Behold, the idiocy that is John Hollinger

[chumdawg]

Originally Posted by LonghornDub

You're right. He discovered this objective formula existing naturally in the wilderness, slapped the "Hollinger" name on it, and threw it on ESPN.com. There was no subjective professional judgment involved in its creation.

Please.

Stats guys have been using MOV for a long, long time. It definitely did exist out there in the wild, as you say.

He may have stamped his name on it as some sort of predictor, but he didn't break any new ground, that's for sure.

[jthig32]

Originally Posted by chumdawg

Stats guys have been using MOV for a long, long time. It definitely did exist out there in the wild, as you say.

He may have stamped his name on it as some sort of predictor, but he didn't break any new ground, that's for sure.

He's applying completely subjective modifiers in his calculation. Modifiers that were either attained from subjective regression analysis or from good old fashioned guessing.

Either way, the root of the formulas is clearly subjective.

[LonghornDub]

Originally Posted by chumdawg

Stats guys have been using MOV for a long, long time. It definitely did exist out there in the wild, as you say.

He may have stamped his name on it as some sort of predictor, but he didn't break any new ground, that's for sure.

What thig said. His power rankings are not straight MoV, and you know it.

[Murphy3]

Here's the bottom line.. It's impossible to quantify everything. The Mavs could win a game by 4 points... This may look like a "close" game, but that might not be the case at all. Perhaps they were up by 20 throughout but the opponent hit 2 meaningless threes in the last 15 seconds or so to get it down to 4... So, it's a 4 point win but the game was never in doubt. How do the statisticians handle this game? Is it a close game? Is it a "coinflip" game? Is it a blowout because the other team was down 20 throughout?

There are obvious HUGE flaws that statisticians HOPE will even out over the course of the season. However, they often don't.

And at the root of ever formula such as Hollinger's is a person that decides how much weight to put into certain categories.

[Murphy3]

Originally Posted by LonghornDub

What thig said. His power rankings are not straight MoV, and you know it.

And regardless, the stat MOV can be deceptive as well. Two to three blowouts that turn into "close" games that were never really close throughout the course of a season can start to have an impact on MOV.

[j0Shi]

I've read somewhere that Dean Oliver is the head of ESPN statistics. I wonder why they are using Hollinger's numbers then ...

[dirno2000]

Originally Posted by joemoeschmoe

Furthermore, the Mavs winning is kind of a knock on his system along with 3 of the top 6 going out in the first round. Yet another reason for him to be defensive.

I still contend that the basic assumption -- that MOV tells us (almost) all about how good a team is -- is fundamentally flawed in basketball.

You tell me this: Mavs vs some scrub team with 2 min to go and tied. You're telling me that this is supposed to be a coin flip? Pick any elite team and any scrub team. I call BS.

It makes sense that the more you shorten games the more random the outcomes. For instance if you were to replay this past regular season but instead of 48 minute games each team got one possession what would the final standings look like? My guess is that every team would gravitate towards .500. Make the games 2 minutes and you would start to see a bit more separation, 10 minutes even more, and so on. It's hard to argue that playing longer games (or in the case of the NCAA vs. the NBA more games) favors the better team.

I think where Holllinger misses the boat is refusing to concede that we may be an outlier when it comes to consistently winning close games. From his pre-season prediction column:

The biggest question in projecting the Mavs is how they'll use the roster. I project each team based on the idea that they'll play their best-rated players the most (provided those players are healthy), but in Dallas' case I feel less assured than in some others. I rated them with Beaubois playing nearly 2,000 minutes after returning from injury, which would require the Mavs to use Butler exclusively as a backup small forward and to reduce Terry's minutes. I'm not sure they're ready to do this, but if Beaubois plays as well as he did a year ago, I don't think they'll have a choice.

The projection for Beaubois is perhaps too optimistic; on the other hand, Kidd's looks far too negative. On balance, it evens out. The other factor to consider is the magic beans the Mavs have been using to spirit close games into the win column in recent seasons. I remain a non-believer, but if you want to give Dallas extra credit for this you should add about three wins to the total below (48).

Maybe you have to watch game in and game out to know that Dirk and Jet have a knack for making big shots. Unfortunately before this past season our success in close games didn't always extend to the post season when everyone is paying attention.

As far as his power rankings I’d bet that most statistical rating systems had us in the 7/8 range so I don’t know if his is any better or worse than say Sagarin or basketball-reference. if he looks bad (or biased) it's because he honestly thought that this was a 48 win team.

[LonghornDub]

Originally Posted by dirno2000

As far as his power rankings I’d bet that most statistical rating systems had us in the 7/8 range so I don’t know if his is any better or worse than say Sagarin or basketball-reference. if he looks bad (or biased) it's because he honestly thought that this was a 48 win team.

I find it hard to fault Hollinger individually in this instance when pretty much everyone had the Mavs around that same spot. I don't think anybody any this board would have predicted they'd win the title. There were a fair amount of people here who had them losing to the Blazers, and at least 40% had them losing to the Lakers.

Hollinger was definitely "wrong," but so was just about everyone else, so I'm not gonna single him out this time.

[OSUmavsfan]

Originally Posted by Murphy3

Here's the bottom line.. It's impossible to quantify everything. The Mavs could win a game by 4 points... This may look like a "close" game, but that might not be the case at all. Perhaps they were up by 20 throughout but the opponent hit 2 meaningless threes in the last 15 seconds or so to get it down to 4... So, it's a 4 point win but the game was never in doubt. How do the statisticians handle this game? Is it a close game? Is it a "coinflip" game? Is it a blowout because the other team was down 20 throughout?

There are obvious HUGE flaws that statisticians HOPE will even out over the course of the season. However, they often don't.

And at the root of ever formula such as Hollinger's is a person that decides how much weight to put into certain categories.

I think an interesting twist on MOV would be considering the margin at every point in the game, meaning basically an integration of the margin with respect to time over the course of each game. So if a team is up by 20 for an entire game, but the lead gets cut to 5 in the last two minutes that would look better than a game that's tight throughout with a team winning by 5 in the end. I don't know why Hollinger or anyone else hasn't adopted this approach yet.

[bobbyfg7]

Hollinger is a wiener.

[endrity]

Originally Posted by LonghornDub

Hollinger adds something new to the discussion. He gets credit for that. He's also a very smart guy. He gets credit for that too.

But that's not the point. The point is whether it's fair to give the guy a pass when the formula he created spits out bad predictions because, hey, it's the formula! Hollinger didn't do it! Please.

Like I said, the appropriate defense for Hollinger is that everybody is wrong frequently. I happen to come at Hollinger more than other talking heads because I don't think his model is particularly worthwhile, he writes a bunch of self-serving ex post facto rationalizations, and he is generally a douche. But he's certainly not the only one who makes bad predictions.

But that's what I wanted to say in the first place. Do you honestly think there is a formula out there that can predict winners 100% of the time. I don't think there is. The Mavs run this year is a complete outlier by any standard. Yet stuff like that happens all the time in sports and no formula is going to be able to predict that.

He does his best, it's about as objective a criteria as there is out there compared to all the other writers. When he goes wrong, at least he gives an explanation as to why his formula failed, what other indicators were out there etc. Most other writers tell you smth like "X team just played with more desperation than Y, and that's why they won" which is ok at times, but hardly insightful.

[LonghornDub]

Originally Posted by endrity

But that's what I wanted to say in the first place. Do you honestly think there is a formula out there that can predict winners 100% of the time. I don't think there is. The Mavs run this year is a complete outlier by any standard. Yet stuff like that happens all the time in sports and no formula is going to be able to predict that.

See:

Originally Posted by LonghornDub

I find it hard to fault Hollinger individually in this instance when pretty much everyone had the Mavs around that same spot. I don't think anybody any this board would have predicted they'd win the title. There were a fair amount of people here who had them losing to the Blazers, and at least 40% had them losing to the Lakers.

Hollinger was definitely "wrong," but so was just about everyone else, so I'm not gonna single him out this time.

[endrity]

Originally Posted by dirno2000

It makes sense that the more you shorten games the more random the outcomes. For instance if you were to replay this past regular season but instead of 48 minute games each team got one possession what would the final standings look like? My guess is that every team would gravitate towards .500. Make the games 2 minutes and you would start to see a bit more separation, 10 minutes even more, and so on. It's hard to argue that playing longer games (or in the case of the NCAA vs. the NBA more games) favors the better team.

I think where Holllinger misses the boat is refusing to concede that we may be an outlier when it comes to consistently winning close games. From his pre-season prediction column:



Maybe you have to watch game in and game out to know that Dirk and Jet have a knack for making big shots. Unfortunately before this past season our success in close games didn't always extend to the post season when everyone is paying attention.

As far as his power rankings I’d bet that most statistical rating systems had us in the 7/8 range so I don’t know if his is any better or worse than say Sagarin or basketball-reference. if he looks bad (or biased) it's because he honestly thought that this was a 48 win team.

Actually Hollinger was a big fan of Dallas early in the season. If he thought we were a 48 win team it's because Butler went out.

[dirno2000]

Originally Posted by endrity

Actually Hollinger was a big fan of Dallas early in the season. If he thought we were a 48 win team it's because Butler went out.

That was his pre-season prediction. The 5th seed in the West with 48 wins. I'd link the article but it's Insider.

[Thespiralgoeson]

Originally Posted by dirno2000

Maybe you have to watch game in and game out to know that Dirk and Jet have a knack for making big shots. Unfortunately before this past season our success in close games didn't always extend to the post season when everyone is paying attention.

More important than that, I think, is the fact that the Mavs have consistently been one of the best FT shooting teams in the league throughout the entire Dirk Nowitzki era. I think that winning so many close game might have something to do with that, but Hollinger has never once mentioned it as far as I know.

[endrity]

Originally Posted by dirno2000

That was his pre-season prediction. The 5th seed in the West with 48 wins. I'd link the article but it's Insider.

I was talking about the Hollinger rankings in the early part of the season. I'll trust you about his preseason prediction.

[endrity]

Originally Posted by Thespiralgoeson

More important than that, I think, is the fact that the Mavs have consistently been one of the best FT shooting teams in the league throughout the entire Dirk Nowitzki era. I think that winning so many close game might have something to do with that, but Hollinger has never once mentioned it as far as I know.

It's one of the big debates in advanced metrics right now. Clutch statistics and the "hot hand", and right now most of the stats guys tend to believe that the stats don't prove anything. Now Cuban said smth about him believing in clutch stats when talking about Kidd, so maybe the Mavs have found some new indicator and ofcourse will not share it.

The unfortunate thing about this debate is that most of the resources (financial and human) are being spent by the teams, yet we have no real idea what they are looking into and probably most teams are looking in completely different directions.

[alexamenos]

'hot hand' and clutch are two different things.

Re 'hot hand', who doesn't believe that technique and confidence affect whether a shot will go in? Those who don't are free to argue that there's no such thing as a 'hot hand' (via the waxing and waning of technique and confidence), but those who do (ie me) have a bit of trouble with the argument.

Re 'clutch' -- if there is no such thing as 'clutch', then there is likewise no such thing as choke. If there is no such thing as choke...well, bullshit, there is too such a thing as choke. I know, I've done it.

[nowhereman]

changing the direction of the thread a little bit, I think we are ready to acknowledge that the advanced stats movement in basketball has nothing to do with ranking teams ahead or behind each other, and everything to do with determining the best lineup iterations possible at various stages in the game and against various lineups. It really doesn't matter what the team MOV is as long as our combined 5 score more than their combined 5. Anyone can throw out a formula with some subjective coefficients and claim it tells you who the best team is, but that has nothing to do with what the numbers actually mean.

[Murphy3]

Personally, I don't fault Hollinger for being "wrong". He should just admit that his formula was "wrong" instead of trying to make excuses for it. I was "wrong" at times this season concerning this team. I won't make any excuses for getting things "wrong" at times... Hollinger just needs to do the same.

His formula has obvious holes. The holes are beyond easy to point out. But that's just it.. no formula is perfect or even close. Just admit that there are incredible limitations to your rankings and move on.

Personally, I like Hollinger a little... I have no issues with using some of his statistical measurements in argument. But I also realize that no singular stat or formula can tell the whole picture when comparing a team or an individual player.

[jthig32]

Originally Posted by nowhereman

changing the direction of the thread a little bit, I think we are ready to acknowledge that the advanced stats movement in basketball has nothing to do with ranking teams ahead or behind each other, and everything to do with determining the best lineup iterations possible at various stages in the game and against various lineups. It really doesn't matter what the team MOV is as long as our combined 5 score more than their combined 5. Anyone can throw out a formula with some subjective coefficients and claim it tells you who the best team is, but that has nothing to do with what the numbers actually mean.

I had a takeaway similar to this from watching some of the panels at the Sloan Sports Analytics conference this year. They had Hollinger on a panel with Cuban and other statistical minds from sports teams and it was painfully obvious that no one cared about Hollinger's contributions one whit. They are purely for consumer consumption. Those panelists could not possibly care any less about arguing which team is better or which player is better based on PER.

I will say though, that multiple on that panel mentioned without hesitation that when they ARE trying to compare teams, MOV is what they use.

[MascisMan]

Originally Posted by LonghornDub

Hollinger adds something new to the discussion. He gets credit for that. He's also a very smart guy. He gets credit for that too.

But that's not the point. The point is whether it's fair to give the guy a pass when the formula he created spits out bad predictions because, hey, it's the formula! Hollinger didn't do it! Please.

Like I said, the appropriate defense for Hollinger is that everybody is wrong frequently. I happen to come at Hollinger more than other talking heads because I don't think his model is particularly worthwhile, he writes a bunch of self-serving ex post facto rationalizations, and he is generally a douche. But he's certainly not the only one who makes bad predictions.

I can't believe Im agreeing with Dub but he's right.

Hollinger chose specific weightings within the formula to predict a team's value or success based on his opinion of what was important and most impactful.

There are mathematical formulas that exist that have been proven to provide factual results 100% of the time. This is not one of those. This is more of a forecasting formula. And as we all know, the weather man is often wrong.

What is really disconcerting is that it seems like he never tweaks his formula from year to year even though his weightings/values are often wrong.

Also keep in mind that despite his formula's rankings (that he believes in) he picked POR over DAL in round 1. How is that for a guy that lived in Portland and is not injecting his opinion into his analysis?

[joemoeschmoe]

Originally Posted by dirno2000

It makes sense that the more you shorten games the more random the outcomes. For instance if you were to replay this past regular season but instead of 48 minute games each team got one possession what would the final standings look like? My guess is that every team would gravitate towards .500. Make the games 2 minutes and you would start to see a bit more separation, 10 minutes even more, and so on. It's hard to argue that playing longer games (or in the case of the NCAA vs. the NBA more games) favors the better team.

Wrong. All teams would not gravitate towards .500. Yes, you get more randomness in the outcomes, but in the end the better teams would tend to have better records. Standard deviation is not the same mean. In the case you gave, you have higher standard deviation, but the same mean.

If you played 100 games with each team getting 1 possession, the better teams would win more than 50% of the time.

You will have have some streaks where good teams lose a bunch in a row and bad teams win a bunch in a row, but in the end, most good teams will have good records and vice versa.

[jthig32]

Originally Posted by joemoeschmoe

Wrong. All teams would not gravitate towards .500. Yes, you get more randomness in the outcomes, but in the end the better teams would tend to have better records. Standard deviation is not the same mean. In the case you gave, you have higher standard deviation, but the same mean.

If you played 100 games with each team getting 1 possession, the better teams would win more than 50% of the time.

You will have have some streaks where good teams lose a bunch in a row and bad teams win a bunch in a row, but in the end, most good teams will have good records and vice versa.

This doesn't make him "wrong". He said "gravitate towards", not "stick right on .500". His point is sound: the shorter the time frame, the more randomness is introduced, and the less the result is based on the quality of the two teams.

[grndmstr_c]

Perhaps some of the confusion here relates to the issue of whether we're talking about predicting outcomes of games from known performance parameters, or are instead trying to estimate performance parameters from known game outcomes. In the former case the number of possessions would make no difference at all unless the number of possessions in a game were itself a predictor of performance (for example, because of incongruities between team ranks on probability of scoring and expected points per possession). In the latter case, however, it would make a huge difference. Though team performance wouldn't gravitate towards .500 (joe's right in that respect), the ability to estimate a team's true quality, and hence to predict their future performance based on extremely small samples such as would be available in a season that had 82 games of one possession each, would be completely shot, and prediction accuracy would undoubtedly slip towards chance levels.

[Kirobaito]

Originally Posted by OSUmavsfan

I think an interesting twist on MOV would be considering the margin at every point in the game, meaning basically an integration of the margin with respect to time over the course of each game. So if a team is up by 20 for an entire game, but the lead gets cut to 5 in the last two minutes that would look better than a game that's tight throughout with a team winning by 5 in the end. I don't know why Hollinger or anyone else hasn't adopted this approach yet.

This.

By george, you've done it.

[joemoeschmoe]

Originally Posted by jthig32

This doesn't make him "wrong". He said "gravitate towards", not "stick right on .500". His point is sound: the shorter the time frame, the more randomness is introduced, and the less the result is based on the quality of the two teams.

No. He is wrong. They will not "gravitate towards" .500. What I presume he's trying to say is that they will converge on or generally be close to .500 -- which is incorrect.

Let's use an extreme example to prove a point. Let's say we put a team of 5 Michael Jordan's vs 5 Antoine Walker's against each other in an 82 game season. Give them one possession or 1 minute or something equally tiny.

Now, the MJ team is CLEARLY better, but Walker could (and would) jack up long, stupid jumpers and win a few in an 82 game season. As a result the Walker team would win a few games from sheer luck. However, it would be nowhere near 50%.

Again, variation is not the same as mean.

[iella]

Can we stop arguing about definitions of statistical terms? Please? All it does is bring back bad memories of my engineering statistics class in college. If people are wrong, let them be wrong. Not like this:

[jthig32]

Originally Posted by joemoeschmoe

No. He is wrong. They will not "gravitate towards" .500. What I presume he's trying to say is that they will converge on or generally be close to .500 -- which is incorrect.

Let's use an extreme example to prove a point. Let's say we put a team of 5 Michael Jordan's vs 5 Antoine Walker's against each other in an 82 game season. Give them one possession or 1 minute or something equally tiny.

Now, the MJ team is CLEARLY better, but Walker could (and would) jack up long, stupid jumpers and win a few in an 82 game season. As a result the Walker team would win a few games from sheer luck. However, it would be nowhere near 50%.

Again, variation is not the same as mean.

Fair enough on the .500 point.

But the larger point is still valid, obviously. Randomness is introduced with the shorter duration.

And to the initial point, on how effective MOV, I would just reiterate that Cuban and other statistical minds are on record that MOV is a big indicator of team quality.

[endrity]

Originally Posted by joemoeschmoe

No. He is wrong. They will not "gravitate towards" .500. What I presume he's trying to say is that they will converge on or generally be close to .500 -- which is incorrect.

Let's use an extreme example to prove a point. Let's say we put a team of 5 Michael Jordan's vs 5 Antoine Walker's against each other in an 82 game season. Give them one possession or 1 minute or something equally tiny.

Now, the MJ team is CLEARLY better, but Walker could (and would) jack up long, stupid jumpers and win a few in an 82 game season. As a result the Walker team would win a few games from sheer luck. However, it would be nowhere near 50%.

Again, variation is not the same as mean.

For everyone else, if I may help, what I think he is trying to say is this:

- if teams played a couple of these types of seasons (one possession,short games, whatever) over and over again, the variation from one season to the other of teams' record would be quite large, even if the teams' abilities remain the same from one season to the other.

What the other guy is saying is that no matter how many times you repeat these seasons, teams would converge around a .500 record.

I think Joe is right.

[endrity]

Originally Posted by iella

Can we stop arguing about definitions of statistical terms? Please? All it does is bring back bad memories of my engineering statistics class in college. If people are wrong, let them be wrong. Not like this:

lol!

[DwD]

I resemble that cartoon good one!

Originally Posted by iella

Can we stop arguing about definitions of statistical terms? Please? All it does is bring back bad memories of my engineering statistics class in college. If people are wrong, let them be wrong. Not like this:

[dirno2000]

Ok here's my logic around teams gravitating towards .500 in the scenario I laid out. Tell me where it goes off the rails:

- It's a zero sum game. For every win there's a loss so the league mean will always be .500.

- What explains variance to the mean from team to team? Mostly skill but there's some luck in there too (i.e. Tony Delk goes for 50). For the sake of argument let’s say that it's 80% skill (team quality) and 20% luck/randomness.

- I think everyone would agree that the shorter the game the more random the outcome. So again for the sake of argument let’s say that in the short game scenario the factors that decide the outcome of a game are reversed (20% skill 80% luck/randomness).

With randomness being the major factor in the outcome of each game, why wouldn't teams gravitate towards the mean in the long run?

[grndmstr_c]

Originally Posted by dirno2000

Ok here's my logic around teams gravitating towards .500 in the scenario I laid out. Tell me where it goes off the rails:

- It's a zero sum game. For every win there's a loss so the league mean will always be .500.

- What explains variance to the mean from team to team? Mostly skill but there's some luck in there too (i.e. Tony Delk goes for 50). For the sake of argument let’s say that it's 80% skill (team quality) and 20% luck/randomness.

- I think everyone would agree that the shorter the game the more random the outcome. So again for the sake of argument let’s say that in the short game scenario the factors that decide the outcome of a game are reversed (20% skill 80% luck/randomness).

With randomness being the major factor in the outcome of each game, why wouldn't teams gravitate towards the mean in the long run?

Think about it this way. This skill component will be the same from one contest to the next, whereas the random component will constantly be shifting around, sometimes causing teams to overachieve, other times causing them to underachieve, and in some cases having no discernable effect on the final score of the game. In the long run the nonconstant nature of the random component leads to an averaging out of the positive and negative fluctuations around each team's mean performance level, and the constant reliable skill component comes to dominate each team's winning percentage. That will be the case even with one possession games. It's just that it will take a lot more games for the skill component to start dominating in that case then it does in the current situation where each game has over 80 possessions.

[chumdawg]

D2K is clearly correct here, and I'm amazed anyone would even think to argue otherwise.

The premise is very simple: good teams will win fewer games in a one-possession format than they would in a 48-minute format...and bad teams would also lose fewer.

The net result is, indeed, that the teams would gravitate toward .500.

There is a really good article on this concept on one of the sports websites, but at the moment I can't remember which. The point was that weaker teams improve their chances by introducing more randomness into the contest.

[chumdawg]

Originally Posted by grndmstr_c

Think about it this way. This skill component will be the same from one contest to the next, whereas the random component will constantly be shifting around, sometimes causing teams to overachieve, other times causing them to underachieve, and in some cases having no discernable effect on the final score of the game. In the long run the nonconstant nature of the random component leads to an averaging out of the positive and negative fluctuations around each team's mean performance level, and the constant reliable skill component comes to dominate each team's winning percentage. That will be the case even with one possession games. It's just that it will take a lot more games for the skill component to start dominating in that case then it does in the current situation where each game has over 80 possessions.

I get what you are saying, but the difference of opinion comes when you say "the skill component will be the same from one contest to the next." When you are talking about a one-possession game versus a full game, you are talking about a markedly different "skill component." It doesn't translate completely in this situation, because elements of a good team's "skill component" no longer come into play. Things like roster depth, coaches' in-game abilities to gain matchup advantages, playing a physical style that wears the opponent down, getting in the penalty early and knocking down free throws, having a balanced offensive attack with several weapons as opposed to relying on one star for the bulk of the offense...and I could go on and on.

[grndmstr_c]

Originally Posted by chumdawg

D2K is clearly correct here, and I'm amazed anyone would even think to argue otherwise.

The premise is very simple: good teams will win fewer games in a one-possession format than they would in a 48-minute format...and bad teams would also lose fewer.

The net result is, indeed, that the teams would gravitate toward .500.

There is a really good article on this concept on one of the sports websites, but at the moment I can't remember which. The point was that weaker teams improve their chances by introducing more randomness into the contest.

I'd like to see the article, because I'm very skeptical of that claim, depending on the nuances of the assumptions on which it's based. The idea that increased randomness could lead to a clustering of teams around .500 would make sense if you're talking about increasing randomness at the level of team performance. But if you're merely talking about reducing the per-game sample size I can't see any way that that would fundamentally change the long-run probability of winning for individual teams.

[Murphy3]

It could definitely chane some teams. Some teams might be better equipped in that specific type of situation. However, over the course of a season, GMC, I have to believe that you are correct.

[endrity]

Originally Posted by dirno2000

Ok here's my logic around teams gravitating towards .500 in the scenario I laid out. Tell me where it goes off the rails:

- It's a zero sum game. For every win there's a loss so the league mean will always be .500.

- What explains variance to the mean from team to team? Mostly skill but there's some luck in there too (i.e. Tony Delk goes for 50). For the sake of argument let’s say that it's 80% skill (team quality) and 20% luck/randomness.

- I think everyone would agree that the shorter the game the more random the outcome. So again for the sake of argument let’s say that in the short game scenario the factors that decide the outcome of a game are reversed (20% skill 80% luck/randomness).

With randomness being the major factor in the outcome of each game, why wouldn't teams gravitate towards the mean in the long run?

Right at that bold part is where you are right for the last time.

Here's how you can think of it. Let's assume we know the exact skill level of a team, and let's assume that this skill level is perfectly measured by team record. So a good team's perfect measurement is, let's say, a 60-22 record. If there were no randomness in any of this than no matter how many times you repeat a season, this is where this team will end up. Likewise, a bad team's perfect measurement is 22-60. You are correct when you say that the league overall is and will always be .500 not matter what, but as you'll see below this doesn't matter at all for specific teams' record.

Now let's add randomness to the model.
Let's say that the way to predict the team's record is by a simple equation:
y = x + e
where x is the perfect record and e is the random error. Assume that e is normally distributed (which means that it tends to be negative just as much as positive, in simple terms if you don't know what I am writing).

What a smaller game will do therefore is make the effect of e larger*. So therefore the long run mean for a team's record will continue to be it's perfect measurement x, not .500, but the band within wich it will variate will grow larger. It will be just as likely for the good team to have a 70-12 season as it will be to have a 50-32 season. The same can't be said about the bad team. Its probability of getting a 70-12 season is extremely small, and smaller than getting a 50-32 season of course, and quite different than the scenario involving the good team. So the means for the teams' records will continue to be their previous ones, x, not .500. Longer games mean that the effect of e gets smaller, and the band within which a team record will variate will diminish. I hope I wrote it clearly. Without graphs this is a bit difficult to explain.

* for those who studied regression analysis I made an incorrect simplification of course, because variables should have coefficients in front of them, but the point is to make it easy for everyone to understand

[grndmstr_c]

Originally Posted by chumdawg

I get what you are saying, but the difference of opinion comes when you say "the skill component will be the same from one contest to the next." When you are talking about a one-possession game versus a full game, you are talking about a markedly different "skill component." It doesn't translate completely in this situation, because elements of a good team's "skill component" no longer come into play. Things like roster depth, coaches' in-game abilities to gain matchup advantages, playing a physical style that wears the opponent down, getting in the penalty early and knocking down free throws, having a balanced offensive attack with several weapons as opposed to relying on one star for the bulk of the offense...and I could go on and on.

You misunderstand me. I'm not claiming that the determinants of winning and losing are the same in a 100 possession game as they are in a one possession game. I'm just saying that skill, as distinct from randomness, is by definition constant.

That said, you make a good point that skill, whatever it is, is surely multidimensional, and there are certainly very complex interactions that go into determining precisely how performance across different dimensions influences the long-run odds of winning in a particular matchup between two specific teams. I just don't view that point as being particularly critical to the discussion at hand.