Regression to the mean

by Peder on Apr 14, 2006 8:58 AM CDT Comment 6 comments

Almost all of the sabermetric previews of the White Sox have noted how they overacheived their pythagorean expected victories last year. Most have added that teams that exceed their pythag will return to earth the next year. Whether that means winning fewer games than their pythag should allow or be back to even isn't clear to me. But I was curious as to how true that really is.
Last year I put together pythagorean standings for the past decade or so. (If anyone knows of historical pythagorean results somewhere on the web, please point me towards it!) I decided to look at only teams that exceeded expectations by 5 or more games. These should be the truly lucky teams. What I found suprised me.
From 1994-2004 there were 41 teams that qualified. Of those 41, 19 played above their expected rate, 5 played even and the other 17 played below. As a whole, the 41 finished 9 games higher than expected.
From this admittedly small analysis, it seems that there's no special reason to expect a large swing of fortune for the White Sox.

SouthSideSox is a community driven site. As such, users are able to express their thoughts and opinions in a FanPost, such as this one, which represents the views of this particular fan, but not necessarily the entire community or SouthSideSox editors.

0 recs | Comment 6 comments

Story-email Email Printer Print

Comments

Hmm...

I've always found the "back to earth" statement when it comes to pythagorean records to be odd. The previous year should have no impact on exceeding or being below the current year's pythagorean record.

The records in and of themselves are supposed to be the indicators without something like karma (lucky one year so it'll come back to bite you the next) factored in.

So good point there Peder.

AIM: BrentBrookhouse

by Brent Brookhouse on Apr 14, 2006 9:12 AM CDT reply actions 0 recs

I've tried to make this point before

but you did it better and more thoroughly. That said, they had key over-acheivement in the bullpen (Cotts, politte, Hermanson Jenks) which ended up being their strength thru the regular season. They may not win as many games but if the bullpen is close to last year they'll make the playoffs.

by dyspeptic on Apr 14, 2006 10:53 AM CDT up reply actions 0 recs

Regression Effect / Regression Fallacy

Here is a great article that explains "regression to the mean"

http://www.tufts.edu/~gdallal/regeff.htm

by Tybor on Apr 14, 2006 12:52 PM CDT up reply actions 0 recs

appreciated

In fact, without putting apen to paper, the regression fallacy might suggest to one that the extrordinary performance in the bullpen will tumble so far back as to more than offset then upwards regressions by those with slightly below averages performances

of course the mean is not destiny as demonstrated by a young guy developing or a an older one one the way out

by dyspeptic on Apr 14, 2006 6:21 PM CDT up reply actions 0 recs

It does show

that some performances aught not to be counted on. A GM who does not realize that and stands pat will likely pay the price.

Comonnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn You White Sox!

by zokmaad on Apr 15, 2006 9:43 PM CDT reply actions 0 recs

My thoughts

Regression to the mean, like OBP, seems to be every "newbies" first find upon stumbling into sabermetrics--and they will glady recite it without really having any understanding of why it happens, and why sometimes it's best to to count on it.

I see "regression" all over the internet being used to discount young players' improvement--which is misguided and silly.

Analagous to this is when "pythagorean" is used to argue a point, when then individual performances behind that RS/RA number are never addressed.

In sum, yea, it's a useful argument, but only when put into a greater context--like most things I guess.

by madvillian on Apr 16, 2006 12:48 AM CDT reply actions 0 recs