THE BOOK--Playing The Percentages In Baseball

I can run that although I am not sure I quite understand the first part, specifically what you mean about breakeven between eighth and ninth. If I move #9 to #6 he can no longer be #8 or #9??

Part 2 is clear.

John:  Nice work.  BTW, I don’t think you can rely on Pinto’s lineup tool (unless it’s been modified since I last saw it).  It isn’t dynamic, so each lineup spot continues to have the same weighted value even as you change the types of hitters before and after that slot.  Obviously, that’s wrong:  if you bat the pitcher 8th, for example, that has to reduce the coefficient for the #9 hitter. 

(I’m also not sure I buy Cyril’s original regression values, but even if you do, the tool can’t work.)

I’m with Guy.

As for my initial request: suppose that the #8 hitter is .310/.370 and the #9 hitter is .200/.240.  In this case, we are probably better off swapping the two players.  Now, what if it was a .210/.260?  .220/.275?  Exactly how good does the #9 hitter have to be for the swap to NOT make sense.  I showed that in the AL, you do NOT want to swap.  I showed that in the NL, you DO want to swap.  Therefore, when does it not matter?

In the ballhype comments, I wrote this:

If you are going to test whether batting the pitcher in any given slot in a particular batting order was correct “before the fact” (which is really the only reasonable or interesting perspective), you can’t use the after the fact stats (OPS or whatever) for each batting slot.  You have to use some kind of projection for each slot if you want to look at a particular team.

Otherwise, you are liable to reach ridiculous conclusions based on amsple fluctuations in the stats for various slots.  For example, let’s say that at the end of a certain time period, say, one season, your number 2 and 3 hitters rarely got on base.  You might conclude (and correctly so) that you should bat your worst hitter in the 4 hole, since no one was ever on base when that slot came up.

Of course, we don’t care about those kinds of analyses (what should we have done given what happened).  We care about what we should have done given what should have happened (on the average) or what was most likely to happen, or what was the distribution of likely events, etc.

So using those particular numbers for the STL lineup, after the fact, is NOT a good way to do the analysis of where the pitcher should have batted, given what was known at the time the decision was made (again, which is really what we are after).

One more thing:  “There is no doubt that LaRussa is a brilliant manager?”  That is news to me.  I think there is lots of doubt, regardless of your definition of brilliant.

Got it tango—will run that.

***

On MGL’s point I concurred, largely, and posted this in response

I agree with you that you are right, if you are doing a strict analysis of whether it was a good decision. I guess I was not—that was a topic that you covered extensively in The Book. What I wanted to know was what benefit (roughly) has TLR achieved since making the switch.

I not sure I agree with you when you say that we don’t care about those analyses. I find some after the fact analysis interesting. WPA is probably the most after the fact stat and, for me, that is interesting, although as a “value metric” it has severe weaknesses.

Finally I agree with you on the definition of brilliant—that is quite a loose term. Successful would have been a more appropriate word.

***
I will re-run at some point as MGL suggested

I am recently on a little bit of a bandwagon regarding attributing team success to the manager, which I think is more than ridiculous.  If Torre, LaRussa or anyone else manages the Devil Rays, they are still going to lose lots of games.  If any manager manages the Red Sox or Yankees they are still going to win lots of games. I am not saying that there are not good and bad managers and that they do not have an influence on team wins.  It is just that we have no idea who they are, no idea how to measure the impact of a manager, no idea how to evaluate managers in general, and a team’s w/l record has ALMOST nothing to do with the talent or impact of a manager.  For example, if Maddon and Tracy were good managers with a 1-2 win influence and Torre and Francona were bad managers with a negative 1 to 2 win influence, how in the world would we know that?  The Pirates and Rays would still lose a boat load of games and the Yankees and Red Sox would still win a boat load of games.  I am sick of hearing, “What a great job (insert manager here) did - look at how well the team did.”  Of course, you often hear that when the team was not “expected” to do well, but the expectations often referenced are poor ones (like the Nats were only supposed to win 50 or 60 games - yeah, right, that may be what Buster Olney thinks), as well as the fact that even if a team overperforms their expectations, the likelihood the manager is responisble for that is slim at best.  And what about teams that were expected to do well and did (like ANA)?  Does the manager get no credit?  Scoscia is just as likely to be as good or even a better manager than Melvin.

End of rant about managers.

Okay. Here is part 1. Part 2 (Pujols) will have to follow tomorrow as I gotta run soon.

#8 in NL hit: .255/.325/.379
#9 hit: .187/.241/.269

Swap = 0.4 runs per 162

#9 .202/.256/.291

Swap = 0.3 runs per 162

#9 .235/.287/.339

Swap = 0.1 runs per 162

#9 = .245/.319/.357

swap = 0.08 runs per 162

***

Basically it tends to zero until the hitters match. By the way doing the same for AL shows that indeed you don’t want to swap. Why? In the AL 2007 was:

#8: .241/.301/.364
#9: .250/.302/.347

****
This assumes that all hitters have the same transition matrix, as pitchers bunt more base runners will move differently which will affect things. I can adjust for that if you want but that requires a bit more work as I have to ajdust the pitcher transition matrix. Did your work in the Book adjust for that?

The 0.4 is alot less than I got.  I think it was 4 runs if I kept the pitcher for the whole game, and 2.4 with the PH.

I’m pretty sure I gave each lineup slot its own transition matrix (which may explain the difference), but I can’t remember right now.

I thought you got 1.9 not adjusting for PH, but haven’t got the book in front of me so could well be wrong. Anyway, assuming that you got 2 runs adjusting I suspect that the transition matrix accounts for the difference.

Only changing the 1XX state for pitcher to account for bunting the 0.4 goes up to 0.8. This accounts for PH in the fact that the #9 stats include pitchers + PH, though the transition matrix may not be quite right. I guess that if you fix the transition matrix for the other states then that will close the gap, probably. BTW when I say the transition matrix I mostly mean when batters run on the out, which will be different for pitchers as they bunt more.

On the AL note, if you make #8=#9 (with rest of batting order the standard NL) then there is no benefit in swapping the two. If you take 1 single away from #9 and then swap to #8 you get an increase in runs 0.0004 per 162 games. This implies that your worst hitter should bat 8. #8 leads off an inning 92% of the time while #9 does so 85% of the time (assuming equality). This barely changes when you swap the hitters. The benefit of having the top of the order to drive you in outweighs the extra PA you get from batting #8

If we’re running this study specifically for the Cardinals, don’t we also need to account for all their non-Pujols hitters being worse than league-average?

James—the Cardinals specific piece was based on Cardinals hitters, not league average.

Straight from 1956:

http://www.tangotiger.net/files/Bragan_Batting_Lineup_1956.GIF

When the image comes up, click on it, to read it.