View previous topic :: View next topic |
Author |
Message |
Ed Küpfer
Joined: 30 Dec 2004 Posts: 616 Location: Toronto
|
Posted: Sat Jun 02, 2007 11:35 am Post subject: |
|
|
DLew wrote: | What Justin is trying to say is that the point of a model generally is to try to predict the future, not describe the past. |
That is not necessarily true. A model can be purely descriptive, and that is perfectly fine. There's nothing wrong with using a model to "retro-dict" past observations with no reference to future results. The regression coefficients can have inherent interest.
wiLQ, what Justin is pointing out is the danger of overspecification. The greater you tune your model precisely to past results, the more error your model will have in the future. Random error is implied in every regression model, although it's not always made explicit. It's worth keeping that in mind. _________________ ed |
|
Back to top |
|
|
DLew
Joined: 13 Nov 2006 Posts: 57
|
Posted: Sat Jun 02, 2007 11:45 am Post subject: |
|
|
That's why I said generally (as opposed to always). I guess it's somewhat unclear what he is trying to do here. |
|
Back to top |
|
|
wiLQ
Joined: 23 May 2007 Posts: 22 Location: Poland
|
Posted: Sat Jun 02, 2007 11:47 am Post subject: |
|
|
Mike G wrote: | You could almost certainly explain all mvp's with one formula, if you include height and skin color. But then you might get results saying Stockton 'should have' won a few. Or maybe not, since his teams never improved dramatically. Tops was 9 wins, from '96 to '97; Malone won that year. |
According to MVP Points Stockton was very underrated in voting. He should have been ahead of Malone in some cases but he wasn't.
But, what is important, that was taking place outside of the Top2 and he never was close to actually win that award.
Mike G wrote: | Some people still swear Kidd should have won in '03 -- based almost entirely on the Nets' improvement. In a best-of-7 Finals, was he anywhere near the best/valuablest player? Was Iverson the b/v in the '01 Finals? What did they have that Shaq/Duncan didn't have? -- Shortness |
That's why I really like that argument. It looks legit in 3 cases. It could have helped Iverson once and Nash twice.
Mike G wrote: | I don't remember -- is incumbency a negative in your formula? |
No, because I wasn't sure that it has negative impact. _________________ regards
wiLQ |
|
Back to top |
|
|
wiLQ
Joined: 23 May 2007 Posts: 22 Location: Poland
|
Posted: Sat Jun 02, 2007 12:31 pm Post subject: |
|
|
Ed Küpfer wrote: | what Justin is pointing out is the danger of overspecification. The greater you tune your model precisely to past results, the more error your model will have in the future. |
I do understand that fact.
But as you can see in this discussion I'm not asking what I should throw out from formula so I'm not creating it from the scratch. The fundamentals are fine because they still work in 20 out of 21 cases.
I'm trying to identify and add those unexpected things that happened in 2006 because I simply wasn't aware of them before.
For example, Billups looked like a great candidate for MVP and he finished only fifth in voting. Why? Because he was playing alongside three All-stars and it really hurt his chances.
IMO with this knowledge I should add new rule: "playing alongside three All-stars = - X MVP Points".
Do you think that if I do that, formula will work less accurate in the future?
IMO it will work in the exactly same way but when other player will be playing alongside three All-stars I will know how many MVP Points I should deduct from his total.
What's more, why do you assume that Nash's MVP in 2006 is an outlier while it can be a sign that formula is not accurate enough? _________________ regards
wiLQ
Last edited by wiLQ on Sat Jun 02, 2007 1:19 pm; edited 1 time in total |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 508 Location: Columbus, OH
|
Posted: Sat Jun 02, 2007 12:55 pm Post subject: |
|
|
Ed Küpfer wrote: | DLew wrote: | What Justin is trying to say is that the point of a model generally is to try to predict the future, not describe the past. |
That is not necessarily true. A model can be purely descriptive, and that is perfectly fine. There's nothing wrong with using a model to "retro-dict" past observations with no reference to future results. The regression coefficients can have inherent interest.
wiLQ, what Justin is pointing out is the danger of overspecification. The greater you tune your model precisely to past results, the more error your model will have in the future. Random error is implied in every regression model, although it's not always made explicit. It's worth keeping that in mind. |
Yes, what Ed said.
I've given my 2 cents, so I'll back off now. _________________ Regards,
Justin Kubatko
Basketball Stats! |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 880 Location: Durham, NC
|
Posted: Mon Jun 04, 2007 6:04 am Post subject: |
|
|
wiLQ wrote: | I'm trying to identify and add those unexpected things that happened in 2006 because I simply wasn't aware of them before.
For example, Billups looked like a great candidate for MVP and he finished only fifth in voting. Why? Because he was playing alongside three All-stars and it really hurt his chances. |
That's basically what Justin and Ed were describing as overspecification. Every year you'll have to add a new "rule" because you'll have discovered something else that seemingly didn't factor into the equation the year before. After 20 more years you'll have (at least) 20 more rules and what will you do with them and will your model still be tractable?
wiLQ wrote: | IMO with this knowledge I should add new rule: "playing alongside three All-stars = - X MVP Points". |
And won't that affect past results? Or will you only apply it to this current year?
wiLQ wrote: | Do you think that if I do that, formula will work less accurate in the future? |
Yes. |
|
Back to top |
|
|
HoopStudies
Joined: 30 Dec 2004 Posts: 533 Location: Near Philadelphia, PA
|
Posted: Mon Jun 04, 2007 6:59 am Post subject: |
|
|
For what it's worth, the overspecification issue Justin et al have raised is legit if we're trying to understand something rational and logical. If you look at it cynically, the MVP vote may not be that way. If you have to add new rules to understand what they're doing, it is possible that the voters are making up new things to justify their voting. _________________ Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com |
|
Back to top |
|
|
Mike G
Joined: 14 Jan 2005 Posts: 1509 Location: Delphi, Indiana
|
Posted: Mon Jun 04, 2007 7:53 am Post subject: |
|
|
HoopStudies wrote: | ... If you have to add new rules to understand what they're doing, it is possible that the voters are making up new things to justify their voting. |
Well, the apple cart was upset by the sudden prominence of Steve Nash; this was a new phenomenon -- small white player leads suddenly-best team. Such rare events -- actually, combinations of events -- don't happen every year (as Gabe seems to suggest). Outliers should become ever rarer.
Eventually, voters might catch on that they've been figured out. Then they might actually vote for the most valuable player, rather than for the most 'electable' one. _________________ 40% of all statistics are wrong. |
|
Back to top |
|
|
Harold Almonte
Joined: 04 Aug 2006 Posts: 224
|
Posted: Mon Jun 04, 2007 8:22 am Post subject: |
|
|
Does all this stuff means that if you try to "over-specificate" a lot (using an analogy), a lot of inputs (more than just boxscore stats), and try to build a more complex formula, then the regression would give you wider error margins for prediction that you couldn't fix even with adjusts? Is it to try to accurate predict 21 for 21 too much expectation for statistics, is it unreal?
Could be that why WP didn't ever take that inconvenience? Are different "logical and rational" things? |
|
Back to top |
|
|
wiLQ
Joined: 23 May 2007 Posts: 22 Location: Poland
|
Posted: Mon Jun 04, 2007 8:55 am Post subject: |
|
|
gabefarkas wrote: | Every year you'll have to add a new "rule" because you'll have discovered something else that seemingly didn't factor into the equation the year before. After 20 more years you'll have (at least) 20 more rules and what will you do with them and will your model still be tractable? |
But we are talking here about really rare event. Four All-stars from one team... it happened only 7 times in history of that league. What's more, it happened only 2 times in the last 2 decades... I could have omitted such impact, right?
However, I agree with you on one thing: if I have to add new rule(s) every year to get a perfect match, I will have to admit that idea just isn't working. But MVP voting in 2007 really gave me a lot of new hope.
gabefarkas wrote: | And won't that affect past results? Or will you only apply it to this current year? |
Beside Billups it will affect exactly one past result but in a positive way.
After comparing MVP Points to actual voting in 1997/98 you can see that Shaq overachieved... and he was playing alongside three All-stars... so adding that rule will help move him down where he should have been.
gabefarkas wrote: | wiLQ wrote: | Do you think that if I do that, formula will work less accurate in the future? |
Yes. |
So do you think that next time when a player will be playing alongside three All-stars it won't hurt his chances the way it hurt Billups? _________________ regards
wiLQ |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 880 Location: Durham, NC
|
Posted: Tue Jun 05, 2007 6:54 am Post subject: |
|
|
wiLQ wrote: | gabefarkas wrote: | Every year you'll have to add a new "rule" because you'll have discovered something else that seemingly didn't factor into the equation the year before. After 20 more years you'll have (at least) 20 more rules and what will you do with them and will your model still be tractable? |
But we are talking here about really rare event. Four All-stars from one team... it happened only 7 times in history of that league. What's more, it happened only 2 times in the last 2 decades... I could have omitted such impact, right?
However, I agree with you on one thing: if I have to add new rule(s) every year to get a perfect match, I will have to admit that idea just isn't working. But MVP voting in 2007 really gave me a lot of new hope. |
Every year will be a new rare thing. This year it's the small white player, next year it's a guy who somehow gets 5 steals per game, the year after LeBron puts up MJ's '88 stats almost exactly, the year after that Kevin Durant learns to pick his nose while dribbling. No one is able to predict the future with 100% certainty.
wiLQ wrote: | gabefarkas wrote: | wiLQ wrote: | Do you think that if I do that, formula will work less accurate in the future? |
Yes. |
So do you think that next time when a player will be playing alongside three All-stars it won't hurt his chances the way it hurt Billups? |
I have no idea. |
|
Back to top |
|
|
Harold Almonte
Joined: 04 Aug 2006 Posts: 224
|
Posted: Tue Jun 05, 2007 7:55 am Post subject: |
|
|
Quote: | Every year will be a new rare thing. |
Basketball and MVP is not that diversity. It's the opposite, a routine, and ocassionally it happens something that seems to be outside the parameters. If a formula for subconcious racial biased refs can exist, and everybody blindly agree, why not measure an ocassional subconcious variable in MVP prize? Isn't even said by some people that the scoring starring, and its reflex on some metrics is actually a subconcious thing? |
|
Back to top |
|
|
wiLQ
Joined: 23 May 2007 Posts: 22 Location: Poland
|
Posted: Wed Jun 06, 2007 4:55 am Post subject: |
|
|
gabefarkas wrote: | Every year will be a new rare thing. |
So you are not convinced that there may be a pattern behind this?
20 out of 21 MVP's described in the same way is not enough to show you that's not about rare things?
Well, I see it differently...
gabefarkas wrote: | This year it's the small white player, next year it's a guy who somehow gets 5 steals per game, the year after LeBron puts up MJ's '88 stats almost exactly, the year after that Kevin Durant learns to pick his nose while dribbling. |
MJ's '88 MVP is included in results of my formula...
gabefarkas wrote: | No one is able to predict the future with 100% certainty. |
That's true, but I guess you don't want to remove, for example, weather forecast from your life? _________________ regards
wiLQ |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 880 Location: Durham, NC
|
Posted: Wed Jun 06, 2007 6:16 am Post subject: |
|
|
wilQ, you're the one who believes something new happened in 2007, not me. You're trying to explain it away with a new parameter in your model.
Secondly, I never look at the weather forecast, to be honest. Since when has a weatherman been able to predict the weather? |
|
Back to top |
|
|
Ben
Joined: 13 Jan 2005 Posts: 202 Location: Iowa City
|
Posted: Wed Jun 06, 2007 4:38 pm Post subject: |
|
|
How big of an anomaly was 2006? Wasn't Nash an anomoly in 2005 too? If I recall correctly, if he got no mvp points for leading the league in the assists, a different player on Phoenix would have won the MVP point race. Also, leading the league in assists never provided the margin of mvp point victory for another player, thus it might as well have been +x points for being Steve Nash. (I could be wrong , but I remember thinking something along these lines at the time.) |
|
Back to top |
|
|
|