The college basketball prospectus is out and he explains the system a bit more in it. To wit:
"I decided to give the system a name. How about we call it Crazy Uncle? I don’t have a crazy uncle, but the output of this system fits the stereotype of someone making stunning predictions just to get some attention. Crazy Uncle provides an objective approach to setting expectations for each of the 344 teams signed up to play Division I basketball this season. Since it’s based on a bunch of equations, it’s insulated from the groupthink that tends to permeate poll voting in the instant-communication age. For example, last season Crazy Uncle’s top four was Duke, Kansas, Ohio State and Pitt. Those four ended up being the No. 1 seeds for the NCAA tournament five months later. It’s also true that none of the 66 AP poll voters had those four teams as their top four."
"This is not to say the Crazy Uncle is better than the voters. As one who prides himself on recognizing the impact of luck, it’s obvious to me that Crazy Uncle benefited from good fortune in matching his top four with the one-seeds. That’s not going to happen every year, or maybe ever again. Any preseason ranking figures to have a lot of uncertainty associated with it. Furthermore, this achievement only has merit if you believe the one-seeds really were the four best
teams in the nation, and you could make the case that Pitt was not. So I don’t want to give the impression that Crazy Uncle is always right, just that you should consider his opinion."
"Crazy Uncle is giving you something different, driven by numbers. Specifically, the ingredients are the last three seasons of efficiency data, information on how much of last season’s roster returns, and the destinations for the top 50 incoming recruits in the country. The algorithm is largely unchanged from last season with two key exceptions. Last year I only included the previous two seasons of efficiency data. It turns out that including a third season can improve the system slightly. In addition, Crazy Uncle is smarter about the use of recruiting information. In this area, less is more. Whereas last season I attempted to include recruiting information for every Division-I team, there is far too little predictive information in recruiting ratings outside the elite recruits for it to be worth the effort it takes to compile this information. Actually, I’ve found that limiting Crazy Uncle’s knowledge to the top 50 recruits (as determined by rscihoops.com) is for the best. For the recruits I do include, their impact is more intelligently determined than last season. Not only are available minutes considered when determining a recruit’s impact, but the team’s recent level of play is considered as well. Thus, even when few minutes are available on a relatively weak team, Crazy Uncle acknowledges that a four-star recruit or two can have a major impact."
"Note that when assessing roster continuity, Crazy Uncle is technically not using returning minutes. Minutes used are not created equal and calibration of the system to past seasons shows that players that carry a higher burden of the offense (as measured through percentage of possessions used) or are more efficient than their team’s offense tend to be more valuable. Players that do both are extremely important and the loss of such players (think Kemba Walker and Jimmer Fredette) is treated much more harshly that their minutes played would suggest. Crazy Uncle invokes similar principles to determine the impact of personnel changes on defense."
