At last! The eagerly awaited results are here!
For those who missed the original post, I asked folks to submit lineups based on (what I thought would be) the probable starters against LHP and against RHP. And, to top it off, folks could also submit their favorite lineups among all Reds personnel. I took these submitted lineups and compared them to those that Dusty has employed already this season using CHONE projections and John Beamer's Markov model.
Let's start by comparing the lineups against LHP. These lineups h ad to use the following players: Harang (as a pitcher), Hernandez, Votto, Phillips, Rolen, Cabrera, Gomes, Stubbs, and Bruce. It is essentially what has turned out to be the primary starting defensive lineup for the Reds. Here are all the proposed lineups using that combination of players, including one of Dusty's:
Lineup vs. LHP
Pretty happily surprised to see my lineup come up first on this list (and honest, I posted that lineup before I ran it through the model). My approach was to follow The Book. Here are the basic arguments put forth in The Book for lineup construction, which are based primarily on MGL's lineup simulator:
1. Your best three hitters should hit 1st, 2nd, and 4th. Lean toward OBP for the leadoff guy. Lean power for the cleanup hitter. The 2nd hitters should be balanced.
Based on CHONE projections, the best hitter projections are Votto and Bruce. Bruce's projection is power-heavy, with the best SLG of the group, so I put him in the 4-hole. Votto has the best OBP and nearly as much power as Bruce, so I judged that he fit better in the 2-slot. By wOBA, Gomes was the next best hitter, but his OBP projection is only 0.333, so I went to the next guy on the list, Rolen, who had a fine .350 OBP to be my #1 hitter.
2. Your next two best hitters should hit #3 and #5. You get more payout for power in the #3 slot than OBP because the #3 hitter comes up with two outs and no one one base more often than any other lineup slot (because he hits third in the first inning).
Gomes & Phillips were the next two best hitters. Gomes is better in both categories, so I decided to give him the PA's and hit him 3rd, with Phillips 5th.
3. Your 6th & 7th hitters should be your next two best hitters.
That's Hernandez and then either Stubbs or Cabrera, who have very similar projections. I went with Stubbs, mostly because I wanted to save Cabrera (who had a slightly higher OBP) for the next step. It also let me leverage Stubbs' speed with a pitcher at the plate...you get more of a speed payout when it occurs in front of anemic hitters, and at least should make sacrificing easier on the pitcher...
4. If a player exists in a lineup that is significantly worse than the other hitters (like a pitcher), it is best to hit that player 8th. The reason is that the 9th hitter can then become a "secondary leadoff hitter," getting on base for the good hitters at the top of the order.
Given that Stubbs and Cabrera are so similar, I opted to put Cabrera in the 9-hole to use his slightly better OBP (0.331) to get on base for Rolen, Votto, and Gomes.
Overall, I was pretty happy with this lineup. The players fit their lineup "roles" well, and it avoided lumping two lefties in a row, which I think is worth doing whenever you can. The other lineups that fared well tended to follow similar sorts of patterns. GregD's lineup was almost identical to mine, except that it flops Bruce and Gomes, as well as Stubbs and Cabrera. It performed equally well. Jason461's is the same as GregD's, except it again swapped Cabrera and Stubbs.
Based on the model (and yes, absolutely, this all assumes that the model is accurate...), there was a 21-run difference per season between my lineup and the one that Dusty Baker used on April 7th, including these same players (bolded above). That's about two wins over the course of the season. Not much, but on last year's free agent market two wins is worth something $7-8 million. That's nothing to sneeze at.
A few other observations:
- A common feature of less successful lineups was hitting Cabrera or Stubbs second. The 2-hole hitter is among the most important spots in the lineup, with big payoffs for both on-base and power abilities. Hitting Cabrera in the 2-hole wastes the chance to leverage much production out of the top of the order. Baker, of course, HAS to hit his shortstop second.
- Phillips hitting cleanup also tended to occur among many of the more poorly-performing lineups. Phillips just doesn't have the power to hit there, at least not when there are much better options (e.g. Votto, Bruce, Gomes).
- I used David Pinto's lineup tool to generate some additional lineups, and used the best and worst lineups it generated in the Markov. Its worst lineups were genuinely terrible (Harang leading off, etc). But it's "best" lineups didn't fare as well. Why? They often hit either Cabrera or Stubbs 3rd. I'm willing to be pretty radical about my lineup construction, but that's far too many PA's to sub-par hitters for my taste. Pinto's work is based on regression models, and I think this shows their limited application--they don't capture the sorts of interactions that can come into play when one spot in a lineup affects how often another comes up to the plate with runners on base, for example.
- I threw in some "idiot" models that were just based on ranks of players according to stats, like OBP, SLG, OPS, etc. This follows the idea that we should give the best players as many PA's as we can, and ignore anything else about lineup construction. Some of these did work out pretty well--again, this emphasizes the importance of getting good hitters at the top of the lineup!
Lineup vs. RHP
I can't win them all, apparently. Still, if you look at the lineups that ranked well, they largely adhere to the guidelines set out in The Book. OBP in the first slot. Balance in the 2nd slot. Power in the 4th slot. Best five hitters hitting in the first five lineup slots. Etc.
Comparing my lineup to those that did better, the major difference is Dickerson hitting 1st in the best lineups, while I had Rolen there again. Dickerson projects for a fine .348 OBP but zero power. Rolen's OBP is basically the same, but with more power. Apparently, and perhaps unsurprisingly, the lineups seem to work better saving that power for when someone like Dickerson is already on base. Based on that, I'd probably modify the general guidelines above to say that you should strongly emphasize OBP in the #1 slot, but try to save power for later in the lineup if possible.
Again, poorly performing lineups (ignoring the Harang leadoff ones) tended to put Stubbs or Cabrera (the two worst hitters) in the first 2-3 lineup slots. Phillips hitting cleanup also tended to not be a very successful strategy, once again.
One critique you could legitimately raise about this study is that even though I asked for "Vs. LHP" and "Vs. RHP" lineups, I did not use left/right splits for these hitters. I have tried to address this by calculating regressed splits for each player and then applying these splits to the above lineups. The types of lineups that win "best lineup" are largely the same (although I don't "win" the LHP competition this time!). I've copied those results to a google spreadsheet for those interested.
Best All-Organization Lineups
Here's where it gets fun: users got to take their pick of players from around the organization, filling the lineup with our best possible combination of offensive talent. Here are the results:
I had to get all clever and put in Balentien in LF, Dickerson in CF, and Frazer at 2B (with Phillips at SS). But again, following The Book, my lineup came out on top. Jason 461's was fairly similar, except that I went with Frazier and he continued to let Cabrera hit at SS. Most of the best-performing lineups have the best hitters hitting in the first five slots, while many of the worst lineups have Stubbs and/or Cabrera (or Hanigan) somewhere near the top of the lineup, and many also have Bruce (who had a very strong projection) down near the bottom.
I put in four of Dusty's lineups. Two finished last (not counting crolfer's funny one), but to be fair they did so more because of personnel (one missing Rolen, another missing Bruce) than batting order. The other two were fairly average. I think that's ultimately were we are, at least with this player set. There is no Willy Taveras or Corey Patterson on this team with a sub-0.300 OBP that Dusty can hit leadoff, and so he just can't do that much damage.
Lineup construction does matter. I'd guess that Dusty's lineups cost the Reds ~2 wins per season, which seems like a lot to me. It's maddening to think that in a business that spends millions upon millions to gain an extra win in the standings, "free" inefficiencies like this can still exist. But that seems to be what we're looking at here.
That all said, we should be a bit cautious here. These results are based on a model, the fine intricacies of which I can't professor to fully understand. I know the principles about how this thing is structured, and it all seems correct--but no matter how good the model, there will be ways in which it cannot match up to reality. And even if Dusty is costing the Reds a few wins per season with lineups, his other merits as a manager--motivation of players, maintaining an organized, cohesive clubhouse, etc--may offset what he loses with the lineups. Dusty's not a great manager. Lineups are probably the biggest problem. But I think think the Reds could do worse.