To walk or not to walk

To walk or not to walk, that is the question.
whether ’tis nobler to in the game to suffer
the slings and arrows of the current batter,
or to walk him and face a sea of troubles.

Intentionally walking a batter can be a high risk, high reward play. In the postseason, moves like this can win, or lose, a game and go a long way in determining who wins the series. As Hamlet puts it, do you let the current batter hit, potentially giving up a key run, or do you walk him and open the door to a possible big inning?

Up until recently, it has been difficult to look at these intentional walks objectively. With the advent of WPA we now have the ability to look up the current situation and how your team’s chances would change if you walked the next batter. The only problem with WPA is it assumes everyone involved in the play is league average. This is great for getting a baseline but what if Alex Rodriguez is up, or Neifi Perez, or Roy Halladay is pitching, or the Devil Rays are playing defense? Clearly the WPA would change in these situations.

Some work has already been done on the subject. Tom Tango in particular has written extensively in the past few years on whether or not you should walk Barry Bonds. What we need is something like what Tom did, but flexible enough to handle any situation. This tool should use WPA as a jumping off point but also fold in estimates for the quality of the players involved. It also needs to be relatively quick as we don’t want to have to wait two hours to figure out whether to walk the batter or not.

What I have done to tackle the problem is create a simple simulation using 2007 data to output a corrected WPA given an estimated OPS of the at-bat. If you want to know more about the process you can read all the ugly details here. Now this system is still not perfect. It still is using a league average for all base runners, and the simulation doesn’t add in some exotic plays like multiple errors on one play, but it does a very good job for the pitcher/batter duel.

That is as long as the OPS of the at-bat is estimated correctly. Estimating this can be hard but I am recommending inputing the average of the batter’s OPS against the pitcher’s handedness and the pitcher’s OPS against against the batter’s handedness. This way the defense behind the pitcher is being added in as well.

If a batter has an extremely good track record against the pitcher, or the pitcher is tiring, or whatever other circumstance can be handled by altering the OPS up or down at your discretion. You can access the tool here. Now that we have this tool, let’s use it to examine a few of the intentional walks handed out during this post season.

Sometimes you push the wrong button and get a bad result.

The first situation I want to look at happened in game two of the Rockies-Phillies series. It was the top of the forth inning with a runner on second and two outs with the Phillies clinging to a one run lead. Yorvit Torrealba was batting and the pitcher’s spot was due up next with the top of the order, Kaz Matsui and Troy Tulowitski to follow.

You might remember this situation from one of my previous articles here. I came down pretty hard on Charlie Manuel and after reading a few comments by the readers I wanted to revisit the play. Manuel called for the intentional walk after a 3-1 count to Torrealba and Clint Hurdle pulled his starter and pinch hit Seth Smith.

If we just look at this play with WPA, the move lowered the Phillies chances of winning from 64.1% (no IBB) to 62.6% (IBB). But, using my tool, and an assist from Jeff Sackman’s excellent website, we can estimate that the Phillies WPA dropped from 65.2% to 63.4%. What if Hurdle had let the pitcher bat? Then the estimated WPA would have been 66.6% but the walk still would have lowered it to 65.7%. Now there was a 3-1 count on Torrealba so that would probably tighten these numbers but it is pretty clear this was a bad move by Manuel and it lead to a huge inning for the Rockies.

Sometimes you push the wrong button and get a good result.

Our second situation comes from the marathon game two between the Indians and the Red Sox on Saturday. It was the top of the sixth with a runner at second and two outs in a tie game. Grady Sizemore was up with Asdrubal Cabrera on deck, and Travis Hafner and Victor Martinez to follow. Hideki Okajima was pitching for the Red Sox.

Terry Francona decided to walk Sizemore and pitch to the switch hitting Cabrera. There are a lot of reasons to not like this move. First, Sizemore is a lefty and isn’t as great against left-handed pitchers. Second, Cabrera does much better against lefties. Third, if Cabrera were to reach Hafner and maybe Martinez would bat.

WPA doesn’t like this move, as it lowers the Sox’s WPA from 52.2% to 50.9%. The simulation likes this move even less with the Sox WPA going from 53.2% to 51.7%. This isn’t even counting the running ability of Sizemore so, if anything, this move was even worse. That said, I have seen almost nothing written about this move. Why? Because it worked, with Hafner lining out to end the inning.

Sometimes you push the right button and get a bad result.

Our third situation comes from game two between the Angels and the Red Sox. It was the bottom of the ninth with two outs and a runner at third in a tie game. David Ortiz was coming up to bat with Manny Ramirez on deck and Mike Lowell to follow. The Angels had their closer Francisco Rodriguez on the hill (by the way, good use of your closer Mike Scioscia).

Scioscia walked Big Papi to face Manny, who promptly ended the game with a walk-off home run. Here WPA thinks walking Ortiz was slightly negative with their WPA going from 37% to 36.4%. The simulation however sees that Ortiz destroys right-handed pitching and Manny is a mere mortal against right-handers. It thinks the WPA went from 35% to 37.8%.

I’d have to agree with the simulation here. No way am I letting Ortiz hit with two bases open. You might be wondering if walking Manny would also have been the right move but the simulation doesn’t think so. Lowell is a decent hitter and if you walk Manny, now a walk or a hit by pitch will end the game as well. So Mike Sciocia made the right call here, it just didn’t work out for him.

Sometimes you push the right button and get a good result.

Our last situation comes from game one of the Yankees-Indians series. It was the top of the fifth with runners on second and third with only one out. The Indians had a scant one run lead at the time. Alex Rodriguez was batting with Jorge Posada on deck, and Hideki Matsui and Robinson Cano to follow. Staff ace, C.C. Sabathia was pitching.

Obviously, this is an extremely dangerous situation and when I saw the play live I thought walking A-Rod was a huge mistake. As good as A-Rod is, now a walk means a run (though the double play is in order) and the guys coming up aren’t exactly chopped liver. WPA agrees, thinking this move cost the Indians, dropping their WPA from 51.8% to 50.2%. The simulation likes this move though. It thinks the Indian’s WPA went up from 51.7% to 52.2%. Not a huge increase, but a large difference compared to straight WPA.

Normally, walking the bases loaded in the middle of the game is a mistake, but with a left hander on the mound the simulation thought this was the right move. And, in the end, it worked like a charm for Eric Wedge as Sabathia was able to get out of the inning with the lead still intact.

You might be thinking to yourself, “Why am I making such a big fuss over plays that change the WPA by a percent or two?” Well, those moves add up. If your manager makes a -1.25% mistake in every game over the season, he has cost your team over two games. Ask the Mets, Padres, or Brewer what two games in the standings meant to them. In a short series things are much more of a crapshoot but the manager’s job is the same. He is there to maximize his team’s chances of winning, even if he has less pull than he does in the regular season.

So what is the moral of the story? Well, first WPA does a pretty good job of estimating most situations. When dealing with close decisions though an improved simulation appears to produce better results. While this simulation isn’t perfect, it is a step in the right direction and it produces a pretty quick answer. If you would like to run a situation and see how a manager did you can use the tool here.