiency. Estimating defense is trickier. We really don't have the stats to accurately capture an individual player's impact on defense. I'm guessing he uses some combination of returning stl%, blk%, dreb%, and overall returning minutes.
3) Pretty much self-explanatory. Teams with incoming recruits ranked in the RSCI top 50 get a boost (which I'm guessing is exponential...i.e landing #1 Anthony Davis vs. #11 PJ Hairston results in a much bigger boost than landing #31 Quinn Cook instead of #41 Ben McLemore). He also does factor in potential playing time (i.e Dorian Finney Smith is worth more to Va. Tech than similarly-rated Michael Gbinije is for Duke). Last year he attempted to evaluate incoming recruiting classes for every school, and he found that forecasting freshmen seasons for recruits ranked outside the top 50 is nearly impossible; its just way too unpredictable.
Intuitively, it makes a lot of sense. It finds the starting point (Part 1), and then accounts for losses (2) and gains (3).
To answer someone else's question, it is gradually phased out. Team strength is not constant; it fluctuates throughout the season. KenPom is designed to forecast how teams
will do, not how they have done. As a result, the more recent a game was, the higher its weight. In March, results from November are nearly meaningless. So, as each game goes by, Crazy Uncle's projection is factored in less and less. Last year, Pomeroy said that the preseason projection was completely zero-ed out around mid/late January. I imagine that'll be the case again this year.
Finally, for those keeping score at home, KenPom went 4-4 today in the Old Spice Classic.
