THE BOOK--Playing The Percentages In Baseball

I have not studied this at all so please file what I am about to say under pure conjecture.  I think what David might be measuring is that players with 5 or more years playoff experience may be better players than those with less experience, and that when better players play against better players their performance falls off less than when not so good players play against better players.  Both of my assumptions can be tested easily.

When I come across these potentially age-based impacted studies, I simply use 29-and-under compared to 30-and-over.  This way, we’ll get most of the 5+ years of playoff experience in the 30+ group, and those with under 5 years will also be 30+ years old.

I can certainly believe that age, rather than playoff years, is what you want.  In The Book, I was able to show that young players, with a runner on 1B, were not that good. 

And we also know that brains/wisdom has a positive aging slope (see: walks, for both pitchers and hitters).

So, anything that requires more than typical use of your brains is good if you are old, and it would seem dealing with the playoffs would require extra brain power.

Pure guess of course.

Peter, I realize you said it was conjecture, but I don’t think that the drop-off versus better (or worse) competition is anything but linear as a function of the level of talent/performance.

I showed in The Book that the good pitching v good hitting is a wash.  So, I don’t buy that one.  But, maybe in the playoffs, it’s different.

David finds that players with a lot of post-season experience see a smaller decline in performance in the playoffs than do those with little experience.  It seems to me this could be a function of regression to the mean.  Post-season teams—and thus post-season players—have been luckier than average in the regular season.  So some of their playoff decline is just regression to the mean (most of it is of course facing better opponents).  However, a player with 5+ years playoff experience is more likely to in fact be a high true-talent player.  So we’d expect less mean reversion in the playoffs than average from these players.

Just a theory, of course.  If David showed us not only the wOBA decline for each cohort, but also those players’ regular season and career rates (or a Marcel), we could see if the inexperienced players over-performed their true talent more than the 5+ year players.

Hmmm. I’d have to go back and look at his numbers and see if your theory makes sense and would/account for some or most (or all) of the difference.

I think the theory is right.  Imagine we take a pool of .300 regular season hitters.  We would clearly find that they hit significantly lower the next year, on average.  But if we look only at those hitters in the group who have hit over .300 five or more seasons, surely they would regress less than the overall group.  The principle is the same here:  those with a lot of post-season experience are very good players and/or are on very good teams, so their presence in this year’s playoffs is less likely to be a function of luck. 

But I don’t know if this could account for most/all of the difference David found, which was .012 wOBA.

#2 You might have it here with the age vice experience.  Scoring is less in the post season, mainly do to colder weather. Players that rely on hitting, younger players, will have less hits (e.g. a flyout that would have been a HR in warmer weather) your older veteran will still be taking his walks thereby keeping his wOBA up.

#8, interesting proposition.

Guy, I looked back at the article.  David compares the difference between ONE regular season and that same post-season for experienced versus non-experienced players.  There should be no regression issue there.  He is not looking at the difference between their career regular season and that post-season, I don’t think.  If he were, then your point would be valid.  But I don’t think he is, although it is not 100% clear.

In other words, every player in his sample, regardless of their post-season experience, is on a post-season team during the period in which he culls the data.  And he only culls the regular season and post-season data from that one year. 

For example, we have 10 players with 5+ years of post-season experience (before the season in question).  Their collective wOBA in the ONE season he is looking at is .380.  They collectively hit .370 in that same year.  There is going to be a little bit of regression because he is “cherry picking” players from playoff teams and playoff teams got lucky, and therefore the players on those teams got lucky.

Then he looks at 10 players who are also on playoff teams in year X but have less (say, less than 2) prior playoff experience.  They also hit .380 in year X, but they hit .360 in the post-season in year X.  They also regress for the same reason, but they will regress exactly the same amount.

The only caveat to my statements, and I have to look at the data again, is that if the players in group II (less playoff experience) were .390 for some reason (and there is no reason for them to have a higher regular season wOBA than the players in group I), then they would regress more simply because the further you are from the mean, the more you will regress.

Now, it could be that the players who had more playoff experience were actually better players, such that they are regressing toward a higher mean.  That could explain David’s results without concluding that players with more post-season experience do better in the post-season.  I’d have to look at the data again, and David would have to look at each group’s career or at least prior season’s wOBA to see if that might be true.

MGL:  I think I’m clear on David’s method.  My point is that we have additional information on the players with 5+ years of post-season experience, i.e. that they are (often) very good players.  I would expect their reg. season performance to reflect less luck than the less-experienced players.  That doesn’t necessarily mean they are better players, though they might be.  But the gap between their current regular season performance and true talent should be narrower, resulting in less regression in the post-season. 

Look at it this way:  we might find that 93-win teams regress to 86 wins the following year, on average.  But if I only look at those 93-win teams which had also won 90+ in the two preceding years, we expect to find less regression.  I think the same principle applies here.

My point is that we have additional information on the players with 5+ years of post-season experience, i.e. that they are (often) very good players.  I would expect their reg. season performance to reflect less luck than the less-experienced players.  That doesn’t necessarily mean they are better players, though they might be.  But the gap between their current regular season performance and true talent should be narrower, resulting in less regression in the post-season.

Yes, that is exactly what I said in the last paragraph of my last post.

Close, but not “exactly.” It doesn’t matter if the 5+ players are “better players” (though they might be).  It matters if their true talent is better relative to this year’s regular season performance.  But sounds like we’re on the same page....

David G was blocked from posting, so here is his post:

==========================

Hey guys,

Just wanted to let you know I’ve seen all your comments and will have an article on THT addressing all of them soon enough.

But sounds like we’re on the same page....

Yup.

I’m not sure the regression and true talent issue are going to be all that large though, as compared to the effects that he found.

I mean, a couple of the relevant questions are:

What do we expect to be the average true talent of a player who has had less than 2 year’s experience in the post season and a player who has had 5+ experience, even after adjusting for age?  Is the former 3 points in wOBA above average and the latter 5, or is it like 10 and 20?  Or is it 1 point and 2 points.  I have no idea.

And what would the wOBA during the regular season be of an average player on a playoff team be?  2 points above league average? 10 points?  (And then how much of that above average number is true talent and how much is luck - IOW, how much will he regress in the post-season, after adjusting for the weather and quality of competition in the post-season?) Again, I have no idea.

The research that David is trying to do is a little tricky.  Maybe a little trickier than he anticipated.

Thinking about it some more, it doesn’t seem likely that pure regression can explain a difference of .012.  But it may explain some of the difference.  On the other hand, another complication is quality of opponent in the post-season.  The experienced players will tend to be on better teams, and thus face weaker opponents than the inexperienced players.  For example, a bunch of David’s 5+ players must be players on the 1950s Yankees, whose entire post-season record was beating up weaker NL squads (and the reverse is true for those NL players).  Indeed, given the fewer opportunities to make the post-season, the 1946-68 period is especially problematic.  (And for the last decade or so, the imbalance in league quality creates an expectation that NL players will decline more in the WS.) When you consider how often the 5+ players will be on really strong teams, I think this could play an even bigger role than the difference in regression rates. 

As you say, it gets surprisingly complicated.  I think a good starting point would be to look at the regular-season performance of these players in year N-1.  It might also help to look at the average W% or pythag record of post-season opponents for the 5+ players, to see if they faced weaker opponents (though that may be more work than it’s worth).  Let’s see what David comes up with.....