If you’ve been here before, or someone like Brian Kenny tweeted a link to this post, you know that we are big proponents of the #KillTheWin movement. We don’t like wins and losses as a pitching statistic for many reasons. You can pitch well and not get a win, you can pitch terribly and get a win, wins don’t even out, and wins are extremely misleading. Put simply, wins are dependent on things that pitchers can’t control and it’s silly to measure them based on something their teammates do. Here are links to all of our formative #KillTheWin work:
- You Can Pitch Great and Not Win
- You Can Pitch Poorly and Win
- Wins Don’t Even Out Over Long Careers
- Wins Mislead You When Comparing Players
- Assorted Facts About Wins from 2013
- Dissecting the Case in Support of Wins
But today I’d like to address a solution that a lot of people are calling for. You see, the old guard won’t let go of the wins and losses concept and language. They can’t accept things like FIP, xFIP, and WAR, or even K%, BB%, GB%. Even ERA is doesn’t satisfy their longing for the “W.” So I’d like to propose a simple idea that simply changes the methodology for awarding wins and losses. Currently, a starter has to pitch at least 5 innings, leave with a lead, and not watch the bullpen surrender that lead. If we invented wins and losses today, no doubt we wouldn’t use such a silly rule.
So let’s use a better one. If there is an appetite for Wins and Losses, why don’t we actually tie wins and losses to performance? Here are two basic proposals that do that while solving a couple of key issues with wins.
The first problem with wins and losses is that it depends on how much and when your team scores. So what we want is something that only measures the impact of the pitcher on the game. Another problem with wins and losses is that the no-decision essentially erases everything you did on a given day. If a pitcher throws 7 shutout innings and gets a no decision, that game shows up in every single one of his stats except wins and losses. We want to judge every start a pitcher makes, not just one in which the right conditions are met by his offense and bullpen.
To partially resolve this issue, let’s turn to the 2013 Tigers as an example. Instead of wins and losses as determined by the current rule, what if we allocate them by Win Probability Added (WPA) or Run Expectancy 24 (RE24)? Those two stats are a bit complicated to calculate, but extremely easy to understand. WPA reflects the percentage by which a player improved his team’s chances of winning. It is very context dependent, but you can still earn positive values even when your team is losing. RE24 is a similar statistic except it doesn’t pay attention to the score of the game and just reflects how many runs above or below average you are contributing. Think of it this way, in a 10-0 game a solo homerun has a pretty low WPA because the game is already decided, but it has the same RE24 in a 10-0 as it does in a 2-0 game. Both allow for the addition of value in a context dependent sense, but both also allow a player to add value even when his team is not. Both of these stats are readily available on FanGraphs and Baseball-Reference.
Below I present the 2013 Tigers with WPA and RE24 “wins.” If a pitcher has a positive WPA or RE24 for a single game they get a win. If it’s negative, they get a loss. No no-decisions and no concern about how the game actually ended. Did the pitcher improve his team’s chances of winning a single game? That’s what wins and losses should tell us, so let’s try this.
There are obvious weakness to this approach, namely that I’m not addressing by how much a pitcher helped his team, but to answer that question, we have season long numbers that are more important. This approach is meant to give people who want to see wins and losses a better reflection of true value.
|Pitcher||Starts||W-L||WPA W-L||RE24 W-L|
You will notice a couple of things. You’ll notice that Scherzer’s no-decisions are primarily the function of his team bailing him out and Fister, Sanchez, and Verlander’s are almost all a case of the Tigers not providing enough run support. Porcello’s are divided pretty evenly. This is interesting because it shows that even on individual teams, wins/losses/no decisions are handed out irregularly despite the same contingent of position players.
For the die-hard #KillTheWin-er, this approach is still too context dependent and derived from an illogical attempt to hand wins and losses to a single player. But for a more traditional observer, hopefully this is compelling. Even if you like wins and losses, surely you can appreciate that the actual way in which wins and losses are assigned is arbitrary and foolish. Why is 5 innings the cutoff? Why do you not get a win if you pitch 8 shutout innings and your team wins in a walkoff? Why should you get a win if you allow 6 runs? Even if you want to track day to day contribution, at least track it in a way that reflects what the player you’re judging actually did.
Now I’m not sure if this is the best way, but this is definitely an improvement over wins and losses as currently defined. The current stat makes no contribution to analysis, this one makes some contribution. I’d still rather pay attention to season long numbers, but if we’re going to judge a player in each individual game, let’s at least do it right.
Fister gains the most “wins” from this different approach. I have always felt that it is silly to give “wins” to relief pitchers. Nobody looks at a closer or setup guy that luckly or otherwise racked up a few wins. What AL pitchers would benefit the most from this measurement, King Felix?
My gut would tell me Felix would benefit a lot from this. I don’t know if he would benefit the most without going through everyone. Felix is 12-9 right now, would be 20-9 in WPA, 20-9 in RE24.
Chris Sale, for example, is 10-12, with a 16-9 WPA, 20-5 RE24.
Great stuff as always, Neil.
This is a similar approach to the one I used for my article on Beyond the Boxscore last week, using RE24 and some quick math to help guide Cy Young voting:
Much like you, I also took a look at WPA in my research, but found that when performing tests for significance, WPA didn’t pan out, but RE24 did.
That being said, RE24 is a nice bridge between the old school and new school and is a great way to quantify superior performances, even with its limitations.
Yeah, I personally like to use context neutral numbers most of the time, but if people have a desire to see context dependent stuff, it should at least be based on something more meaningful like RE24.
I appreciated your article (and anyone down here reading this should go read it), but wanted to parse it down for people who might run at the phrase “t-test.”
These are very rational & thoughtful proposals – and I agree that if you need to keep the win, these are both huge improvements – but it will be years, even decades, before wins are awarded by any advanced metric like WPA or REL24. If you have to keep the win, the only realistic change I could see happening in the near future is having the official scorer assign the win (& loss) to the most (& least) effective pitcher. Not perfect, but at least it would end nonsense such as a pitcher giving up 1 run in 7 innings, and getting a no decision because his team scores twice in the 9th for the win.
[…] A Replacement for Wins […]