David Pinto is on to Something: Using (Adjusted) Wins & Losses to Assess Starting Pitchers

David Pinto over at Baseball Musings took a very interesting approach in an analysis of Bert Blyleven the other day.

In fact, I might call it brilliant.

I’m sure you’ve heard these:

Every stat geek boils with rage when they hear the average fan assess a starting pitcher based on his wins and losses. We all know wins and losses are a team statistic. A pitcher can lose 1-0 or win by allowing six runs in five innings… yadda yadda yadda. We’ve heard/made that argument a million times.

What if there was a way to adjust winning percentage based on actual performance?

There just might be. Pinto says:

I decided to look at games scores in a probabilistic fashion. Given a particular game score, or range of game scores, what is the probability of winning that start? The probability of losing that start?

Ooooh, game scores. Pinto is onto something.

What’s so brilliant about this?

Stat geeks can talk about all the advanced stats out there until we’re blue in the face. It will all get lost on the average fan. However, now we can take these game scores and adjust win-loss records accordingly. The average fan is happy because we’re talking a language they understand (wins and losses). The stat geeks are happier because the wins and losses are based on actual pitcher performance, not team performance.

What are the flaws in the system?

Pinto compared pitchers’ game scores to a probabilistic model built on all game scores from 1957 to 2008. As it stands now, there’s a couple problems with the results. Those can be corrected, but only with a ton more research. They are:

If you wanted to go completely defense-independent, I suppose you could modify the game score to only address walks, homers, and strikeouts. For the era, to do it right you’d probably need a model based on each season. So, you’d take each pitcher’s game scores for a certain year, run them against the average for that year, and then move onto the next year. Lots more work, but far more accurate.

As for the park factor, I’m not a huge park factors guy. A couple really good pitchers can turn a hitter’s park into a pitcher’s park. But in some cases, like Coors in the 90’s, you have to address it. Perhaps just the obvious outliers should be addressed.

Can a tool be built to do all of this? Sure. Can I do it? I’m not so sure about that. I’ll be looking into what it will take, though.

What were Pinto’s findings?

You can view the complete set of data. First, here are the players that Pinto himself discussed:

Bert Blyleven: Bert was the reason the research was done. And the hypothesis was proven:

He started 685 games, and compiled a record of 286-248 in those games. His expected won-loss record was 309-205. The loss number is somewhat more interesting. Not only did Blyleven fall short on wins, but his ability to go deep in games while his team didn’t score saddled him with an extra 43 losses.

Don Sutton: People pick on Don Sutton. They think he’s a cheap 300 game winner (only got it because of longevity). Well he wasn’t really a 300 game winner. More like a 350 game winner.

Don compiled a 321-253 record in his starts, when he should have been 347-224. By reaching the magic 300 level, Don made the Hall, but there were quite a few people who felt the same way about Sutton that they did about Bert.

Nolan Ryan: Nolan Ryan is funny. People either think he’s the greatest pitcher ever or the most overrated pitcher ever. He was my favorite player as a kid, and probably my favorite pitcher ever to play the game. I’ve always looked at some of his seasons and thought… “How is that possible?” Like these:

Of course, next to Nolan Ryan, Bert can’t complain too much. [Ryan's] predicted record was 384-209! (He compiled a 318-291 record in his starts.)

384 wins. That’s how well Nolan Ryan pitched.

Tom Glavine: The news wasn’t quite as good for Mr. Glavine.

Of the 300 game winners in the study, only one pitcher is shown as being undeserving, Tom Glavine. His record of 305-203 in starts should have been closer to 267-232. Glavine will go down as the anti-Blyleven, someone good but not great who makes the Hall because he played on very good teams during his career.

I’d still say that winning 267 is pretty darn great, though.

Some other players I noticed:

As you can see, there are some flaws without factoring era (and to a lesser extent, park) into the model. But I love the concept. There’s a bunch of work that would have be done to get an accurate model, but it could be a very valuable statistic for assessing starting pitchers.

Thanks for writing the post, David!