Reply With Quote
Hooligan
Kevin Garnett injures knee. Out for season. How you like your numbers now Hollinger? LOL
Hooligan
i see no problems with his odds
I've tried to put these kinds of regressions together before. You really have to have a well-validated model, and a mountainload of data, or it's GIGO.
I always knew Hollinger was a tool. But unless I'm misreading his odds, he's a broken one, too. Somebody please tell me I'm not reading this right, because if he's making a living doing this, I'm changing careers tomorrow.
He's got the Pistons and Orlando both 100% GUARANTEED to win their divisions, the presence of the Cavs and Heat notwithstanding. One-hundred-percent total confidence. Boston gives up a one-in-a-thousand shot to the Raptors, but it's a mathematical impossibility for Cleveland to win more regular-season games than Detroit.
And even though they would have to get through the Boston juggernaut to even make the finals, the Pistons have better odds of winning a le than anyone from the West. AND SO DOES ORLANDO. A Spurs repeat is less likely than the Magic getting past Boston and Detroit on their way to a le. LOL is right...
Man, Hollinger thanks.
I needed a good laugh today.
Hollinger should stick to fantasy basketball.
I would be embarrassed to post this if I were Hollinger. Sad.
/signed
Here:
The Playoff Predictor: Explaining how it (doesn't) works
He tries to save his face explaining how useless his new toy is.
Pathetic
We have a new word:
Hollinge (v.): to put together exceedingly complex mathematical models whose predictions consistently fail to be confirmed by simple empirical observation.
I think Hollinger designed the BCS ratings system."Three things that I hate most in life. Lies, damn lies and statistics" Disraeli - English PM
It can be statistically proven true that the Mississippi River at one time in its life was over a million miles long. -Mark Twain
Anything can be proven "true" with numbers. All you have to do is get them in the right order and they will "prove" anything.
What a joke.
Hollinger is the best, he gave me the chance of getting nice contract from the well known financially conservative team in south Texas.
Sincerely,
Jackie Butler
.
.
.
I'm going to do something I've never in my life done before, and probably never will do again.
I'm going to sorta defend Hollinger on this one. A little bit, anyway.
Mainly because timvp attacked him first.
No, but seriously. Without knowing anything about the model, you have to cut it slack because it's still very early in the season. The model is only as good as the data you give it. If you gave the most perfect model possible data from only the first couple of days of the season, it's going to spit out numbers based on that data, and tell you that the New Orleans Hornets rank highest, or whatever.
Okay, now the attacking part.
What he should have done was applied the model to past seasons, and determined the minimum number of games that typically need to be played before the data isn't garbage. And maybe he did that, but he doesn't seem that savvy. But that's not the really bad part.
The really bad part is that I am reasonably sure that you are not supposed to use regression methods (like he did) to predict who should win a championship. The reason, as I explained in a previous thread (which I can't link because search is down) is that regression is for drawing correlation to independent outcomes.
Like, suppose I looked at all the people in the world who could dunk, and all who couldn't, and measured all the factors that distinguished dunkers from nondunkers. Like height, weight, age, gender, shoe size, country of birth, etc. I could think develop a regression equation and use it to predict whether bdubya could dunk or whether wildbill2u could dunk or whether timvp could dunk (). But these are all independent, that is, whether or not bdubya can dunk plays no factor in whether wildbill2u can dunk (unles wildbill2u is laughing so hard after watching bdubya fail that he has no energy to jump).
That's not true when you have outcomes that are dependent on one another. If instead I was trying to find an equation that will predict who can jump the highest, that's a completely different modeling question, because only one person can win. And that's what you're trying to model when you try to predict a champion (meanwhile, predicting who's going to make the playoffs is something that falls between the two extremes). This is a problem I've been off-and-on trying to generate a solution for over the last couple of years, and I think I have a method, but I need a real statistician to help.
I can't dunk.
Hollinger swears that he's introducing this concept, but I could swear that he did the same thing towards the tail-end of last season. I tried to find something to support that, but ESPN's archives on Hollinger (as best I can tell) only go back to mid-May.
I'd be curious if anyone could find that model -- if I'm remembering correctly -- so we could how good it was last season. Of course, it's much, much, much easier to assess which teams are likely to win les in mid April than in early December, faltering 67-win teams aside.
Edit: I did at least find this, which was Hollinger's odds to win the le in 2007 (though it appears that he may have used a different methodology). That methodology made the Spurs the odds-on favorites to win last year.
Here are Hollinger's odds from 2006, too.
Last edited by FromWayDowntown; 12-06-2007 at 10:04 PM.
If you run your model in December and it gives you a 100% certainty that a given team will win its division, I can tell you right now there's a bigger problem than the dataset.
And I can dunk. Just need a leg up and a low rim. And a few tries. And no defender.
I guess I'm in the minority, but I like Hollinger's analysis. The key is to not think of it as a prediction, but rather as a statement of what has happened so far. For example, it's not unreasonable to give Boston such odds of winning, if you think of it only as a reflection of what has happened so far.
Once you think that way, then it can reveal some interesting insights.
it's a neat tool. not very useful this early in the season, but entertaining nonetheless. i don't see anything to get upset over.
he explains, "this is a tool that becomes more valuable the later in the season we get, because that's when schedule discrepancies between playoff contenders really become important."
once the celtics are exposed as being overrated thanks to their charmin-like schedule, i'm sure their %'s will fall back to earth.
Yeah, the "100%" (again, no idea how he calculates that) is probably just another indication that at core, Hollinger doesn't really understand what he's doing.
In the last 4 years the chances of a team from the east winning a championship was 100% after the spurs won a le.
even his explanation is fubar. he states that if orlando comes out of the east, with the way theyve been playing, they will be favored over whoever in the west is... even the spurs and suns who have both beaten them.
Not to beat a dead..ok, I'm beating a dead horse. But WRT your point about the outcomes being interdependent , when I heard about this, I figured Boston's great odds were based on the relative weakness of the East making them much more likely to be in the Finals than any one team out of the West. But he's also got Orlando and Detroit as better bets to win a le than any West team. I just can't figure that whole package.
While it's still laying there, dead . . .
Another mistake that probably explains these results is that Hollinger used an equation with coefficients fitted to stats over a complete 82-game season. If you really want to predict who's going to do what based on 20 games, then he should have gone back into each year, looked at just the first 20 games, and fitted his coefficients to that data.
In other words, the coefficients should change based on how far you are into the season. The coefficients this early in the year should have much greater uncertainty built in to the than the end-of-year coefficients.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
