Do pyth records correct themselves in-season?

by Imperialism32 on Jun 16, 2009 6:34 PM EDT

Depending on who you ask, the Rays are either 4 or 5 games under their pythagorean record entering play on June 16th. Playing below a pyth record is generally attributed to luck, though I'm sure people smarter than me have other reasons(for instance, a poor bullpen can often cause a team to play below it). My question is: Does this luck "normalize" over the course of a season? That is, is it fair to say "The Rays have played better than their record, their luck will even out the rest of the year?"

The best way I could figure to test it was to see if first-half pyth record had any correlation to second-half pyth. I first thought to test at June 16th and compare it to end-year data, but obviously there are more games played in the June 16-end year period and more game ostensibly leads to more variance in pyth records.

So. To test this, I took data for the previous 5 seasons(2004-2008), using X WL on MLB.com's standings. First I checked how teams were playing on June 30th(which is a rough estimate of mid-season. This year the season started later so 2009's midpoint is sometime in early July) and compared it to how they did at the end of the year. For example, in 2008, the Rays were +3 over their pythagorean record on June 30th, and +6 at the end of the season. Thus, they played 3 games their pythagorean record in both the first and second halves of the season. Let's see if everyone was this consistent.

via i39.tinypic.com

Note that the extremes(8 games over and under) are 1 case and the 6/7 ones have very limited data

I found it somewhat surprising that on average, teams who played 3, 2, or 1 games over their pyth and 1, 2, or 3 games UNDER their pyth all played worse(in regards to their pyth) in the 2nd half. Of course, playing even in the 1st half doesn't fit under that. That's probably just statistical noise and it basically tells me 5 years isn't enough of a sample.

Basically what I figured coming in is that pyth records are part luck, but what part depends on each team. Has any research been done into this? I'll probably go back further than 5 years when I have the time.

This post was written by a member of the DRaysBay community and does not necessarily express the views or opinions of DRaysBay staff.

Tweet Comment 11 comments | 0 recs |

Do you like this story?

Comments

I forgot to mention

Average really isn’t a great way to judge this, but it’s quick and dirty. What is tells me particularly for the -1 to -3 range is that there’s a pretty even distribution of teams that player higher and lower in the 2nd half.

www.raysprospects.com

by Imperialism32 on Jun 16, 2009 6:38 PM EDT reply actions

Very interesting...

Did you notice if the quality of the team has any effect? It seems that higher scoring teams would be more likely to win less actual games where low scoring/great pitching teams would out produce? Does the data reflect that at all?

by tallyray on Jun 16, 2009 6:44 PM EDT reply actions

I don't think it's so much that the pythagorean evens out

As that it will ‘catch up’ with the expected output going forward. If a team has a .450 pythag, and a .500 actual record, then it’s fairly likely that the team will go somewhere close to .450 the rest of the way. But the gains (or losses) are locked in – there’s no catch-up, except on a rate basis.

Bad Left Hook - The SB Nation boxing blog
"Baseball is played on the field, not on a calculator."

by Brickhaus on Jun 16, 2009 8:24 PM EDT reply actions

Sounds like the gambler's fallacy.

by R.J. Anderson on Jun 16, 2009 8:41 PM EDT up reply actions

?

I don’t think so. I think I’m saying exactly the opposite actually.

Assuming that your pythag is normal, the reason there tends to be regression is because it has a correlation with future success.

In a gambling context, what I’m saying is that just because you’ve flipped heads 5 times in a row, you still have a 50% chance of hitting tails, so of course there will be regression, but it’s just a function of what happens on a going-forward basis. The fact that you hit heads 5 times in a row doesn’t mean that you’ll bounce back.

Bad Left Hook - The SB Nation boxing blog
"Baseball is played on the field, not on a calculator."

by Brickhaus on Jun 17, 2009 12:49 AM EDT up reply actions

I meant this part:

But the gains (or losses) are locked in – there’s no catch-up,

Which is what most people fall into. Just because a .500 true talent team went 3-7 during a stretch doesn’t mean they’ll go 7-3 at some point to “even out”.

by R.J. Anderson on Jun 17, 2009 1:05 AM EDT up reply actions

And that's where most people have trouble with regression...

by tallyray on Jun 17, 2009 7:55 AM EDT up reply actions

Yes, we're in agreement

Just different ways of saying the same thing. Maybe I wasn’t clear.

Bad Left Hook - The SB Nation boxing blog
"Baseball is played on the field, not on a calculator."

by Brickhaus on Jun 17, 2009 1:48 PM EDT up reply actions

I probably should've pointed out which part I meant.

by R.J. Anderson on Jun 17, 2009 2:05 PM EDT up reply actions

Pythagorean records tend to correct themselves the next season

I’m sure if you looked at Pythags mid season for teams, and compared them to the end results, you’d see some variation. But as Brickhaus points out, the results often lock in.

Where you’ll often see some correction is in subsequent years. A team can significantly outperform their pythagorean win expectancy for a full year, but it rarely happens the next year. The 2008 Angels, for example, finished 12 games above their Pythag (100 actual wins to 88 expected wins). The 2009 Angels are 33-29, or two games above their Pythag.

"It's just a tiny little nick, but it hurts when I get champagne in there."
- Jason Bay, on getting spiked scoring the winning run in ALDS Game Four.

by 0157H7 on Jun 16, 2009 10:39 PM EDT reply actions

I think the key is whether the current Pythag record is a good indicator of future performance

We know what their record should be in the past, but if the team is expected to keep up its current pace then it is more likely that the W/L’s would be more in line with the Pythag numbers than the real life numbers (going forward).

by matthan on Jun 17, 2009 12:05 AM EDT reply actions