THE BOOK--Playing The Percentages In Baseball

Included in there, and I don’t see any indication that it was controlled for, is that very good young pitchers often get to pitch out of the bullpen for a few years. This could be because teams have innings limits on them, or their secondary pitches aren’t ready for starting, etc, but how much of that age 21-22 peak is due to future starters airing it out in the bullpen for a year or two? Bard and Feliz come to mind this year as possible moves to the rotation.

This has two effects - the obvious one is the bullpens have a lower percentage of the “good arms” as age gets older. Starting rotations, however, also gain from this, since that’s where these guys end up. That could explain the maintainence of SP K% over those age ranges, at least in some small part.

Along those same lines, once a pitcher loses effectiveness in the rotation, he is often forced to pitch in the bullpen. I imagine if you look at all relievers over 35, a significant subset of them were starters in their 20s. However, for starters over 35, it’s going to be difficult to spot a guy who was primarily a reliever in his 20s.

To be clear, the author is using a “Delta” method, meaning comparing K rates of the SAME players, in any pair of years.

Granted, but is he accounting for year to year changes in role? Even comparing the same player won’t fix this unless he accounts for role changes. Is the effect significant? Probably not, but it should be included.

Wouldn’t we expect the relievers to show more mean regression, because of the selective sampling problem in the delta method, simply based on the smaller sample sizes?  It’s easier for a year 1 pitcher to get lucky over 60 IP than 150 IP.  (But I haven’t read the methodology here.)

Presumably, it’s reliever stats in both years.  Otherwise, how would he classify it?

There is the obvious selection bias to contend with, which I’ve already talked about at length in The Book, and here in the past.

Just to clarify, I guess what I meant is if a player goes to the bullpen from 35-36, then his 34-35 will be a starter and his 36-37 a reliever. In other words, you should exclude all players who could fit into both starters and relievers at some point in their career.

Of course, their exclusion will lead to other biases.

5/Guy - Yes, but there’s also a chance they have a horrible season and then rebound. Now, the lucky player is more likely to be employed next season than the unlucky player, but it should average out if the ratio of next year employment isn’t too lopsided.

7:  But it won’t even out.  Players who do well in year 1 will be much more likely to meet the year 2 inning threshhold than those who pitch poorly.  And this is especially true for young relievers, many of whom are “cusp” players who often end up in the minors if they don’t perform.

Is there a bigger percent change because the relievers start out with a higher K/9 rate than starters?

I can’t seem to find it, but what (if any) innings threshold does the author use?

I think what you actually mean is that player 1 is more likely to play at all in the second year. Let’s assume there are two players are of equal value, one has a lucky first year and the other unlucky. However, they are equally skilled, so each has just as good of a chance of reaching the inning threshold in the second year. Granted, one may have a multi-million contract and the other is a NRI, but they are equally likely to hit the threshold. The unlucky player may not hit the threshold in the first year though.

And the ‘cusp’ players are that for a reason - they aren’t very good. Bad players may be more likely to play when they are young (when they have ‘upside’, which will affect the aging curves, but I’m not sure this matters for your argument in 5.

I did the calculations.

No innings threshold, but the more innings you threw, the more they are counted. There is a link in the article to the method used to calculate them.

I used only the innings the pitcher threw in Starting and Relief appearances, not the main roles of the pitcher in that year. If the pitcher started and relieved in year 1 and only relieved in year 2, I only ran the data for relief innings.

Thanks, Jeff.

The problem of selective sampling remains, however.  Players who over-perform in year 1 are more likely to return in year 2.  And even those under-performers who do return are probably likely to pitch fewer total innings. 

And this pattern will especially occur among the least talented pitchers, i.e. relievers.  If a 6.0 K/9 reliever (true talent) delivers 7.5 K/9 this year, he’ll get a chance to pitch next year. But if the same guy delivers 4.5 K/9, he probably finds himself in AAA.  This can happen to starting pitchers as well, but I’d expect the impact of selective sampling to be smaller because 1) starters throw more innings, reducing the role of luck in their performance, and 2) starters are more talented, so teams are more likely to give them repeated chances to succeed.