An improvement to "Quality Starts"
I'd like to present a new way to quantify the contribution of starting pitching. I've tried to improve upon the (deservedly) maligned "Quality Start" statistic.
Using data from Baseball Reference, I assigned each start a Start Quality (SQ) from 1 to 4:
- <5 IP or >4 ER is a "1". In the 2016 NL, 29% of starts were 1's.
- <6 IP or >3 ER is a "2". In the 2016 NL, 25% of starts were 2's.
- <7 IP or >2 ER is a "3". In the 2016 NL, 28% of starts were 3's.
- 7+ IP & less than 3 ER is a "4". In the 2016 NL, 18% of starts were 4's.
Summarizing a Season
Top Right: Start-by-Start. Each of the teams 162 starts is shown in the largest, top right section; darker colors indicate better SQ. They are organized by 5-day, once-through-the-rotation 'baseball weeks' and the pitcher is indicated at left.
Top Left: Starter-by-Starter. The histograms at the top left show how many of each SQ were made by each pitcher. The longer the bar, the more starts they made. As usual, the more purple, the more better.
Bottom Right: Week-by-Week. These graphs show how many of each SQ were made for the team in each week of the season. This chart makes it easy to see when your starters had strong and weak stretches. You can see the 2016 Rockies were quite strong in the middle of the season, weeks 10 to 22. Before and after, the starters were much less consistent.
Bottom Left: Season Summary. These bars show the total number of each start type (bottom row) and the total number of each start type that were wins (top row). Teams will win the vast majority of SQ=4 starts, over 75%, and most of SQ=3 starts. SQ=2 games are competitive as the starter has at least kept the team in the game. Teams lose most of the duds (SQ=1).
What we learn about the Rockies season
From the top left pitcher summary, you can see that the Rockies had a solid core of 5 good, but inconsistent pitchers. Bettis, Gray, Chatwood, JDLR and Anderson took most of the starts this year and they weren't awful. Of the 5, only Gray had a significant number of dominant (Rating=4) starts. All 5 contributed a good number of SQ=3 starts. Unfortunately, this group also turned in a fair number of duds (SQ=1). To make things worse, Lyles, Butler, and Hoffman started about 20 games and they were mostly awful. You can see the difference in the week-by-week numbers. There is a section of mostly purple after week 12, when Anderson came up and JDLR returned. Their starts were solid and they removed Butler and Rusin.
The bottom left team summary shows we had too many duds - about 50 games worth of SQ=1. The Rockies will always be a bit high in awful starts because LOLCOORZ but still, there were too many. The team actually had a solid number of SQ=3 and SQ=4 starts, at almost the league average. Its worth noting that this year was WAY better than the recent Rockies past - but more on that another time.
To compete next year, the Rockies need to cut down on the duds. The path to this isn't so hard to imagine:
- Jon Gray eliminates many of his duds. If the kid stays healthy, I'm optimistic he will improve (and have some better luck).
- Replace early-season JDLR with something better ... from the get-go. One of our young guys needs to be solid this year, starting on Opening Day.
- A full season from Tyler Anderson without much regression. Stay healthy Tyler!
- Bettis and Chatwood continue to be solid MOR guys (though Chatty had the crazy Coors splits).
- Better Depth - To account for injuries and/or regressions, we will probably need at least 2 solid contributors from our collection of TINSTAPPs. One will replace JDLR. If our depth improves, we will win more of the games (and spare our bullpen) when our top 5 are out of action.
What it looks like if everything goes perfectly
This is what it looks like when you have 5 good pitchers who are consistent and healthy for an entire season. Just. Wow. The Cubbies only had ~10 starts outside of their top 5.
Also, take a look at Kyle Hendricks. In 30 starts, he NEVER threw a dud. Not a one.He made it through at least 5 IP in every start and through 6 IP in 20 of the 30.