Jemile Weeks Should Stop Batting Right Handed

But not for splits. The problem with splits is that not only do you have to resolve the lines with small sample sizes cut in half…but you also have to resolve the error bars to the point where they don’t overlap with each other. Sure, in 300 PAs you can say that he’s generally gonna be a .320 wOBA guy, plus or minus, say, .030 points. But for the splits? You have to have an enormous sample size to shrink that .030 error. Something like 1000 PAs against LHP.

Even if these sample sizes were infinitely large, so that these numbers are absolute measurements of his talent level, we still have no idea if the problem is him batting right handed, or just left-handed pitchers in general. You have to isolate them. To really conclude this, you’d have to have this:

vs R as R
vs L as R

In this case, the only variable is the pitcher handedness, not two variables.

of at least trying for this last month of the season. What is there to lose?

See below

Again, fail. Sorry, I’m tired. See below for 2nd embarassing mistaking on this site today. As I type this, all I can hear in my head is this. Hahah unreal

It was the coaching staff at the U of Miami going into his frosh year who wanted to make him a switch hitter, and I’m fairly certain it was the LH side they asked him to learn.
Correct me if that’s wrong, but wouldn’t it be reasonable to think he can still improve since he’s only in his 3rd month as a big leaguer, even if negative splits were apparent on his way up. Also, he should bunt more RH, especially since the LHP fielding would have the awkward throw to 1B. With Weeks flying down the 1B line, that’s how balls get thrown away & he ends up on third without getting the ball out of the infield.

To a point yes. We have lots of his splits from minors. Last year, for example, as OP mentioned is available. And although he goes through some strange gymnastics in weighing OBP and SLG, the fact is that last year Weeks was an equally successful batter from both sides of the plate:

wOBA vs. LHP = .322
wOBA vs. RHP = .324

In fact the only time he was significantly better from left side is his 2009 stint in Stockton, where he posted ridiculous splits for half a season. But then he went on to play in Midland in the same year and actually hit better there from right side. Shows you how volatile such small samples are. His Stockton stint pulls his overall minor league splits in favor of batting left-handed, by 37 points of wOBA.

This is much closer to our estimated level of his true talent split, which is .028 (.335 – .307), than to the current observed level of his talent, which is .069 (.349 – .280). If we had numbers to use for MiLB regression, we would probably come very close to matching the same .028 difference we came up with.

See below

Knowing that there are people who are interested in knowing about it makes the effort worthwhile

I think you have to ask yourself two questions when you face such a problem:

1) Is an appropriate group definable?
For example, I would say that your first definition isn’t really, because it is open to too much interpretation. Second, depending on how one defines “excellent minor league numbers” is. Or one could just say “#1 overall picks” and then it sure would be.

2) Does such a group have notably different observed level of talent (meaning: results) than the overall group does?

(Then there is also an issue if the group you have defined has enough people in it and how easy it is to find the data for such a group)

But, in general, if you can answer both 1) and 2) with a yes than you should definitely use a subset group to regress to.

Couple more things that are important to understand.
1. By regressing someone with little observed talent towards the mean talent of the group he belongs to, one is not saying: “I think so and so’s real talent is exactly the average of that group”. One is saying, instead: "I don’t know what his real level of talent is because he has played so little. But instead of just assuming that his real talent level is exactly the level at which he played in the little time he had (for example .100 BA for Harper or 1.000 if he starts 10 for 10), our best guess is the combination of his own results and the average of the group we believe he belongs to. It is far from exact, but it is much better than the alternative.

2. This is pure math and on mathematical level it works. It is more accurate to regress than to draw conclusions on small samples. However, regression is only the best way to estimate talent when only using the results to do so. When there simply are very few results to work with, probably the most accurate is the evaluation by professional scouts. So, in the very early stages of a players career, I’d say scouts >> regressed >> observed, and as time passes regressed goes to the top, and as even more time passes, regressed and observed become very similar.