Regressing to the mean

[Heinz57]

I was thinking about something and wondered what any of you reading think about this. When people use statistics to project outcomes for games they try and use the most current and if that sample is not large enough will look at historical as well. But certainly current form outweighs what a player did 2 years ago. However there is a simple truth that players who are far surpassing their historical norms generally regress to the mean at somepoint assuming they aren't young and improving still. So here's my question: If you value how a team (which is composed of players) based on how they have been preforming recently, but know in general that these overachievers should return to earth at somepoint is that something that can be weighed? It would seem the two counteract each other but are both true...Sorry if this was confusing thanks for reading through it I'm interested in any thoughts you all have.

[sportypicks.com]

Heinz,

I'm not 100% sure but I'm guessing the gist of your question is how much weight to apply to recent data in order to accurately project future outcomes?

There are probably endless possibilities when trying to do this. Are yesterday's outcomes more correlated to future results than outcomes from 2 days ago? At what point does data not become recent? Does this very by player/team?

I don't have the answers but one possible way would be to use "credibility" weighting where the number of samples needed for full credibility is determined by the mean&variance of the variables you are trying to predict. There is probably alot of work involved with this, but it is one possibility.

[Heinz57]

Thanks, yeah you hit the gist of the question thanks for expressing it more clearly. I was just wondering how the to variables played into each other given they point in different directions. A credibility weighting is a pretty good idea.

[bugstud]

so you're wanting to fade the white sox for the rest of the year?

[Heinz57]

Heh, certainly not! I'm simply wondering where the balance between current form and the tendency to regress to the mean is. I live in Cleveland and like the White Sox to win the central, unless Minnesota really gets it together or Chicago implodes. Because the Tribe is not gonna do it. KC is out already in my opinion and DET is playing well but not going to be there for the long run.

[DougOzzzz]

[ QUOTE ]
Heh, certainly not! I'm simply wondering where the balance between current form and the tendency to regress to the mean is. I live in Cleveland and like the White Sox to win the central, unless Minnesota really gets it together or Chicago implodes. Because the Tribe is not gonna do it. KC is out already in my opinion and DET is playing well but not going to be there for the long run.

[/ QUOTE ]

If you believe that Chicago is likely to win the division, then the answer to your original question is: Recent results are not nearly as important as you think they are.

The team has an OBP of .300. That is a TERRIBLE number, even for just 22 games. Their pitching performances are simply not going to hold up. Their K/BB ratio is actually BELOW average - maybe their fielding has been absolutely dominant, but more likely their opponents have just been hit unlucky.

They've also:
- Won more games than their record suggests (16 vs. 13.8).
- Scored more runs than their statistics suggest (100 vs. 93).
- Allowed fewer runs than their statistics suggest (93 vs. 81)
- Faced weaker than normal opponents (probably not important, since the whole division has it relatively easy)

You give up on the Indians WAY too easily. Their adjusted record is 10.6-9.4 vs. 12-10 for Chicago. Chicago has a head start, but I'd say the Tribe are 40% to pass them, and Minnesota is about 75% to do so. Can we say 2003 Royals all over again?