by Ken Pomeroy on Saturday, October 26, 2013
Pre-season ratings have been posted for the upcoming season. When I first started doing these before the 2011 season, I thought I was pretty awesome. It was kind of a big deal to get every team’s lineup data, mix in some limited recruiting info, and produce a rating that wasn’t laughably horrible. But then Hanner came along with his lineup-based approach and TeamRankings did something that is probably fairly sophisticated, and my preseason ratings became the simplest algorithm possible without being a complete joke.
The system is largely the same as in recent seasons. It independently predicts a team’s adjusted offensive and defensive efficiency. As a reminder, it uses information split into two categories:
- Base level of the program. This takes into account the last five seasons of data for the same unit (offense for predicting offense) and the last season for the opposite unit (defense for predicting offense). It also includes data for how much money the program has spent on men’s basketball for the last three seasons. The bulk of this component is determined by the most recent season’s performance of the unit.
You can make a decent predictive system just by knowing what is normal for a program. If we were predicting the Big 12 standings in 2025 (assuming the conference exists), it would bereasonably safe to say that Kansas will have a winning record and TCU will have a losing record. We can say that with some confidence even though some of the players on those rosters haven’t picked up a basketball yet.
- Personnel. This component handles who’s coming back from last season’s team and which impact recruits are being added to the roster. More impact is given to returning players from earlier classes. And minutes played by those with a high-efficiency/high-usage profile are particularly important. Recruits in the RSCI top 100 have some influence here as well, although most of the influence is in the top 50.
The goal here is really to get each conference’s pecking order correct and to predict end-of-season ratings. To that extent, if a player is expected to be available by late-January or so, he’s included in the personnel calculations. This applies to Louisville’s Chane Behanan and Florida’s Chris Walker, while Georgetown’s Greg Whittington is not included although he may well see action later in the season.
You can find additional discussion in last season’s piece.
Now let’s get to the question a lot of people will be asking.
Why is [state your favorite team] rated lower than it should be?
It’s because one or both of the components is missing something. Perhaps recent seasons are not representative of your team’s normal level. The personnel component doesn’t consider transfers or recruits outside the top 100. It does have knowledge of players that played two seasons ago but missed last season, but that is a small influence. So if your team has players that the personnel component can’t see (transfers, junior college players, and non top-100 recruits for mid-majors), then it’s possible your team is underrated. Keep in mind, though, that the first component handles some of this. It effectively sets a “replacement level” for new players on the roster that aren’t accounted for in the personnel component.
The system doesn’t think as highly of freshmen as AP voters will and it likes good teams that return a lot of players. Hence Oklahoma State, Iowa, UConn, Creighton, and Stanford are ranked higher than the humans and Kentucky, Kansas, and Arizona are ranked lower. (Hey, the Fab Five were ranked #20 in the preseason by the humans, so leave me alone.) Andrew Wiggins and Julius Randle are not your typical first- and second-ranked recruits, so perhaps I could have made some subjective adjustments here, but I chose not to.
Last season, the system managed to hold its own against others, with a mean absolute error of about 2.14 on predicting conference wins. It had some good calls and some bad ones, some of which were discussed in the linked piece. Refer to your local message board archives for additional details.
It’s worth mentioning that at the end of the season, any conference’s standings will not look like what is currently predicted. Meaning, it’s obviously going to take more than 12 wins to win the Big East or 13 to win the Big Ten. But the top teams in those conferences are similar enough that a reasonable expectation for the win total of each of those teams can not be very high.