Rofl Hollinger Ratings

[B-Roy]

From the Bulls game we went from 6th seed to 3rd seed:

http://sports.espn.go.com/nba/hollinger/playoffodds

And in the power rankings moved from 16th to 5th:

http://sports.espn.go.com/nba/hollinger/powerranking

Who knew one game could do this much?

 

[Nikolokolus]

Hollinger heavily weights his models to account for point differential, so big wins and big losses have a large effect scores ... not to say that big wins or losses aren't important, but I think he weights them too heavily.

 

[Minstrel]

Hollinger heavily weights his models to account for point differential, so big wins and big losses have a large effect scores ... not to say that big wins or losses aren't important, but I think he weights them too heavily.

I think weighing point differential so heavily is the right approach. I think it's the best measure of true team quality. Big wins and losses have disproportionate effect early in the season, due to the standard sample size problems. For the first month or two, one should expect a lot of volatility in the standings. Over the course of the season, the effects of any single blowout will be pretty minimal.

 

[Foulzilla]

It's because the sample size is still so small. Any statistical modeling system is going to be fairly inaccurate until it has a large amount of data points. I suspect his ratings will still be fairly questionable until around all-star break, at which point I suspect they'll make more sense.

 

[Wizard Mentor]

I think weighing point differential so heavily is the right approach. I think it's the best measure of true team quality. Big wins and losses have disproportionate effect early in the season, due to the standard sample size problems. For the first month or two, one should expect a lot of volatility in the standings. Over the course of the season, the effects of any single blowout will be pretty minimal.

I think depending on point differential so heavily is OK, but he obviously does it wrong. He should clamp the
point differential so that a 50-point win is the same as a 20-point win.

 

[Crimson the Cat]

He should clamp the point differential so that a 50-point win is the same as a 20-point win.

But a 50 point blow out is more impressive than 20 point blow out.

 

[Minstrel]

I think depending on point differential so heavily is OK, but he obviously does it wrong. He should clamp the
point differential so that a 50-point win is the same as a 20-point win.

Why? A 50 point win has the same effect on the standings as a 20 point win (or a 1 point win) but a 50 point win speaks to greater dominance than a 20 point win. Even if your issue is that everything after a 20 point lead is "garbage time" (which it isn't, the Blazers built their lead to 30 and even 40, I believe, while the Bulls still had starters in), a team's reserves outplaying the other team's reserves still provides useful information.

It's not perfect, perhaps, but I think the bigger the sample you have the better, and that includes all of the game, not just the part up to the 20 point lead, essentially.

 

[Wizard Mentor]

But a 50 point blow out is more impressive than 20 point blow out.

No, it isn't. The game has turned into a blowout, and the players aren't playing the same way they usually do.
It's barely even the same game. Players who wouldn't normally even play for both teams are playing. That's
bad data when used for predictive purposes.

 

[BrianFromWA]

I think the Pythagorean method of point differential being a strong predictor (brought from baseball) is much more suited to baseball than basketball, and has been one of my heavier Hollinger criticisms in the past. In baseball, there's not a lot of matchup differences that are unaccounted for, everyone is basically on the same "6 games in 7 nights" schedule, so run differential ends up being a strong predictor of wins. But IMHO I think basketball has a lot more fluidity and doesn't statistically account for a lot of things that happen during the game (such as refereeing, fatigue due to back-to-back nights, more "day-to-day" injuries, etc.)

So while I think prediction models are fun, I'll continue to express my issues with Hollinger's. Until he tweaks it.

 

[Minstrel]

In baseball, there's not a lot of matchup differences that are unaccounted for, everyone is basically on the same "6 games in 7 nights" schedule, so run differential ends up being a strong predictor of wins. But IMHO I think basketball has a lot more fluidity and doesn't statistically account for a lot of things that happen during the game (such as refereeing, fatigue due to back-to-back nights, more "day-to-day" injuries, etc.)

Why would that make point differential less correlated with team qualityin basketball? All NBA teams have to deal with basically the same refereeing, fatigue issues and attrition over the course of the season.

 

[Crimson the Cat]

No, it isn't. The game has turned into a blowout, and the players aren't playing the same way they usually do.
It's barely even the same game. Players who wouldn't normally even play for both teams are playing. That's
bad data when used for predictive purposes.

Well then I guess Portland didn't get the memo. Because when up 20 Portland didn't let up and allow Chicago to go on any runs. Chicago sure looked like they were trying. Portland was just denying them at every turn. Instead of just maintaining a 20 point lead, they applied even more defensive pressure and increased their lead. This killer instinct and deep bench must be accounted for.

This team performance was greater than a 20 point win and should be quantified appropriately.

 

[Wizard Mentor]

Well then I guess Portland didn't get the memo. Because when up 20 Portland didn't let up and allow Chicago to go on any runs. Chicago sure looked like they were trying. Portland was just denying them at every turn. Instead of just maintaining a 20 point lead, they applied even more defensive pressure and increased their lead. This killer instinct and deep bench must be accounted for.

This team performance was greater than a 20 point win and should be quantified appropriately.

Actually, I watched the game. They stopped playing any semblance of team ball and played 1-on-5 ball.
I haven't seen the bulls play this year before that game, but I assume it's not normal.

 

[Crimson the Cat]

Actually, I watched the game. They stopped playing any semblance of team ball and played 1-on-5 ball.
I haven't seen the bulls play this year before that game, but I assume it's not normal.

I watched it too. I was there. Portland's defense never allowed them to play team ball.

 

[elcap15]

the sample size is still small, so there will be big swings.

Also, the Blazers had a negative margin of victory, even though they had a winning record. So a big win really helped to correct that anomoly in the system.