Starting Pitcher Ratings: GSDev Single-Season Scores

So far in this series, we’ve discussed the reason to develop a new metric for starting pitchers, explored Game Score as the basis for such a metric, and introduced the calculations used for the metric itself. Now, let’s get to some results.

As a reminder, the formula for Game Score Deviations (GSDev) is as follows:

As an example, let’s look at 2021 AL Cy Young winner Robbie Ray. Ray posted an average GS2 of 59.0, which increases to 60.2 when adjusting for opponent and park. The league average adjusted Game Score that year was 48.8, and the regressed five-year scrub deviation was 8.67. The scrub pitcher baseline therefore works out to 41.1, which is 19.1 points below Ray’s average. Ray made 32 starts, so the league deviation (8.67) is divided by the square root of 31, coming out to 1.56. Ray’s GSDev score is the ratio between 19.1 and 1.56, which is 12.26.

Of course, a single season for a single pitcher doesn’t tell us much. Let’s examine the top 15 pitchers from 2021:

Rank	Pitcher	Starts	Adj GS2	GSDev
1	Corbin Burnes	29	65.2	14.71
2	Zack Wheeler	32	62.9	14.00
3	Max Scherzer	33	61.7	13.48
4	Jacob deGrom	15	72.2	13.45
5	Walker Buehler	37	60.0	13.12
6	Brandon Woodruff	31	61.0	12.60
7	Robbie Ray	32	60.2	12.26
8	Kevin Gausman	34	58.9	11.81
9	Gerrit Cole	31	59.4	11.59
10	Nathan Eovaldi	36	58.0	11.55
11	Logan Webb	28	60.0	11.34
12	Carlos Rodon	25	60.5	10.95
13	Charlie Morton	37	56.6	10.77
14	Julio Urias	34	57.1	10.60
15	Sandy Alcantara	33	57.2	10.54

Note that while Ray’s seventh-place finish may seem disappointing for a Cy Young winner, the six pitchers ahead of him were all National Leaguers, so GSDev agrees with the voters (in both leagues, since Burnes won the NL Cy).

The start totals for the top 15 get as high as 37, which may look awfully high for a season as recent as 2021. Remember, these numbers include the playoffs, for good (Webb, 2 postseason starts, 14.2 innings, 1 run, 17 strikeouts, 1 walk) or for ill (Cole, 1 start, 2 innings, 2 HR, 3 runs allowed).

To get a better sense for the unique aspects of GSDev as a system, here are 2021’s top 15 pitchers listed along with their results and rankings in bWAR and fWAR:

Pitcher	GSDev	GSDev Rnk	bWAR	bWAR Rnk	fWAR	fWAR Rnk
Corbin Burnes	14.71	1	5.3	8	7.5	1
Zack Wheeler	14.00	2	7.5	1	7.2	2
Max Scherzer	13.48	3	6.1	4	5.4	5
Jacob deGrom	13.45	4	4.5	15	4.9	9
Walker Buehler	13.12	5	7.1	2	5.6	4
Brandon Woodruff	12.60	6	5.3	10	4.7	11
Robbie Ray	12.26	7	6.9	3	3.9	24
Kevin Gausman	11.81	8	5.2	11	4.8	10
Gerrit Cole	11.59	9	5.7	5	5.2	6
Nathan Eovaldi	11.55	10	4.3	16	5.7	3
Logan Webb	11.34	11	3.9	25	4.1	19
Carlos Rodon	10.95	12	5.1	12	5.0	8
Charlie Morton	10.77	13	3.9	24	4.5	12
Julio Urias	10.60	14	4.8	14	5.0	7
Sandy Alcantara	10.54	15	4.2	18	4.3	15

The top 15 in GSDev includes 11 of the top 15 in bWAR, and 13 of the top 15 in fWAR. The highest-ranked pitcher in either WAR system who’s absent from the GSDev list is Ranger Suarez (ranked #6 in bWAR), who was a swingman in 2021 (12 starts, 27 relief outings); GSDev only counts his starts and therefore places him at #40. Next up is bWAR #7 Lance Lynn, who lands at #19 in GSDev; his FIP was notably worse than his ERA, and he got blown up in his only playoff outing. The only entry on the GSDev list who doesn’t make either of the others is Logan Webb, who had the two spectacular postseason starts we mentioned earlier. (If playoffs are removed, Lynn closes about half the distance to the top 15, and Webb falls out of it entirely.)

Most of the major discrepancies between bWAR and fWAR (Ray and Eovaldi in particular) see GSDev end up somewhere in between. The major exception there is Burnes, in whose case GSDev sides with fWAR in putting him at the top of the list. Here we have another aberrant instance of bWAR’s handling of the fielding adjustment. Burnes led the majors in ERA, at 2.43, but bWAR adjusts for him benefiting from a good Milwaukee defense to the tune of 0.36 runs per 9 innings. However, Burnes’s BABIP allowed was .309, worse than the major league average of .292. And his FIP? 1.63, about a third lower than his already-major-league-best ERA.

Well, Burnes had the best ERA among qualifiers, at least. The biggest discrepancy between GSDev and the combined efforts of the two WAR systems is Jacob deGrom, who GSDev sees as the fourth-best starter in the league despite his start total being barely half of the next lowest on the list. deGrom’s 2021 rate stats are the stuff of legend: 1.08 ERA, 1.24 FIP, 0.554 WHIP, 13.27 K/BB ratio. His respectable ranking in both WAR systems speaks to that high level of effectiveness, but even with that, there’s a limit to the number of wins you can contribute in 92 innings, and deGrom hits it.

But remember, GSDev doesn’t handle playing time the same way WAR does. As you can see from the formula, the playing time adjustment isn’t a straightforward multiplication by either innings or starts; instead, it increases with the square root of starts. (For the sake of comparison, Walker Buehler’s 207 regular season innings are 2.25 times deGrom’s total. Even adding Buehler’s four playoff starts to the mix, his playing time adjustment in GSDev is only 1.60 times higher than deGrom’s.) As such, GSDev will be more lenient with pitchers who miss time, as long as their rate stats are sufficiently impressive.

Remember, GSDev isn’t estimating wins added, it’s estimating statistical confidence in the pitcher’s ability. And if you’re asking the question “which pitchers definitely didn’t stink in 2021,” it’s tough to get too far down the list before landing on deGrom.

All right, that’s one season of GSDev totals, which gives a general idea of the scale we’re looking at (basically, over 10 is very good). But even one season isn’t an ideal sample. Let’s look at the top pitchers in GSDev for a smorgasbord of other seasons – say, every 15 years working backward from 2021:

Year	#1	GSDev	#2	GSDev	#3	GSDev
2021	Corbin Burnes	14.71	Zack Wheeler	14.00	Max Scherzer	13.48
2006	Johan Santana	16.05	Brandon Webb	13.98	Chris Carpenter	13.11
1991	Roger Clemens	16.67	Tom Glavine	13.19	Jose Rijo	12.23
1976	Frank Tanana	15.64	Vida Blue	13.73	Tom Seaver	13.71
1961	Whitey Ford	11.79	Camilo Pascual	11.53	Sandy Koufax	10.95
1946	Hal Newhouser	18.35	Bob Feller	17.67	Spud Chandler	12.58
1931	Lefty Grove	18.83	George Earnshaw	14.63	Lefty Gomez	11.29
1916	Walter Johnson	15.51	Grover Alexander	15.12	Babe Ruth	12.60
1901	Cy Young	14.43	Christy Mathewson	10.39	Vic Willis	10.35

That gives a good sense of the typical performance of high-level pitchers. The average #1 season from 1901-2021 has a GSDev of 16.1; #2 stands at 14.4, and #3 at 13.1. But unsurprisingly, there is significant variance from year to year in terms of the top scores. Ford’s #1 season in ’61 wouldn’t have made the top three in several of the other years listed above, while Feller’s second-ranked score in ’46 was easily good enough to win many years.

Of course, not every pitcher can stand among the league’s elites, and even the ones who can won’t manage to do it all the time. What do some of the lower scores look like? Let’s go back to 2021 and take a look at some pitchers who end up in the neighborhood of some useful benchmarks.

10 points can be casually labeled as an ace-level season; the number of pitchers with a GSDev of 10 in a given season has never matched or exceeded the number of MLB teams in the year in question. For 2021, our sample 10-pointer will be Jose Berrios (who actually scored a 9.79). He made 32 starts and pitched 192 innings, with an ERA of 3.52 (123 ERA+) and an FIP of 3.47. He had 3.4 bWAR, 4.0 fWAR, and got a few down-ballot Cy Young votes. Berrios might have ranked a bit higher had he not hit 15 batters, which pushed his average adjusted Game Score down by almost a point to 56.3.

8 points is a very solid year, and 2021 gives us a pitcher with a GSDev of 8.00 to work with in Luis Castillo: 33 starts, 187.2 innings, 3.98 ERA (117 ERA+), 3.75 FIP. Castillo was quite sturdy (average Game Score 53.3) despite leading the NL in both losses and walks allowed. Once GSDev dips below 10, you also start to encounter more part-timers, such as Trevor Bauer (8.30). His 2021 season included only 17 starts (107.2 innings), with an excellent 2.59 ERA but a 4.03 FIP, largely thanks to 19 homers allowed. Despite the longballs, Bauer’s average Game Score was 59.1, a smidge better than Kevin Gausman (who had double Bauer’s starts and made an appearance in the middle of the top-15 table that opened this post).

A GSDev of 5 is respectable but not especially impressive. Take Casey Mize’s 2021 season (5.14 GSDev): 30 starts, 150.1 innings, 3.71 ERA, 4.71 FIP (thanks to 24 HR and a relatively modest strikeout rate). Not someone you dump from the rotation right away, but the average Game Score of 49.4 means you’d rather not start him in a playoff game if you can avoid it.

Once you get down to 3, the pitchers are scuffling quite a bit. 2021 gives us a fun pair of 3-point GSDevs for comparison: a struggling Tarik Skubal (29 starts, 143.2 innings, 4.39 ERA, FIP north of 5, over 2 homers per 9 innings, average Game Score 46.2, 3.10 GSDev) and an injured Chris Sale (12 starts including the playoffs, only 51.2 innings with an ERA of 4.01 and a closely matching FIP, 48.7 average Game Score, 2.93 GSDev). At that point, what odds could you have gotten on Skubal and Sale both winning Cy Young awards three years later?

We can go lower! 2021 provides an excellent example of a scrub-level pitcher, in this case one with 30 starts and a GSDev of 0.04. JA Happ, come on down: 30 starts, 152.1 innings, 5.79 ERA, 5.13 FIP, 177 hits and 30 homers allowed. As you’d expect, Happ’s average Game Score was right at the baseline level of 41.1. There’s a reason this was his last year in the big leagues.

And speaking of final seasons, our last entrant here was also on the verge of retirement. The pitcher with the lowest GSDev of the 2021 season was Jake Arrieta, who produced 24 starts, 98.2 innings, a 7.39 ERA and a 6.17 FIP. Per 9 innings, he allowed 11.9 hits, 2.2 homers, and 4.0 walks; those results range from “bad” to “very very bad.” It adds up (or subtracts down) to an average Game Score of 34.3 and a GSDev of -3.73, one of the worst ever for a pitcher with double-digit starts.

With the scale established, we can get to some of the really fun stuff. Next time, we’ll look at GSDev’s version of the top 100 starting pitching seasons since 1901.