How to optimize a lineup, sabermetric style

Remember when I looked at lineup order on this blog? Baseball Musings has posted an online lineup analysis tool which allows you to predict the strength of an offense based on OBP and SLG numbers for each player. Even more interesting is that it gives sabermetrics-optimized alternative batting orders which would actually yield more runs than the lineup you put in.

Of course, this is kind of a naive projection, because you don’t have anything in there about left-right splits, how often players might be injured, the quality of your reserves, etc. But it still provides you with the overall quality of an offense, and is interesting to think about, right? If there’s a platoon at a position, you can plug in composite numbers, due to the freeform fields on the page. I think the coolest application is to plug in players to see what their effect might be if they were signed by or traded to your team. BM is already using it to project some 2008 offenses, like the Detroit Tigers.

Here’s what the prediction was for some offenses using 2007 numbers, to test how accurate it is:

2007 New York Yankees (best in the AL with 968 RS)

1. Damon
2. Jeter
3. Abreu
4. Rodriguez
5. Matsui
6. Posada
7. Cano
8. Giambi
9. Cabrera

Plugging in this batting order gave a result of 6.049 R/G. Their actual performance was 5.98 R/G, down due to at-bats by subs such as Doug Mientkiewicz and Andy Phillips. Interestingly enough, the optimized lineup, according to the calculation, would’ve looked like this:

1. Posada
2. Matsui
3. Abreu
4. Rodriguez
5. Jeter
6. Cano
7. Giambi
8. Cabrera
9. Damon

Some obvious problems here. Posada gets that many at bats, he wears out as a catcher. I don’t see a lot of players going for that, and Joe Torre would have been fired a lot earlier if he posted that sort of lineup. But that’s the traditional bias in baseball, isn’t it?

2007 Chicago White Sox (worst in the AL with 693 RS)

1. Owens
2. Iguchi
3. Thome
4. Konerko
5. Dye
6. Fields
7. Pierzynski
8. Podsednik
9. Uribe

With this lineup, we get 4.793 R/G. Actual output was 4.28 R/G, due to injuries and the trade of Tadahito Iguchi. My feeling is that bad lineups will move players around and substitute more during a season, which hurts production (if the subs are so great, why aren’t they starting?).

The optimal lineup given by the calculation was:

1. Thome
2. Konerko
3. Iguchi
4. Dye
5. Pierzynski
6. Fields
7. Uribe
8. Podsednik
9. Owens

That would have yielded approximately 4.960 R/G. The general strategy seems to be simple: get your highest OPS hitters the most ABs, and cluster the sluggers early-middle in the lineup. The Chicago offense suffered by leading off the very player they should have been batting 9th.