Baseball Professor header image
≡ Menu

How to Project Wins and Quality Starts

When it comes to drafting a pitching staff, we’ve always spouted two general rules:

  1. Don’t pay for saves
  2. Don’t chase wins

Obviously there comes a point when even the most reliever-averse drafter decides to dive in and take a closer, and sometimes it’s a good idea to consider a pitcher’s win potential when deciding to select him.

CC Sabathia was a workhorse starter for a very good team and won at least 17 games in five straight seasons from 2007-2011. During that span, he was the pretty much a sure-bet for 17-20 wins.

But what’s the best way to target pitchers with good win potential? Which stats should you look at? How much does his team’s offense, defense, and bullpen impact his chances at winning games? And what about quality starts? Is there a different way to go about projecting how many quality starts a pitcher should have?

I looked back at data from 2010-2012 to answer these very questions.

Projecting Wins

There are five primary, measurable factors that affect whether or not a pitcher picks up a win in any given outing:

  1. Does he last at least the minimum five-inning requirement?
  2. Does he hold the opposition to few enough runs?
  3. Does his offense back him with enough runs?
  4. Does the defense play solidly behind him?
  5. Does the bullpen hold the lead?

These five factors can each be measured statistically. To do so, I’ve chosen the following statistics, each of which corresponds to the number above:

  1. Pitcher’s IP/Start
  2. Pitcher’s ERA
  3. Team’s offensive rank in runs scored
  4. Team’s defensive rank in UZR/150
  5. Team’s rank in bullpen ERA

The next task was to see how important each of these factors was to a pitcher’s chance at winning games. To do this I found the relationship between each of these statistics and a pitcher’s win percentage (W%), measured in wins per game started (wins divided by games started). The results are below (provided as r-squared values).

winpercentage

A value of 1.000 would mean a perfect relationship between that stat and W%, however here the best value we get is 0.310 between W% and ERA. This means that 31.0% of the fluctuation in W% can be attributed to fluctuation in ERA.

Now, what exactly are we supposed to do with these numbers? To make them more useful, I’ve converted these five values to percentages to show how you should weight each of them when trying to project a starting pitcher’s win total. Then I summed up offense, defense, and bullpen to yield a cumulative “team” component.

winpercentage2

When trying to project a starting pitcher’s win total in a given year, ERA and IP are almost equally valuable and account for about 80% of the pitcher’s expected win total. The other 20% is a combination of the offense, defense, and bullpen backing him, and surprisingly the quality of the bullpen was more important than the quality of the offense! The defense’s impact wasn’t negligible, but it was the lowest of the three.

Consider Jon Lester and Jason Vargas for a moment. Over the last three years they’ve each made 96 starts with similar innings totals (Lester: 605, Vargas: 611) and ERAs (Lester: 3.85, Vargas: 3.96). Lester has won 43 of his 96 starts, Vargas has won 33. That means Vargas has won 23.2% fewer starts, right in line with our 20.1% from the table above, and he’s played for the inferior team.

Felix Hernandez and Justin Verlander match up rather closely as well with Verlander winning 59 games to Hernandez’s 40, a 32.2% gap, however Verlander’s ERA over the last three years is about a quarter-run lower than Hernandez’s.

Projecting Quality Starts

Aside from marginal impacts that a pitcher’s teammates have, things like making errors to extend innings, quality starts are almost completely within the starting pitcher’s control. Only his average innings per start and ERA really matter.

Like with wins, I did the same thing with quality start percentage (QS%), measured in quality starts per game started (QS divided by starts).

QSpercentage

As expected, offense, defense, and bullpen were all irrelevant to a pitcher’s QS%. Innings pitched per start showed a slightly higher correlation than ERA did, which is different than we saw above. With W%, ERA showed a stronger correlation than innings pitched. Here it’s the opposite.

This suggests that when trying to project a pitcher’s number of quality starts, his ability to go deep into games is slightly more important than his ability to prevent the other team from scoring. My off-the-cuff explanation for this is simple; a pitcher can pick up a quality start by giving up three runs in six innings, which would yield a 4.50 ERA. Even if a pitcher is below league average on any given day, they can still pick up a quality start. Their ability to go six-plus innings is of primary importance.

So, how much should you weigh each category?

QSpercentage2

It’s almost a 50-50 split, but innings pitched gets the nod.

If your league is converting from wins to quality starts, you need to change the way you value a starting pitcher’s statistics. Relative to ERA, innings pitched becomes about 20% more important when trying to project quality starts rather than wins, so pitchers that go deep into games are better targets than pitchers who do well in a smaller amount of innings.

Completely ignoring ERA, this chart below shows you what innings totals yield what QS% on average. It should be noted that ERA is indirectly included because there aren’t a lot of pitchers with 4.50 ERAs throwing 200+ innings in a season, but just stick with this and you’ll be a-OK!

QSconverter

Happy drafting!

facebooktwittergoogle_plusredditpinterestlinkedintumblrmail

About the author: Bryan is the co-founder of Baseball Professor and works as a consultant specializing in operational metrics and efficiency analysis. When he’s not working, blogging, or tending to basic human needs, he enjoys pondering the vastness of the universe, rewatching episodes of Breaking Bad, and avoiding snakes. (@BaseballProf)

Comments on this entry are closed.

  • Guest

    Very nice analysis here. I’ve been looking for something along these lines as my league is shifting from Ws to QS this year. Thanks!

    • http://www.baseballprof.com/ Bryan Curley

      Hey thanks, someone on our Facebook page asked me how changing from W to QS affects draft strategy, and that spawned this post. Glad it was of use.

  • DJS

    Exactly how I would have done it. But I’m too lazy to do it. So thanks! Brilliant work.

    • http://www.baseballprof.com/ Bryan Curley

      You’re welcome!

  • DJS

    Why not use FIP, xFIP, or SIERA, since ERA inherently factors in defense and you have a separate defensive factor?

    Also I feel like ERA and IP/Start have to be pretty closely related right?

    I’m running some numbers…I’ll post results tomorrow.

    • DJS

      For qualifying starters from 2010-2012 r-squared between IP/Start
      ERA: .435.
      FIP: .290
      xFIP: .280
      SIERA: .261

      • http://www.baseballprof.com/ Bryan Curley

        Thanks, I used ERA because in the end the quality start is determined by earned runs the pitcher actually allows and not theoretical earned runs the pitcher should allow. Unfortunately defense does factor into QS and can sometimes prevent (or maybe even aid) a pitcher in picking up a QS. Thanks for the numbers though, they confirm what I already suspected!

  • Paul D’Amore

    end result of 21.8 Qs assuming 6.2 IP/GS and greater than 220 IP seems pretty low. Those guys will typically be in the high 20’s no? I too am switching to QS over wins but my projections has highest QS at 28, and I take into account IP/GS and ERA in my QS calc. Am I reading the above wrong?

    • http://www.baseballprof.com/ Bryan Curley

      Last year the league leader in QS was Verlander with 27. Five guys had more than 24, 10 guys had more than 22, but keep in mind the chart takes ERA completely out of the equation and looks at just how IP and QS relate. The guys with mid-to-high 20s QS numbers are pitching with 2.80 ERAs over 230 innings, but there are also guys like James Shields (227.2 IP, 3.52 ERA, 20 QS) and Clayton Richard (218.2 IP, 3.99 ERA, 18 QS) who serve to bring down the average. If we broke the chart down into >6.2 IP with a 3.20 ERA or better (for example) we’d definitely see higher projected QS averages than 21.8.

      Thanks for the great question, though!

  • DJS

    I guess I’ve got a lot of question…did you actually use teams’ “ranks” in runs scored and UZR/150, or did you use the actual metrics runs scored and UZR/150.

    Also, is there a projection system that includes team UZR projections?

    • http://www.baseballprof.com/ Bryan Curley

      I used team ranks. In hindsight it might have been better to go with the actual numbers, but at the time I remember there being a reason for why I didn’t do it. I think it was because the data takes into account three separate seasons and team ranks were a way to view the data on the same scale. For example, if a team’s bullpen had an ERA of 3.50 in 2010 and that was good for 10th in the league, that 3.50 ERA might be good for 15th in the league in 2012. I wanted to look at the relative performance. It’s hitting me now that using relative performance (team ranks) might be good for wins where one pitcher is battling against another, but concrete numbers (specific bullpen ERA) might be better for QS where it’s about pitchers hitting certain thresholds (6 IP, 3 ER). This is something I’d like to look into further.

      Thanks for the comments!

  • drew

    Great work. Excuse my ignorance, but regarding W%, is there anyway to take the information you got and turn it into a formula? Thanks.

    • http://www.baseballprof.com/ Bryan Curley

      Of course. My research above determined that W% is about 38% the pitcher’s IP/start, 42% their ERA, and 20% the team that backs them (broken up pretty evenly between offense and bullpen with defense holding much less weight).

      Every season we project how many games started, how many innings, and what ERA we think a pitcher will have (most projection systems do this as well), which means we have everything but the 20% team portion included. If you were to project the team’s offense and bullpen, which you can also find by looking at the projected stat lines for a team’s batters and relievers, you can come up with a projected win total for the starting pitcher.

      Of course, any formula that relies on 4 variables (IP/start, ERA, offense, bullpen) is prone to severe variation, but even still it would be possible. Would it be useful? Maybe. Eyeballing it might honestly have just about as good a correlation to observed win totals as the formula would, but to answer your question, yes, it is possible.

      Thanks for reading!