(Photo by Abelimages/Getty Images)
As those of you who read my game recap know, a funny thing happened on the way to Phil Hughes delivering his first strong outing of the season on Sunday. Following the game, I noticed that his curveball — the one that he told everyone he was going to revert back to utilizing his original grip on, an admission that I was admittedly pretty skeptical of — had racked up -1.1172 linear weights*.
A quick check of the spreadsheet I pulled together this past offseason that contains ridiculous amounts of data from every start Hughes made in 2010 for a series of PitchFX posts breaking Hughes down pitch-by-pitch confirmed that this linear weights tally was better than his curveball in every single start he made least season.
A few days later I went back and checked the game logs for Hughes’ starts in 2009 and 2008 (unfortunately there isn’t reliable PitchFX data for 2007), and it turned out that Hughes’ -1.1172 linear weights from this past Sunday’s game was his third-best ever among the data set available — the only starts he registered better linear weights on the curveball were his final two in 2009 — May 25 against Texas (-1.1946) and May 31 against Cleveland (-1.3905). He was then transitioned to relief and of course pitched out of the bullpen the remainder of that season.
I decided to put together a table showing Hughes’ average linear weights for the curve from his starts during the 2008 season, 2009 season, 2010 season, April 2011 and then the individual linear weights from his two starts back (July 6 and July 17) since returning from the DL.
Now according to Fangraphs, Hughes’ wCB hit a career-high in 2007 at 2.1 runs above average. Brooks and Fangraphs do not utilize the same pitchFX data, but I’m going to go out on a limb and say that his 2007 avg. linear weights were probably roughly in the -2.1000 vicinity, which would make ’07 his most effective curveball year — anecdotally you probably wouldn’t get too many people to disagree with you. However, even without having the exact data available, it’s pretty clear his curveball this past Sunday was among the best it’s ever been.
So what changed?
Horizontal- and Vertical-Break alone don’t tell the entire story, but presumably there’s something to be said for the fact that his curveball broke closer to the plate both horizontally (4.49 inches of H-Break, compared to an average of 6.71 inches across the other five data points) and vertically (-4.66 inches of V-Break, compared to an average of -7.92 inches) than it ever had on average in the past. That’s more than two inches horizontally and more than three inches vertically closer to the plate — it may not sound like much, but that’s a fairly substantial difference. Closer to the strike zone — but not too close — will presumably induce more swings and subsequently lead to weaker contact.
Sure enough, this past Sunday Hughes threw his curve for a greater percentage of strikes — 68% — than at any previous time in the selected data sets, and also recorded the highest Swing%, at 48%. While the relatively low Whiff% might be slightly discouraging, I’d take it as a positive sign that his Foul% was also the highest among these data and the In Play% of 28% was second-highest — given his sterling linear weights on the pitch, this tells me that hitters just weren’t making good contact on the curve.
Now before we get too carried away, this obviously was only one start. Given how ineffective Hughes was at the beginning of the season, I don’t think it wouldn’t surprise anyone if he got beat up in his next outing — he still has a ways to go before he wins back much of the good will he relinquished. However, if this new-and-improved curveball command/control is here to stay, Hughes will hopefully get back on track sooner than we might have expected him to.
*For those not familiar with the concept of pitch type linear weights, Brooks Baseball provides the following definition: “Pitch Type LWTS correspond to how many runs were likely to score on a particular pitch based on average run expectancy when each pitch was thrown and what happened as a result. Negative scores indicate more effective pitches.” All data in this post is from Brooks and TexasLeaguers, and for a more thorough primer on pitch type linear weights, check out this great post on Fangraphs.
If you enjoyed this article, please consider sharing it!
11 Responses to Phil Hughes and the Case of the Curiously Effective Curveball
Leave a Reply
TIQ IQ
Great stuff as always Larry. Was the velocity on the curve noticeably different (faster, presumably), and could that play a role in its success?
Thanks Eric. I didn’t look at velo, though I probably should have. A quick glance at Hughes’ Fangraphs page shows the following avg. velocities:
2007: 71.2
2008: 72.2
2009: 77.1
2010: 75.8
2011: 74.1
Career: 74.8
This past Sunday the curve’s avg. speed was 74.97mph.
Given that his avg. speed on Sunday ostensibly matched up with his avg. speed for his career, I don’t really know what to make of this data. As best I can tell, it would appear that location was really the key for Hughes, at least on Sunday.
Cool, good to know. I was interested because the new curve was supposed to be more of a “power curve,” so I was curious to see if he threw it any harder. Location’s definitely important though.
Larry,
“Now according to Fangraphs, Hughes’ wCB hit a career-high in 2007 at 2.1 runs above average, and so if we assume that Brooks utilizes the same data as Fangraphs, then his 2007 linear weights should be roughly -2.1000, making ’07 his most effective curveball year — anecdotally you probably wouldn’t get too many people to disagree with you. However, even without having the exact data available, it’s pretty clear his curveball this past Sunday was among the best it’s ever been.”
The data is not the same here. Fangraphs is using BIS data for their curveball classifications and brooks uses pitch f/x. The linear weights may not be equivalent as well unless both have updated linear weights. However they should be close enough.
And evaluating pitch effectiveness on a game-by-game basis is pretty iffy. Too much noise, very small samples. Linear weight run values are also subject to mercy of BABIP gods. In addition linear weights do not account for the effect of other pitches.
However, there is reason to believe that less movement is a good thing. Means the curve is less recognizable, smaller hump. I would still like for his curve to be thrown harder.
Thanks Josh — I knew that Fangraphs and Brooks utilized different data are different; I didn’t word that paragraph properly. I also didn’t want to exclude 2007 entirely, but I know it wasn’t a perfect comp.
While you make a good point about the noise involved in evaluating pitches on a game-by-game basis, I still find it useful to know the relative effectiveness of a pitch in a given start. Obviously the key problem with linear weights is that they have no predictive power, which is why you can’t get terribly excited over one outing, but I still think it’s noteworthy that Hughes’ curve was arguably the third-best its ever been on Sunday.
Larry,
Linear weights are strongly tied into babip and the number of hits given up. We don’t pay attention to batting average game to game, so we should apply the same discretion to something that’s strongly tied to babip like linear weights.
A different way would be to use expected run value, where you substitute league average values based on batted ball outcomes or to create a model that predicts run value based on various pitch f/x variables. Maybe just using whiff rate game to game is a better measure, though I haven’t checked to make sure. To test whether his linear weight value from sunday is significant, we could do a hypothesis test.
I get what you’re saying, but not sure why you’re being so dismissive of linear weights. They may be subject to the whims of BABIP to a certain extent, but you can’t write strikes and balls off due to BABIP.
Your analogy of paying attention to BA game to game would make linear weights on an individual game basis essentially worthless. I’m not sure I get where you’re going with that — your average batter has four, maybe five plate appearances in which to accumulate a batting average in one game; your average pitcher throws around 100 pitches, maybe half of which are fastballs and the remainder divided up by whatever else he throws. The sample size is a good deal healthier.
I get that analyzing linear weights on an individual game basis doesn’t tell us anything close to analyzing linear weights on a seasonal basis, but I don’t think they’re as useless as you seem to think they are. What other method is there right now of evaluating a pitch’s relative effectiveness on a game-by-game basis? I for one like knowing that if I watched CC Sabathia throw an amazing start, that if I check out the numbers on Brooks chances are his linear weights will line up with what I saw, and I think there’s value in that. If you know of a better stat I’d be happy to check it out.
Well, if Hughes throws 20 curves in a game, then that’s like 5 games for a batter I guess. We don’t care what a batter does in a week either.
Yes, I think linear weights on a game by game basis are essentially worthless.
And you’re right, it’s hard to get helpful information from one game’s worth of data. That’s something inherent to game by game analysis, and there’s not a whole lot we can do about that. The best things I can suggest to deal with that are as I said in my previous post, expected run value or just straight whiff rate.
I can provide you with the linear weight run values for various batted ball types if you want to do expected run value. It’s like tERA.
Fascinating; I didn’t realize that linear weights on a game by game basis weren’t considered useful. In that case, why does Brooks even carry them?
Appreciate the info; you certainly know more about this stuff than I do. Will keep in mind for future analysis. Thanks Josh.
The most useful stuff on Brooks is the actual data. Results based stats like linear weights and whiff rate make it kind of like the boxscore of pitch fx stats I guess.
Keep up the good work.
[...] Phil Hughes and the Case of the Curiously Effective Curveball … Negative scores indicate more effective pitches.” All data in this post is from Brooks and TexasLeaguers, and for a more thorough primer on pitch type linear weights, check out this great post on Fangraphs. [...]