Comments on: Conference Season Review, Polynomial-Based Trendline Style

By: Penn State Conference Only Efficiency Numbers

Penn State Conference Only Efficiency Numbers — Fri, 09 May 2008 14:53:41 +0000

[...] Weblog did a rather complete game by game graphical analysis, including trendlines of all the Big Ten schools efficiency numbers for the past [...]

By: DMP

DMP — Tue, 25 Mar 2008 19:20:07 +0000

The trend line is nice to look at and *may* be informative. The analysis is still very qualitative, though, as kj pointed out, and as is pointed out in the FO site where kj got the idea from. He even pointed out how limited this is for MSU in particular owing to their inconsistency, even in the microcosm of the last 3 games of the season. 18 conference games is used here, 18 at most is used in the FO site owing to the short football season. That it doesn’t pass strict engineering/statistical/technical muster doesn’t mean it can’t be a nice topic of conversation.

By: kj

kj — Tue, 25 Mar 2008 14:45:27 +0000

A couple links for archival purposes:

1) Example of Football Outsiders’ use of this approach:

http://www.footballoutsiders.com/2008/01/05/ramblings/game-previews/5979/

2) Turns out I’m not the first blogger to think of doing this with basketball efficient stats:

http://phogblog.com/2007/02/26/efficiency-preview-kansas-at-oklahoma/

Rats.

By: kj

kj — Fri, 14 Mar 2008 13:08:01 +0000

Well, since I noted in the post itself that this was not guaranteed to be predictive, I guess we’re more or less on the same page.

By: Erik

Erik — Fri, 14 Mar 2008 04:33:31 +0000

Calculating a mean value (say, points per game allowed) does not have the same restrictions on data quantity as does calculating a trend. Even the most naive Taylor Series analysis suggests this will require N times the points for the same accuracy, N being the order of your polynomial fit. More rigorous analyses do not improve the situation. Just because you can fit an N-th order polynomial in Excel doesn’t mean you should.

Really, the math is not on your side. Don’t be offended, don’t take them down, just be aware that the figures you generated do not give valid small-scale predictions.

By: kj

kj — Fri, 14 Mar 2008 03:23:52 +0000

Whether a trend represents a statistically-significant phenomenon is one thing. But the trends are what they are (and I don’t think the trendlines are all that detailed; most of them have a single peak/valley). Interpret them as you will.

If the graphs are worthless, then I can only assume analysis of single-season college basketball data should be avoided entirely.

By: Erik

Erik — Fri, 14 Mar 2008 00:52:49 +0000

These plots are disingenuous. You can’t get trend information that detailed from so few data points. Consider Illinois’ offensive score, which swung from 120 to 60 on a game by game basis. Or Iowa’s defensive score, which was even worse. You’re telling me that a plot that indicates trends on the order of single digits per game is worth paying attention to?

They’re cool, and neat to look at, but they’re worth about half a grain of salt as far as this engineer is concerned.

By: kj

kj — Thu, 13 Mar 2008 16:36:16 +0000

I’ve thought about something like that–an adjusted off/def efficiency number for each game accounting for the opponent’s def/off and home/away status.

The advantage would be you could look at nonconference performance, too. The disadvantage would be . . . it would take a lot longer.

Maybe I’ll take a stab at this next season.

By: Dave

Dave — Thu, 13 Mar 2008 16:12:48 +0000

What if you graphed the deviation from the average for each opponent? In other words, what percent of their usual PPP did they get against you, and how many PPP did you score relative to their normal defensive efficiency? For example, if a given opponent averaged .90 PPP and you allowed them to score .99, then you would graph this at 110% for your defense for that game.

This would for the most part divorce trends in actual play from schedule difficulty trends, which currently muddy the water.

By: kj

kj — Wed, 12 Mar 2008 20:51:14 +0000

Hard to say how much to blame Dakich for the IU free fall. It sounds like the Indiana players were ready to give up on the season as soon as Sampson was ousted.

Your comments on Neitzel are on target. I’d add only this: Part of the reason Neitzel’s numbers have been so bad in the MSU losses is that he’s their last resort when the offense isn’t clicking. I think it was easier for him to get in rhythm last year when he was always the primary scoring option. It’s been harder this year when he’s expected to step up mainly against the better defenses in the league.