Notes on optimization
Posted by disappointmentzone on 28 July 2006
There are a number of caveats that I should point out with respect to optimizing lineups. First, the sample size below is too small to make general assumptions about 1) the volatility of lineups and 2) the skill of managers. This season Eric Wedge has done poorly relative to his peers, but there isn’t enough evidence to draw the conclusions that other managers would have done better with the Indians roster or that Eric Wedge would have butchered the lineups of other teams. All we can fairly do at this point is compare Eric Wedge against the optimal lineup he could have produced.
The other important caveat is that the optimal lineups have the benefit of hindsight, a luxury no manager has. To mitigate hindsight as much as possible I have taken the season averages of all of the players for this season. By this point in the season OBP and SLG% are fairly static–the values we see now are probably going to be very close to the final statistics for most of the players. If I were to go back and for each game optimize the lineup based on how each player performed in that game–optimize on a game-by-game basis rather than a season-by-season basis–we would find that the ORP of the Indians (as well as every other team) would be significantly higher than it is now. Using the season statistics for each player is a good way to mitigate the hindsight bias as much as possible.
Now, there are certain things Wedge has done to cost the Indians runs that most people who follow baseball probably would have changed. His most costly mistake was not moving Peralta out of the top of the order fast enough. The most important part of the lineup is the top–which is one of the reasons why Seattle has performed so close to optimal; Ichiro is the optimal leadoff hitter against both righties and lefties–and leaving a slumping Peralta in the top of the order for so long (until late May) really hurt the Indians. Wedge, however, was justified in batting Peralta in the top of the order at the beginning of the season.
Since Wedge was correct in batting Peralta at the top of the order in April, should one fault him for doing so? No. But optimizing based on the season statistics does just that. So let’s push a bit further into what Eric Wedge should have done and what the results would have been had he done so.
Based on 2005 performance (and for Jason Michaels, this season), the opening day lineup should have been: Martinez, Hafner, Belliard, Peralta, Sizemore, Blake, Boone, Broussard, Michaels. Based on their April statistics, this lineup (this actual lineup, not this group of players in two platooned lineups) would have produced 154 runs in 25 April games, or 20 more runs than the actual team scored in April. 20 games equal about two additional victories. The UPP is 13%.
After April it was clear that Peralta was struggling and should have been dropped in the order, and Wedge didn’t do so. In May the Indians scored 135 runs, or about .55 runs per inning. If the May lineup were based on the optimized April lineup Peralta would bat seventh instead of third, Blake would leadoff, and Sizemore would have batted third in place of Peralta (among other changes). Based on how each player actually performed in May, the optimized April lineup would have produced 144.3 runs, .59 runs per inning. This is equals about one additional victory and a 6.4% UPP. In May Wedge moved Blake higher in the lineup to take advantage of his hot start to the season. This moved helped the Indians.
In June the Indians scored 129 runs, or about .55 runs per inning. If Wedge sat down on June 1st and adjusted his lineup based on how his players did in May, the new lineup would have scored about 147 runs, or 18 more runs than the team actually scored in May. This equals just under two more victories and a 12.2% UPP.
In July, however, things have changed. If Wedge had sat down on July 1st and adjusted his lineup based on how his players did in June, the new lineup would have scored about 130 runs, or .656 runs per inning. The actual lineup has scored 130 runs. The 19-run game against the Yankees really helps. But this is still fairly good. I’m not certain what to attribute the recent success to (other than 19 runs in one game), but here are two ideas: Jhonny Peralta has now been batting in the bottom of the order for over a month and during that time his average has risen and he has gotten fewer at bats with runners in scoring postition. Wedge also moved Belliard into the three hole for a number of games. Based on his statistics in June, Belliard should have been batting fifth. Moving him even higher in the order worked out well; Belliard got hot and the gamble paid off.
Of course, none of these lineups has also been optimized against the handedness of the starting pitcher. Had Wedge done that–sat down on May 1st and optimized his lineups for righties and lefties based on the stats of his players in April, for example–there would certainly be an even larger optimal run differential. And once again it should be noted that this is a small sample. One cannot extrapolate from these last four months Wedge’s ability to manage a lineup. But the findings are interesting and speak to what’s happened this season.
Also, if anyone knows where I could find monthly splits for major league players I could extend this analysis to include platooned lineups. All I can find are season splits.