Espn rankings

[MBunge]

It's just an indicator of how well the teams have played overall in the regular season.

If that's all it indicates, then what good is it?  Did anybody need Hollinger to convince them that Orlando and Cleveland were really good teams during the regular season?

Mike

[Fafnir]

My objection is people waving around point differential as though its the only thing that matters or the only thing that demonstrates how good a team is.

Mike

Point differential is a better predictor of future success than record. This has been shown to be true when it comes to predicting both regular season future success and also post-season success. Its not "waving it around" to cite the evidence of 30+ years of NBA history. (a recent example is the Mavs flaming out in 06-07, point differential had that team as being weaker than the Spurs that year, though no one predicted what the Warriors did)

We'll just have to disagree about this in the end.

[Celtics4ever]

Ratings are better now, than they were the last week.

[MBunge]

My objection is people waving around point differential as though its the only thing that matters or the only thing that demonstrates how good a team is.

Mike

Point differential is a better predictor of future success than record.

What do you mean by future success?  Orlando and Phoenix had the biggest point differentials in last season's playoffs.  Do they have rings?  Orlando, Cleveland, Utah, Phoenix and San Antonio had the biggest point differentials last regular season.  Did they make the Finals?

Mike

[Fafnir]

My objection is people waving around point differential as though its the only thing that matters or the only thing that demonstrates how good a team is.

Mike

Point differential is a better predictor of future success than record.

What do you mean by future success?  Orlando and Phoenix had the biggest point differentials in last season's playoffs.  Do they have rings?  Orlando, Cleveland, Utah, Phoenix and San Antonio had the biggest point differentials last regular season.  Did they make the Finals?

Mike

When you go back and look at the history of the league without cherry picking data, yes its a better predictor than just record.

The Celtics last year were outliers, as were the 05 Heat, and the Lakers (Shaq/Kobe team) the year they had a ton of injuries. Of course if you're using record to predict things, those teams wouldn't be predicted to win as much as they did in the playoffs either.

Furthermore, no system or person gets everything right. Its all a matter of guesses and/or probabilities. A clearly inferior team can still pull of an upset, it happens.

[celtics2]

the ESPN ranking system is flawed for sure. How can they put the team that beat them twice lower? It's like they have a steak in it along with the Heat Watch they have. Yahoo Sports has  LA 1st Celtics 2nd. More understandable.

[Fan from VT]

interesting read here at least:

http://www.basketball-reference.com/blog/?p=8159

Essentially, better future success is predicted, in order, by:

blowing out good teams
blowing out bad teams
eking out close games against good teams
eking out close games against bad teams

[Fafnir]

interesting read here at least:

http://www.basketball-reference.com/blog/?p=8159

Essentially, better future success is predicted, in order, by:

blowing out good teams
blowing out bad teams
eking out close games against good teams
eking out close games against bad teams

*cough*

BUMP!

I just read an article at one of the most useful basketball sites out there, Basektball-Reference. Neil Paine addresses one of the very topics we've debated in this thread:

http://www.basketball-reference.com/blog/?p=8159

Basically as a predictor of post-season success (in the CFs/Finals) the number of 10+ point wins is more important than your record in close games. (whether or not you're playing a good or bad team)

*cough*

[Fan from VT]

Alright, I'm a little late, but I said I'd bring it back.

Anyway, on the 1/10/11 mark:

The question was,
"Starting on 11/15/10, over the next month (or more like 2, sorry), which will better predict team record over said span (11/15/10 to 1/10/11): win-loss record on 11/15/10 or Hollinger's rankings on 11/15/10?"

I did this very simply as a rough estimate. First, I ranked the teams 1 through 30 based on win percentage. Then I made a second list 1 through 30 based on Hollinger's rankings.

Today, I calculated each team's win percentage from 11-15-10 to 1-10-11, and ranked those in order 1-30. I then subtracted the 1/10/11 ranking from the 11/15/10 ranking for both the standings rank and the Hollinger rank, corrected for absolute value, and averaged them all to find the average error each ranking was off in terms of predicting how good Team X would be in the subsequent ~two month period. In other words, if the standings predicted a team was the 10th best in the league, but that team had the 15th best record from 11/15 to 1/10, the "prediction" was off by 5 spots.

Here's what I found:
In terms of predicting "who would be the best team from 11/15/10 to 1/10/11," "who would be the worst team," etc. ranked 1 through 30, using the actual standings as a predictor yielded an error of 6.07 ranking spots off from actual. Hollinger's predictions were off by an average of 5.13 rank spots.


As a bonus, I also checked to see if actual standings or Hollinger's rankings would be more accurate in predicting actual overall standings at the 1/10/11 mark, not just the period of games played between 11/15 and 1/10. By the same method above, I found that the actual rankings on 1/10 were different from the win-loss record on 11/15 by an average of 4.3 ranks, while Hollinger's 11/15 rankings were only off from the 1/10/11 standings rankings by 4.1 spots.

[Fafnir]

Interesting stuff, you could also use predicted win percentage rather than just ranking. That way you could differentiate more between the really bad teams and the large clump in the middle.

TP for the follow up.

[pearljammer10]

It would probably be more interesting if I understood what you were talking about. Unfortunately I dont. however, good job taking the time to put out a thorough analysis. For that ill at least throw you a TP.

[Brendan]

If I may summarize: You used Hollinger's rankings and Win % to see who would predict future success (as defined as wins from 11/2010 through now) better. Advantage: Hollinger.

Basically you're small sample confirms what "stat heads" were saying: how you win >> if you win in future predictions.

Win % = if you win model
Hollinger = how you win model

While Hollinger's model may not be perfect, I'm unsurprised that it beats the simplistic Win % model.

TP

[fairweatherfan]

TP from me as well - good analysis, and a good example of how these advanced models can provide better predictions than the simpler statistics (like Win %) that many opponents of this kind of approach favor.  

It's nice to see someone putting in the effort to put some substance behind the arguments that break out from time to time (especially since I'm too lazy to do it myself  )

[Fan from VT]

Interesting stuff, you could also use predicted win percentage rather than just ranking. That way you could differentiate more between the really bad teams and the large clump in the middle.

TP for the follow up.

That will be for the next step. On 11/15 it was too early for Hollinger to release his expected win % in his playoff odds thingy. Now that is out, so in 2 months we can check both this ranking method and specific win-loss percent predictions.

[mgent]

I like what you did.  I don't think Hollinger is a good way to rank teams but W/L is worse.  My problem is not that blowing out bads teams is better than close wins over good teams, but what about teams that blow out bad teams and lose more games to good teams than they win?

AKA LeBron teams.