Hollinger's 2010 Playoff Odds

[j_fran]

Someone needs to explain to me how a computer that runs 5000 season scenarios has Boston winning the most games, having the highest percentage chance of making the Finals in the East, the second highest chance of making the Finals in the league but be tied for 6th with Atlanta for chances to win the Championship if, as Hollinger claims, the only information it is given is the Hollinger Daily Power Ranking's numbers?

How can a computer mathematically and randomly come up with the team that is second most likely to get to the championship game come up with that team having only tied for 6th best chance of winning it all unless there is definite matchup tendencies or other info being plugged in?

This is why I hate Hollinger's stats. They make no sense logically.

If he's running a monte carlo then the standard deviation probably accounts for the difference nick.

That's just it. If he's running that then you might figure another team might just as well be off. But what appears are numbers that shows his preseason picks of LA and Cleveland being the most likely winners. Coincidence?

Again, I don't think Hollinger is the most honest of mathematicians or statisticians. But you already knew that, Faf.

They're the most likely winners by a few percentage points, its all pretty even among the contenders. Denver has the same chance as Cleveland after all.

I also thing you're making a mountain out of a mole hill about a Monte Carlo that's likely on autopilot. We can always just see how it keeps fluctuating.

So what do you think of today's playoff odds:

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

The Celtics keep winning, their Daily Power Rankings numbers go up and their position in the Power Rankings improve and yet their odds to win it all decrease??

Still believing the discrepancy on these numbers are a simple Monte Carlo? I think he is inputting further data and not admitting it.

I have problems with Hollinger at times as well, specifically with his devotion to using only PER to determine his player rankings and end of year awards.

But in this case he's fine.  What reason would he have to "cook the books?"  If he was found to be messing with his stastical analysis he wouldn't have a job.  We can disagree with his methods but there's no chance he's cheating so to speak.  And one of many possible reasons is that while the Celtics play in the worst division in the NBA and should get the top seed in the East, they've also won one and lost four games to the other top seven teams in the league.  They beat Cleveland and lost at home to Orlando, Atlanta, and Phoenix.  So while the Celtics overall record is stellar, when Hollinger runs the simulations for the playoffs, the Celtics poor record against the other contenders probably hurts them.  Therefore the lower chance to win the title.

[nickagneta]

I'm guessing that is the team's age is has their Title chances so low.

Otherwise, it would be quite strange given the Celtics impressive statistical performance so far.

Not sure that works Who as Orlando is actually a team with an average age higher than that of the Celtics, they have worse seasonal stats than the Celtics, the computer random generation of their season shows Orlando with a likelihood of finishing 3rd in the East but with the highest percentage chance of winning it all of all Eastern Conference teams and second best chance overall.

Dallas is another team that is older than the C's that is showing better chances of winning it all even though they have worse seasonal stats.

A Monto Carlo or or other type number generation that comes up with odds of this type would not regularly have the team with the best stats being input into the system coming out 6th to 8th highest in the odds to win it all unless numbers were being cooked some. It's almost a statistical impossibility.

[Fafnir]

Someone needs to explain to me how a computer that runs 5000 season scenarios has Boston winning the most games, having the highest percentage chance of making the Finals in the East, the second highest chance of making the Finals in the league but be tied for 6th with Atlanta for chances to win the Championship if, as Hollinger claims, the only information it is given is the Hollinger Daily Power Ranking's numbers?

How can a computer mathematically and randomly come up with the team that is second most likely to get to the championship game come up with that team having only tied for 6th best chance of winning it all unless there is definite matchup tendencies or other info being plugged in?

This is why I hate Hollinger's stats. They make no sense logically.

If he's running a monte carlo then the standard deviation probably accounts for the difference nick.

That's just it. If he's running that then you might figure another team might just as well be off. But what appears are numbers that shows his preseason picks of LA and Cleveland being the most likely winners. Coincidence?

Again, I don't think Hollinger is the most honest of mathematicians or statisticians. But you already knew that, Faf.

They're the most likely winners by a few percentage points, its all pretty even among the contenders. Denver has the same chance as Cleveland after all.

I also thing you're making a mountain out of a mole hill about a Monte Carlo that's likely on autopilot. We can always just see how it keeps fluctuating.

So what do you think of today's playoff odds:

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

The Celtics keep winning, their Daily Power Rankings numbers go up and their position in the Power Rankings improve and yet their odds to win it all decrease??

Still believing the discrepancy on these numbers are a simple Monte Carlo? I think he is inputting further data and not admitting it.

Additionally, the results regress to the mean. This is more important early in the season, and what it essentially means is that even though a team might start 10-0, it's not necessarily bound to go 82-0. The effect of this will reduce sharply after the first quarter of the season or so, but in the early going of most seasons, it's necessary to prevent us from projecting 77-win seasons and the like.

This is the likely culprit. The mechanism he uses to "regress to the mean" is the reason he has the C's that low, especially if that mechanism uses the age of the team. The C's are playing the best and therefore would regress the most out of the teams. The Lakers have a weaker conference so this factor doesn't hurt them as much in the "championship odds".

I still think you're paranoid, Nick. Why in the world would he be input anything extra?

Also the std deviation of such a Monte Carlo data set is likely to be as high as the difference between the Celtics and Cavs title chances.

[Celtics17]

Statistics are just that, statistics and nothing more. You just cant measure such things as desire, heart, how much the players drank the night before etc. Consequently, these things are always bogus, especially this early in the season. I once heard and explanation of statistical analysis this way. If your head is in an oven and your behind in a freezer then on average you are doing pretty well.

[Fafnir]

Statistics are just that, statistics and nothing more. You just cant measure such things as desire, heart, how much the players drank the night before etc. Consequently, these things are always bogus, especially this early in the season. I once heard and explanation of statistical analysis this way. If your head is in an oven and your behind in a freezer then on average you are doing pretty well.

Its been more than 25% of the season, stats can tell you plenty about the league.

Your example of a head in the oven and but is funny, but only really works when you have a two point data set.

[Who]

I'm guessing that is the team's age is has their Title chances so low.

Otherwise, it would be quite strange given the Celtics impressive statistical performance so far.

Not sure that works Who as Orlando is actually a team with an average age higher than that of the Celtics

Orlando and the other contenders have fewer key players as old + with as many miles on them as the Celtics have. I don't know how Hollinger works his numbers ... age might have nothing to do with it at all.

I was just wondering because he's been negative on the Celtics in the past due the age(s) of their main players. Guessing that might be due to whatever his numbers tell him about an older team staying healthy + maintaining their performance throughout.

---------------------------------------------------

Rebounding perhaps?

The only team that I can think of that won a Championship in the last 30 years with a negative rebounding differential was the Houston Rockets of the mid 90's.

[Who]

Did Hollinger ever test his title predictor on past seasons?

Was it more accurate than W-L, point differential, or efficiency differentials?

[nickagneta]

But a standard Monte Carlo would not consistently on a day in and day out basis have the team with the best statistical input come out on the lower side of the mean for eventual winner odds. Orlando is an older team than the C's and they are consistently on the high side of the chances for a championship.


Faf, you're leaving this part out:

The computer starts with the day's Hollinger power rankings. Then, in each of the 5,000 times it replays the season, it makes a random adjustment up or down to allow for the possibility that a team will play better or worse than it has done thus far. (I call this the Anti-Dennis Green Postulate; i.e., maybe they aren't who we thought they were.)

The only stats he is inputting are those compiled for the Power Rankings:

Strength of schedule
Strength of schedule last 10 games or last 25% of season
Scoring margin
Scoring margin last 10 games or last 25% of season
Team record
Team record last 10 games or last 25% of season
Home and road games played
Home and road games played last 10 games or last 25% of season
Total games played

If this is true then I find it tremendously difficult to see the consistent 7th-8th place finish in the probability of winning the championship of the team that has been consistently 1st or 2nd in stats being input.

Call me paranoid but the mathematician in me says he is adding in other data. I find it impossible to believe otherwise.

[Fafnir]

If this is true then I find it tremendously difficult to see the consistent 7th-8th place finish in the probability of winning the championship of the team that has been consistently 1st or 2nd in stats being input.

Call me paranoid but the mathematician in me says he is adding in other data. I find it impossible to believe otherwise.

But Nick you've completely ignored my post, the higher your performance the more you have to regress. Nor are you a mathematician, like I've said I don't see these results with the C's a few percentage points as an indicator that he's cooking the books. I just don't see it. The power rankings and playoff odds loved the C's the last two years, why would he suddenly be forcing his model to make the C's look bad?

Anyways its impossible for me to defend a model when I don't have access to it in its entirety.

(Oh and Orlando is not an older team than the C's if you weight their best players for minutes played)

[Who]

Strength of schedule
Yes, this matters in the NBA, too. It is not as profound in the pro game as in the college game, because the 30 NBA teams are more evenly matched, but it still affects a team's results.

This comes into play mainly in the early part of the season, when there can be wide disparities in the quality of competition, but even at the end of the season, there will be differences among teams -- particularly when one conference is far better than the other.

Celtics are 29th in SoS with opponents having a winning percentage of only .452. Only Dallas has an SoS of lower than 50% (Dallas = 48%) and a higher chance at a title according to Hollinger's numbers.

Edit: Ah no, Denver is 30th. How did I miss that? And they have a higher title shot than Boston (7% to 5.8%). So it's not this statistic either. Very strange numbers altogether.

[Fafnir]

Strength of schedule
Yes, this matters in the NBA, too. It is not as profound in the pro game as in the college game, because the 30 NBA teams are more evenly matched, but it still affects a team's results.

This comes into play mainly in the early part of the season, when there can be wide disparities in the quality of competition, but even at the end of the season, there will be differences among teams -- particularly when one conference is far better than the other.

Celtics are 29th in SoS with opponents having a winning percentage of only .452. Only Dallas has an SoS of lower than 50% (Dallas = 48%) and a higher chance at a title according to Hollinger's numbers.

Edit: Ah no, Denver is 30th. How did I miss that? And they have a higher title shot than Boston (7% to 5.8%). So it's not this statistic either. Very strange numbers altogether.

Yes and no, consider that Sacramento has a non-zero chance to win the title. Same with the Knicks, Toronto, and Washington.

The std. deviation of this model is more than big enough have me shrug off the C's winning only 40% of the time they "make" the finals.

The Celtics have been consistently predicted to the have the best regular season record when they've been on top of the "rankings".

[nickagneta]

If this is true then I find it tremendously difficult to see the consistent 7th-8th place finish in the probability of winning the championship of the team that has been consistently 1st or 2nd in stats being input.

Call me paranoid but the mathematician in me says he is adding in other data. I find it impossible to believe otherwise.

But Nick you've completely ignored my post, the higher your performance the more you have to regress. Nor are you a mathematician, like I've said I don't see these results with the C's a few percentage points as an indicator that he's cooking the books. I just don't see it. The power rankings and playoff odds loved the C's the last two years, why would he suddenly be forcing his model to make the C's look bad?

Anyways its impossible for me to defend a model when I don't have access to it in its entirety.

(Oh and Orlando is not an older team than the C's if you weight their best players for minutes played)

So now you are defending him by breaking age data down to the ridiculous to prove a point? Do you really believe in order to come up with these odds in a Monte carlo methodology he input data down to the final detail of average age per minute played?? And if you do, then you prove my point, he's inputting data he isn't admitting to.

As for the regression to mean, I understand the concept. I also understand such regressions and their equal and opposite aggressions away from the mean, would mean that on a day to day basis, other teams with similarly excellent data being input would have inconsistencies. The Lakers would not consistently always have the best chances of winning. Orlando wouldn't have the best chance of winning in the East every day.

Fluctuations would occur. They don't. You're a statistician. Explain away two teams with almost identical data being input, assuming that data is ONLY that that is used to determine Power Rankings, having such consistently different odds coming out at the end. Can I understand the Lakers always having a higher percentage chance? Sure. They've had better stats up until Saturday. But what I can't understand is how the other team consistently has what can be considered poor comparative output.

I just think he's inputting more data than he is admitting to. Team efficiencies, rebounding rates, turnover rates, eFG% and eFG% against, past winning percentage against every team, etc. If he is doing that, then the Celtics lower numbers make all the sense in the world. But he isn't saying that.

As for the crack about me being a mathematician, I have my BSME and took enough calculus, applied mathematics and statistics courses to call myself a mathematician, IMHO. I might not have an MS in Applied Mathematics or Advanced Algorithms but I still think I have enough mathematical knowledge to question what I am seeing in his output.

I would love, like you, to see the engine he is working with and the full extent of the data being input though.

[Fafnir]

So now you are defending him by breaking age data down to the ridiculous to prove a point? Do you really believe in order to come up with these odds in a Monte carlo methodology he input data down to the final detail of average age per minute played?? And if you do, then you prove my point, he's inputting data he isn't admitting to.

I'm not arguing that, I was merely rebutting you point about Orlando.

As for the regression to mean, I understand the concept. I also understand such regressions and their equal and opposite aggressions away from the mean, would mean that on a day to day basis, other teams with similarly excellent data being input would have inconsistencies. The Lakers would not consistently always have the best chances of winning. Orlando wouldn't have the best chance of winning in the East every day.

How many data points have we actually seen Nick? Have you been checking the power rankings daily to see how the C's odds fluctuate? Are you looking at all of the projections or just the "championship" odds.

You're arguing that Hollinger has to be forcing his model because the C's are 2.4% below Orlando when the same data has the Knicks winning 2.4% of time! You and I both know the Knicks have a 0% chance of being the last team standing.

Its pretty clear to me that the std deviation of the championship odds is pretty high. Its not like this model hates the C's compared to the Magic/Cavs. The C's have consistently been picked to have a better record! So clearly the power rankings do correlate pretty well with overall record. Take a look at the teams, they line up by projected record to current power rankings very closely. SAS are a bit low, but considering they're currently two games behind the two teams ahead of them and have nearly identical power scores that's not anything that's too suprising.

Considering that Hollinger directly addresses your complaints in his post, and has nothing to gain from lying.

As a reminder, this tool is completely, 100 percent automated, so my obvious, long-standing bias against your favorite team is not a factor here.

As always, the output of a product is only as good as its input, so let's explain a little about how this is derived. The computer starts with the day's Hollinger power rankings. Then, in each of the 5,000 times it replays the season, it makes a random adjustment up or down to allow for the possibility that a team will play better or worse than it has done thus far. (I call this the Anti-Dennis Green Postulate; i.e., maybe they aren't who we thought they were.)

Additionally, the results regress to the mean. This is more important early in the season, and what it essentially means is that even though a team might start 10-0, it's not necessarily bound to go 82-0. The effect of this will reduce sharply after the first quarter of the season or so, but in the early going of most seasons

Your claim that regressing to the mean should hurt the Lakers as much as the C's though isn't necessarily true. We have no idea what sort of regression he uses.

As for the crack about me being a mathematician, I have my BSME and took enough calculus, applied mathematics and statistics courses to call myself a mathematician, IMHO. I might not have an MS in Applied Mathematics or Advanced Algorithms but I still think I have enough mathematical knowledge to question what I am seeing in his output.

I do have to apologize for that crack. I'll be frank Nick and say I don't have a high opinion for your mathematical/stat bona fides. Our previous conversation on basketball statistics and the repeated basic misconceptions you had in that thread on possession based calculations among other things still stand out to me. But that's beside the point I shouldn't have put that in my post, I won't do that again.

[nickagneta]

To be honest....it's been 20 years since I was in a statistics class, so I might be rusty.  Also, our last discussion was based on what certain stats meant regarding their interpretation to real time basketball. Put 20 coaches with their personal stats experts in a room and you'll probably get 20 interpretations of the "sabermetric" basketball stats and what they mean.

[Fafnir]

To be honest....it's been 20 years since I was in a statistics class, so I might be rusty.  Also, our last discussion was based on what certain stats meant regarding their interpretation to real time basketball. Put 20 coaches with their personal stats experts in a room and you'll probably get 20 interpretations of the "sabermetric" basketball stats and what they mean.

Oh I understand trust me, if you don't use it you lose it. After two years away from school I forgot so much... As for NBA GMs and Coaches, we both know how dumb they are! Look at Jared Jeffries, Diop's, and Gilbert Arenas's contracts!

I say we table this discussion until we see some more "playoff odds". If my theory about the regression to the mean factor is right it should be fading out of significance soon. TP for the debate.