mrjames
Professor
Posts: 6062
Loc: Montclair, NJ
Reg: 11-21-04
|
08-16-18 10:59 AM - Post#260358
In response to TheLine
Hi all -
Figured this was as good an entry point as any to step away from my summer hiatus (sort of - more on that at the end).
Rather than try to answer all of the different questions/topics of debate one by one, I just want to share some thoughts about preseason rankings.
1) ACTUAL PERFORMANCE: Preseason rankings have a ton of predictive ability. Even those that are based on subjective means like the AP poll ( Preseason AP Poll Performance ). But those that are based on objective rankings of all 1-351 teams are even more on point ( Objective Preseason Model System Performance).
Of course, on a team-by-team basis, there can be huge misses, but if you look at it from a macro level, it's pretty impressive what a limited amount of information can produce when it comes to an accurate prediction.
2) WHY DO THEY EXIST? - Aside from fodder for rabid fans to digest, they actually serve an important ratings purpose. It used to be that most ratings systems "floated" early in the year - assuming all teams were equal to start - and thus took a long time to settle. From a Bayesian perspective, though, ignoring our priors (we kinda know that UNC is gonna be better than UNH) leads to unnecessary uncertainty early on. So, by establishing an objective preseason benchmark (and one that's pretty darn accurate to boot), we can more strongly credit teams that register close losses to great teams early and more strongly debit those teams that register narrow victories over the dregs of D-I.
3) WHAT DO THEY DO? - As has been mentioned here, they roll forward player performance, add in new player expectations, sprinkle in some team trends (plus some minor secret sauces by system) to spit out an objective offensive rating. Defensive rating tends to be more based on team style, continuity and general historical trend and is the less accurate of the two. It's important to note that the objective preseason rankings (Hanner, KenPom, Bart Torvik, etc.) do NOT involve the producer getting involved in any subjective way. So, it's not "Torvik thinks" so much as it is "Torvik's empirical model says."
4) PRESEASON RATINGS AREN'T STATIC - At this point last year, Harvard was looking at a full year of Bryce Aiken and Tommy McCarthy as the PG depth chart, Yale was going to be led by Makai Mason and Jordan Bruner, Dartmouth was going to have Evan Boudreaux. And so on...
Some of these systems will account for further player movement (Bart did along the way), some won't. We never survive October and November without some MASSIVE losses as a league, so these will assuredly change before tip off in November.
5) BELL CURVES ARE THICK AROUND THE MIDDLE - When I used to do detailed projections, my targeted miss was anything better than +/- .1000 in Pythag. At the top and bottom of the curve, that can mean the difference in 50ish ranking spots. Around the middle, that rises to about 80ish. Given that we often have a lot of teams around the middle, it takes a BIG rankings difference to imply a significant difference between two teams.
Hopefully, the above was helpful. I wasn't terribly surprised by Bart's initial cut.
Harvard lost nobody off a team that had the best conference adj efficiency margin and highest slope to its performance start to finish. Add to that players returning from injury and a strong recruiting class, and the modest increase over its stretch run performance seems right.
Leaving aside subjective expectations, Penn makes sense if you understand the model. When the model looks at Foreman and Wood, it sees about 41% of Penn's performance above replacement offensively going out the door. Bart's estimates of the ability of Penn to replace that production are quite strong, actually, but sum total the offense and defense get trivially worse.
Princeton and Yale are the same stories - strong, returning answers offensively drive the expected ORAT up, model generally favors long-run trends defensively so some improvement (regression to mean) boost them each there.
Not going to go through the rest, except to say that I expected Brown to be a little more ahead in 5th of 6th-8th.
Now to return to my comment about my hiatus from the top. My wife and I are about to have our third child, which is incredibly exciting. Between #dayjob and #dadlife, something has to give. Covering the league the way I'd want to cover it (checking in with folks around the league, watching a bunch of games from across the league, booting up the code), just doesn't seem like it will be possible - at least in the short run. So I'm going to power down the Twitter feed and my active participation on these boards for this season. I'll still be checking in from time to time, and I hope to get back to covering the league in the future. But I need to be realistic about my time, and this seems like it has to be on my cut list.
I'll still be watching, and if we can survive the usual devastation in October and November, it could be a HUGE year for our league!
-Mike
|
