How does Hollinger still have a job?

[BballTim]

If you don't care about "stats" enough to understand them, its best not to mock them.

I understand stats quite well actually.  Which is why I dismiss them.

Subjective observation always will trump stats because there just aren't any stats that can account for the intangibles nessecary to be successful in sports.  Until someone can develope a way to measure a man's heart, determinantion, poise under pressure, etc., it will always be this way.

If you understood stats you'd know that statistical prediction consistently beats subjective prediction, in basketball and elsewhere.

  They seem to be discussing the understanding of specific basketball stats, not the theory of stats in general. You don't need to know about statistical prediction accuracy to understand PER or the like. Statistical prediction consistently beating subjective prediction would obviously be dependent on the accuracy of the prediction model and consistency of the data. I might be able to come up with a prediction model for the nba based on the height or age of the players but it wouldn't necessarily be more accurate than informed subjective prediction.

True though in this case if we're talking about score differential, how this thread started, I think it has been proven to be more accurate than subjective predictions on the aggregate.

Not really looking to argue the point anymore, just want to throw this out there.

Did not our very own beloved Celtics prove the scoring margin theory to be wrong last season?

Statistics deal with probabilities not certainties, what the Celtics did was unexpected by the majority of NBA observers. (including me!)

That doesn't mean you throw away 30+ years of data.

  What the Celtics did last year wasn't randomly outperform their expectations. There were specific reasons that the predictions were wrong and there were many people that would have had no problem pointing this out at any point before or during the playoffs. In other words, the "post mortems" of the many matched the predictions of a number of people (myself included).

[BballTim]

If you don't care about "stats" enough to understand them, its best not to mock them.

I understand stats quite well actually.  Which is why I dismiss them.

Subjective observation always will trump stats because there just aren't any stats that can account for the intangibles nessecary to be successful in sports.  Until someone can develope a way to measure a man's heart, determinantion, poise under pressure, etc., it will always be this way.

If you understood stats you'd know that statistical prediction consistently beats subjective prediction, in basketball and elsewhere.

  They seem to be discussing the understanding of specific basketball stats, not the theory of stats in general. You don't need to know about statistical prediction accuracy to understand PER or the like. Statistical prediction consistently beating subjective prediction would obviously be dependent on the accuracy of the prediction model and consistency of the data. I might be able to come up with a prediction model for the nba based on the height or age of the players but it wouldn't necessarily be more accurate than informed subjective prediction.

True though in this case if we're talking about score differential, how this thread started, I think it has been proven to be more accurate than subjective predictions on the aggregate.

  In general yes, but at the very least it needs to be weighted for strength (and probably current performance) of opponents. I think there are enough variables that aren't accounted for that a person, if they spent the time and effort, could beat it for accuracy.

I think that people are too subject to their own biases. I suppose there is some one out there who does that. But I have my doubts.

  Not me, mind you, but I would bet that there are a number of oddsmakers or gamblers that could outperform Hollinger's models simply because they make the effort to factor in things that Hollinger doesn't account for.

[Fafnir]

 What the Celtics did last year wasn't randomly outperform their expectations. There were specific reasons that the predictions were wrong and there were many people that would have had no problem pointing this out at any point before or during the playoffs. In other words, the "post mortems" of the many matched the predictions of a number of people (myself included).

Those specific reasons still made them an outlier though didn't it? There are always reasons for why events happen, but that doesn't make them the most likely outcome.

[KGs Knee]

In the end, I guess this is the beauty of sports (for me personally anyways).

No matter how many stats people want to come up with to try and determine or predict outcomes in sports, the games still have to be played, and the predicted outcome is not always the actuall outcome.  It sure would be boring otherwsie.

[the_Bird]

If you don't care about "stats" enough to understand them, its best not to mock them.

I understand stats quite well actually.  Which is why I dismiss them.

Subjective observation always will trump stats because there just aren't any stats that can account for the intangibles nessecary to be successful in sports.  Until someone can develope a way to measure a man's heart, determinantion, poise under pressure, etc., it will always be this way.

If you understood stats you'd know that statistical prediction consistently beats subjective prediction, in basketball and elsewhere.

  They seem to be discussing the understanding of specific basketball stats, not the theory of stats in general. You don't need to know about statistical prediction accuracy to understand PER or the like. Statistical prediction consistently beating subjective prediction would obviously be dependent on the accuracy of the prediction model and consistency of the data. I might be able to come up with a prediction model for the nba based on the height or age of the players but it wouldn't necessarily be more accurate than informed subjective prediction.

True though in this case if we're talking about score differential, how this thread started, I think it has been proven to be more accurate than subjective predictions on the aggregate.

Not really looking to argue the point anymore, just want to throw this out there.

Did not our very own beloved Celtics prove the scoring margin theory to be wrong last season?

Statistics deal with probabilities not certainties, what the Celtics did was unexpected by the majority of NBA observers. (including me!)

That doesn't mean you throw away 30+ years of data.

  What the Celtics did last year wasn't randomly outperform their expectations. There were specific reasons that the predictions were wrong and there were many people that would have had no problem pointing this out at any point before or during the playoffs. In other words, the "post mortems" of the many matched the predictions of a number of people (myself included).

The models are inherently going to fail when the team being studies isn't really *trying* their hardest to win.  We all know that the C's basically coasted into the playoffs, with the goal being to be healthy rather than to have the best possible record/playoff seeding.  That's going to skew every single statistic of theirs, and I'm not sure how any model is going to adjust for that.

It's like trying to predict regular-season performance based on preseason games.  I remember a couple of years ago, a certain northerly NFL team went 4-0 in preseason.  In situations where one team isn't really giving it their all to win, the stats are going to be more-or-less meaningless.

[kozlodoev]

In the end, I guess this is the beauty of sports (for me personally anyways).

No matter how many stats people want to come up with to try and determine or predict outcomes in sports, the games still have to be played, and the predicted outcome is not always the actuall outcome.  It sure would be boring otherwsie.

This is more or less true just about anything. While some events are more probable than other, there is a certain degree of randomness involved in all outcomes.

Based on this thread, it seems that US colleges do a thoroughly poor job in teaching kids how and why we should study said  randomness.

[mgent]

His system looks at scoring differential not wins and losses. The Heat have beaten up the bad teams they've played by a large margin.

Does anyone else actually agree with this system?  In my opinion scoring differential doesn't mean too much in basketball.

  I think in general scoring differential is more important than records in predictions, but obviously there are more factors involved.

I agree.  But I think about all the times that a game is much different than the score says it was.  Garbage time has got to skew the results.

[the_Bird]

In the end, I guess this is the beauty of sports (for me personally anyways).

No matter how many stats people want to come up with to try and determine or predict outcomes in sports, the games still have to be played, and the predicted outcome is not always the actuall outcome.  It sure would be boring otherwsie.

This is more or less true just about anything. While some events are more probable than other, there is a certain degree of randomness involved in all outcomes.

Based on this thread, it seems that US colleges do a thoroughly poor job in teaching kids how and why we should study said  randomness.

I don't know.  Do you really think the number of people responding to this thread represents a statistically significant sample size?

[Boris Badenov]

His system looks at scoring differential not wins and losses. The Heat have beaten up the bad teams they've played by a large margin.

Does anyone else actually agree with this system?  In my opinion scoring differential doesn't mean too much in basketball.

  I think in general scoring differential is more important than records in predictions, but obviously there are more factors involved.

I agree.  But I think about all the times that a game is much different than the score says it was.  Garbage time has got to skew the results.

What's curious, to me, is not that Hollinger uses point differential rather than W/L. It's that he chooses between the two rather than incorporating both.

From a statistical standpoint, the goal of his system should be "goodness of fit," meaning its ability to explain as much variation in championship success as possible.

And from that same standpoint, it HAS to be true that a model incorporating both W/L and differential will outperform a model using only one of those. It's very simple: the more information you use, the better your model. This will be true even if point differential is a much better predictor of who wins championships, as long as W/L contains some information that differential does not contain.

[LooseCannon]

What's curious, to me, is not that Hollinger uses point differential rather than W/L. It's that he chooses between the two rather than incorporating both.

From a statistical standpoint, the goal of his system should be "goodness of fit," meaning its ability to explain as much variation in championship success as possible.

And from that same standpoint, it HAS to be true that a model incorporating both W/L and differential will outperform a model using only one of those. It's very simple: the more information you use, the better your model. This will be true even if point differential is a much better predictor of who wins championships, as long as W/L contains some information that differential does not contain.

Why does that have to be true?  Maybe W/L contains some information that is less useful than point differential, so incorporating W/L makes the model worse.  I hope you are aware that, in baseball, pitcher ERA is a better predictor than pitcher W/L of future pitcher W/L. 

People like to talk about intangibles such as "heart", but the biggest intangible is probably just plain luck.

[ballin]

I came in here simply to state that PER is the worst "advanced" stat of all time. It really tells you nothing.

Give me win shares or win production any day of the week. Hell, I'll look at true shooting percentages and rebound percentages etc. if I have to before I turn to PER in order to evaluate a player. It's garbage.

[Boris Badenov]

What's curious, to me, is not that Hollinger uses point differential rather than W/L. It's that he chooses between the two rather than incorporating both.

From a statistical standpoint, the goal of his system should be "goodness of fit," meaning its ability to explain as much variation in championship success as possible.

And from that same standpoint, it HAS to be true that a model incorporating both W/L and differential will outperform a model using only one of those. It's very simple: the more information you use, the better your model. This will be true even if point differential is a much better predictor of who wins championships, as long as W/L contains some information that differential does not contain.

Why does that have to be true?  Maybe W/L contains some information that is less useful than point differential, so incorporating W/L makes the model worse.  I hope you are aware that, in baseball, pitcher ERA is a better predictor than pitcher W/L of future pitcher W/L. 

People like to talk about intangibles such as "heart", but the biggest intangible is probably just plain luck.

It has to be true because a model like this can choose whether to incorporate or ignore a piece of information, and the importance to place on it. The model would never use W/L in a way that made its explanatory ability worse, because simply ignoring W/L is a possibility. That possibility guarantees always doing at least as well as a model without W/L in it.

[Fafnir]

His system looks at scoring differential not wins and losses. The Heat have beaten up the bad teams they've played by a large margin.

Does anyone else actually agree with this system?  In my opinion scoring differential doesn't mean too much in basketball.

  I think in general scoring differential is more important than records in predictions, but obviously there are more factors involved.

I agree.  But I think about all the times that a game is much different than the score says it was.  Garbage time has got to skew the results.

What's curious, to me, is not that Hollinger uses point differential rather than W/L. It's that he chooses between the two rather than incorporating both.

From a statistical standpoint, the goal of his system should be "goodness of fit," meaning its ability to explain as much variation in championship success as possible.

And from that same standpoint, it HAS to be true that a model incorporating both W/L and differential will outperform a model using only one of those. It's very simple: the more information you use, the better your model. This will be true even if point differential is a much better predictor of who wins championships, as long as W/L contains some information that differential does not contain.

Too many variables often reduces the predictive power and goodness of fit of a model.

[Donoghus]

Hollinger is an interesting read and I have no doubt he's an incredibly smart guy.  I just take many of his articles with a grain of salt; heavy on the statistics while sometimes removing some common sense aspects you can see with your eyes or gauge by years of simply watching basketball and understanding how its played at the professional level. 

He's certainly a lightening rod for pundits, though.  Everytime someone posts a thread linked to one of his articles, we can get pages of debate.

[BballTim]

 What the Celtics did last year wasn't randomly outperform their expectations. There were specific reasons that the predictions were wrong and there were many people that would have had no problem pointing this out at any point before or during the playoffs. In other words, the "post mortems" of the many matched the predictions of a number of people (myself included).

Those specific reasons still made them an outlier though didn't it? There are always reasons for why events happen, but that doesn't make them the most likely outcome.

  Yes, they were clearly outliers. But the fact that they were outliers didn't make them unlikely contenders, it just made it difficult for that type of model to predict what was likely to happen. The fact that they didn't fit into the typical model didn't make them inherently less likely to win.