Baseball Prospectus Needs Your Help! Check out our call for contributors!

On Monday’s edition of MLB Now, anchor Brian Kenny once again made the case against using wins as a measure of pitcher quality. Citing recent games such as Matt Harvey’s brilliant nine-inning, one-hit no-decision, he argued that the win is an overrated statistic that doesn’t do a good job of describing the pitcher’s performance. After Kenny’s presentation, former pitcher Al Leiter came out to give a rebuttal. Leiter had an interesting take on the issue. He said that Kenny wasn’t respecting the human element of the game, and he suggested that the win statistic might actually make starters perform a little better in some key situations.

Leiter cited the memory of a time when he was pitching and the game was tied. With two outs and his pitch count north of 95, he knew that he wouldn’t be out for the next inning, but that if he could retire the next hitter, he could get the game to his team’s offense in the next half-inning. If they scored, he would pick up the win. Leiter recalled one time that Bobby Valentine came out to the mound to check on him in such a situation and asked whether Leiter had “one more” in him. Leiter said yes, and the possibility of getting a W next to his name was what drove him.

Leiter never directly addressed the main question of whether wins are a silly statistic, but he brings up a very good point about whether the existence of the stat has some very real behavioral effects. Certainly, pitchers would love to get a win in general (and in that situation), and they may feel more motivated as a result, but does their performance actually get better in these situations because of the tantalizing lure of a possible win?

Warning! Gory Mathematical Details Ahead!
I isolated all situations from 1993-2012 in which the starter was still on the mound, it was the fifth inning or later (even if the pitcher finishes the fourth on a high note, he doesn’t get the win), the game was tied, and his pitch count was over 95. We will call these situations “Leiter situations.”

Using the logged odds ratio method of controlling for pitcher/batter quality, I controlled for the likelihood of seven events (strikeout, walk, HBP, single, double or triple, home run, or out on a ball in play). Because we’re also specifically dealing with pitch counts here, I entered pitch count prior to the at-bat as a control. I also controlled for whether the pitcher had a handedness advantage over the batter. Also, I restricted the sample to plate appearances in which a batter who had logged 250 PA in that season faced a pitcher with 250 batters faced.

I ran a series of logistic regressions to check whether the “Leiter” variable had any predictive power beyond the control variables. So, below when I talk about differences in performance, I’m talking about differences from what we might expect given that we know how good the pitcher and batter are, and the fatigue level of the pitcher. Does our starter gain (or lose) anything extra that makes us believe that he is more (or less) than we would expect of a tired pitcher nearing the end of his outing?

It’s important to find a good set of control plate appearances to use as a comparison group. So, I tried a couple of different approaches.

Leiter situations (starter, game tied, high pitch count, fifth inning or later) vs. similar situations in which the game was not tied
Leiter suggested that the fact that the game was tied was important. It means that the pitcher has extra incentive to keep the game tied, because all his team has to do is score in the next inning to make him eligible for the win. In theory, even if he does give up a run, his team could come back in the next inning and re-take the lead, positioning him for the win, but holding the line is a better outcome.

There was an effect of being in a Leiter situation here, only it’s not the one that Al suggests. Pitchers in a tie game were more likely to walk the batter, but less likely to give up a home run or record an out in play. The pitcher was not actually better at getting outs—he was just more careful. Seeing that it was a tied game, I can’t say I blame him, but the lure of the win did not make him more effective.

When I restricted the sample to situations when the game was close (within one run either way), but not tied, being part of a tie game continued to increase the chances of a walk, although no other outcomes were affected. Being in a tie game in your last gasp for the day doesn’t make you better. It just makes you more careful.

Leiter situations vs. similar situations when it was not the fifth inning or later
Here we have as our comparison group a pitcher who has a high pitch count and is in a tie game, but it’s probably the fourth inning, so no matter what he does, he will not get the win. Sure, he wants to hold the game here for his team, but there’s little personal glory to be gained.

With the pressure on and no dubya on the line, do pitchers perform any differently than we would expect given everything else we know? Not really. Strikeout rates go down a bit if there’s a win on the line, but that’s about it.

Leiter situations vs. similar situations where the pitch count isn’t quite as high, so he might be able to go another inning
Here, we’re isolating the urgency of the “this is my last inning” facet of Leiter’s formulation. In the control group, the game is still tied, it’s the fifth inning or later, and certainly, the starter wants to hold the other team here. But his pitch count isn’t quite so high. It’s between 70 and 85, and he might have another inning left in him if all goes right. (Note: Of course, earlier in his pitch count, our starter will be more effective, but we already have a control in place for that).

Is there any difference in his performance because in a Leiter situation, it will be his final chance to have an impact on the game? Nope. He’ll be better off for not having thrown as many pitches, but the fact that it’s his last chance doesn’t seem to make him better or worse.

Leiter situations vs. everything else a starter does
Maybe there’s something about the totality of the Leiter situation. I’ve tried isolating various parts of the Leiter theory to see if I can find the “active ingredient.” Maybe what we need to do is simply look at what happens in Leiter situations vs. everything else that a starting pitcher does. Maybe the fact that it’s close, it’s his last chance, and there’s most certainly a chance at a win on the line all work together to make it so that he’ll be better than we might expect.

There were differences between Leiter situations and the rest of what a pitcher does. Strikeouts and walks went up above expectations. Singles, extra-base hits, home runs, and outs in play all went down. It’s hard to say that this is an improvement in the end, because walk rates jump about 1.6 percent, while strikeout rates go up only 0.7 percent and rates of outs in play drop by almost two percentage points. Hits do go down by just shy of two percent, but in the end, fewer batters make outs. In Leiter situations, compared to all others, pitchers are apparently valuing strikeouts and accepting the additional risk of a walk, but more to the point, just trying to keep the other guy from hitting the ball.

The first control group that we looked at was to see whether the fact that it was a close game made a difference, and the results were similar in that pitchers got more careful. I’d suggest that what we’re seeing here is an extension of those findings. When the opportunity for a win is on the line, pitchers get into a different strategic mindset. Maybe they’re reaching back for something here, but there’s a difference between accepting a different risk profile for the likely outcomes and being better at retiring batters, which is the pitcher’s job.

What it Means
I have no doubt that when pitchers are in a Leiter situation, they are aware of the enormity of what’s going on. This will probably be their last inning, and the game is close. They probably do realize that whether or not they get a win depends on whether or not they do well against the next batter or two. And maybe in their heads, they do try something extra fancy to marshal all the strength that they have left to get to the win.

There just isn’t evidence that all of that will to win actually translates into any benefit in terms of the outcome of the at-bat. A starter in that situation is nothing more (or generally, less) than we would expect him to be, and that’s a tired guy nearing 100 pitches. That may still be a good pitcher, and he may get the batter to ground out to second, and his team might get him the win in the next inning. But if there’s a lesson to be learned here, it’s that one should not confuse a desire to win a game or to get an extra attaboy on the stat sheet with some sort of extra talent that will help accomplish that goal.

Thank you for reading

This is a free article. If you enjoyed it, consider subscribing to Baseball Prospectus. Subscriptions support ongoing public baseball research and analysis in an increasingly proprietary environment.

Subscribe now
You need to be logged in to comment. Login or Subscribe
The flip side of the Leiter argument is that he is assuming that a professional pitcher is not trying his hardest every inning, and thus has the ability to 'reach back' and give full effort when the situation is critical.

It's like the 'clutch' argument. If a player can focus during a critical moment to deliver exceptional results, shouldn't he work to maintain that focus every time he hits/pitches?
It strikes me more as the Jack Morris 'pitches to the score' argument. And we all know how that turned out.
It also seems like he's implying that if the win statistic didn't exist, that pitchers wouldn't put in as much effort in "Leiter situations," which rings incredibly hollow.

Very interesting article. Instead of calling them "Leiter situations" maybe you should have called them "Me First." Is a guy really going to focus better and work harder to pad his own personal win stat?

Maybe it would make sense to contrast these "Me First" situations with other "Empty the Tank" scenarios, those points in the game in which a pitcher has thrown x number of pitches, there are two outs, he knows he's facing his last hitter, and the game is not in the balance (his team is way up or way down).

Theoretically he is going to "empty the tank" to get this last hitter, even though it may not have a direct impact on the score. Maybe it's just personal pride on the line, the desire to finish on a high note, the desire to walk off feeling good about yourself?

Is there any difference between this situation and one in which there is a statistical W on the line?
Could actual odds ratios be of any use here? Logistic regression results in Stata produce what it calls odds ratios but those are the log odds ratios. Odds ratios could tell the increase/decrease in magnitude of the odds for different variables. Great stuff - BP is the best!
As opposed to all the times when he said, "take me out coach. This guy's got my number and my wet noodle arm is about to go all Dave Dravecky on me. I suck and I'm tired and my lady parts hurt."
The added proclivity for the two true outcomes makes sense to me. You don't just hear about pitchers reaching for something; you can see it. Watching games, you see pitchers toward the end of their outing disregard any inclination to pitch to weak contact and limit pitch counts, and they just start aiming for the corners, trying to throw perfect pitches. Problem is, a lot of guys don't make themselves better when they do so, as you note, Russell. They just make themselves different and less efficient. This is when the time between pitches drags, and a good hitter starts to foul balls off and frustrate the guy. Carlos Zambrano used to see way too many outings end with crucial mistakes after seven-pitch at-bats in tight games, because he started throwing 98, missing his spots, trying to bury his slider instead of throwing it at the bottom of the opponent's bat. Yeah, this finding makes sense.
Really interesting work. Would be curious to see if a similar effect could be seen with two outs in late innings regardless of game run difference and test if there's an effect of a pitcher (arbitrarily) finishing a full set of innings, be it 7 or 8.