About All These Velocity Spikes
Yesterday, I wrote about Madison Bumgarner looking like he might have been throwing harder than usual on Sunday. Then Jeff chimed in, noting that most everyone in that game had increased velocity, and so perhaps there was something going on with Arizona’s radar calibration. In the comments of those pieces, several people noted that the ESPN broadcasters had mentioned that MLB was changing the way velocity was tracked this year, and thus we should expect higher readings in general this year than we’ve seen in the past.
Now that we’ve had the second “Opening Day” of the season, we now have data from 14 parks instead of just three, which gives us a chance to look at whether Arizona was an abnormality on Sunday, and whether the broadcasters were correct that we’re going to be seeing higher velocity numbers this year than in the past. And while we’re still dealing with small samples, if there was a systematic measurement change, it should show up pretty quickly. So let’s take a look at some numbers.
Last April, the average fastball was thrown 92.2 mph, based on the data we pulled from MLB’s PITCHF/x files. Through the first 14 games of this season, the average fastball was thrown 93.3 mph. And it’s not just fastballs. Curves have gone from 77 to 78 mph, sliders from 84 to 85 mph, and cutters from 88 to 89 mph. Looking at April to April velocity, everything is up about one tick from the same month a year ago.
To eliminate the chance of this just being the result of a new wave of hard-throwing pitchers who happened to pitch in their team’s first game of the season, but didn’t pitch last April, I pulled the data for every pitcher who has already pitched this year and threw a pitch in the first month of the 2016 season. 97 different pitchers have already recorded data in both April of 2016 and 2017, and the collective average fastball for these pitchers has gone from 92.3 last year to 93.2 this year, basically matching the league-wide change.
And it’s not just a few guys with huge moves at the top pulling up the average. 47 of those 97 pitchers had a first game fastball tracked at least one mph faster than their April fastballs were a year ago, and 70 of the 97 were up at least 0.1 mph. Only eight pitchers were down at least one mph from where they were last April, while 20 guys were up at least two mph.
I should note that Bumgarner is 11th on the list of bumps up from last year, and the top few guys (Raisel Iglesias and Jorge de la Rosa) are relievers who were starters a year ago, so among guys pitching in the same role, Bumgarner has had one of the largest measured changes in velocity so far. But it seems pretty clear that this isn’t just a bunch of guys throwing weighted balls over the winter and coming into camp throwing harder now; the data supports the idea that the velocity readings we’re getting this year are just going to be faster than the ones we’ve gotten previously.
In our live blog yesterday, a reader noted that the Brewers broadcasters made a similar comment to the ESPN broadcast on Sunday, suggesting that MLB was changing the point at which velocity was measured this year, so viewers should expect higher velocity numbers due to that change. Since two different TV broadcasts made the same comment, it seems pretty clear that this was something they were told in preparation for the season, so I asked Mike Petriello of the MLB Statcast team for some clarification.
Mike noted that MLB hasn’t actually changed the point at which they’re measuring velocity, but that this year, the conversion from PITCHF/x to Statcast has been completed, and all the tracking data in Major League stadiums is now coming from the Statcast system. For the last two years, PITCHF/x cameras continued to capture data as they had previously, and since sites like ours and BrooksBaseball had been setup to collect data from the MLB logs during the PITCHF/x era, our sites continued to pull PFX data up through last year.
This year, however, with Statcast officially replacing PITCHF/x in the big leagues, the data being pulled publicly is now from the Trackman radar. While PITCHF/x velocity numbers were reported at a defined point along each pitch’s trajectory — usually at the 55 foot mark — so that velocity didn’t have to be calculated at every x/y point along the pitches path, Statcast outputs the highest velocity that Trackman records along the flight of the pitch, which due to physics, is going to be immediately after the ball leaves the pitcher’s hand.
So what the broadcasters have said is true; MLB is essentially measuring velocity out of the hand this year, while the publicly collected data from prior years was measuring velocity at a point a few feet closer to the plate, so the readings showing up on TV and on FanGraphs in 2017 will be higher than they were in the past. But this wasn’t MLB moving the point at which they were measuring velocity; this was just the result of switching from PITCHF/x to Trackman, and the different outputs from a camera-based system to a radar-based system. If you compare Statcast velocity from last year to Statcast velocity from this year, you can get around this systematic shift, and more easily compare apples to apples.
But since we don’t have Statcast velocity for as many years as we have PITCHF/x velocity, it’s still useful to know how to compare them on the same scale, and it looks like the adjustment is on the order of about one mph. So if a pitcher’s recorded velocity is one tick faster this year than it was a year ago, he’s probably the same. And the magnitude of the change for a guy like Bumgarner, who was up 2.5 to 3.0 mph on all his pitches in his first start, is less extreme than it might seem on the surface, though it still seems likely he was throwing harder on Sunday than he was a year ago.
The other side of this coin is that guys who are just down a little bit this year are probably more worrisome than the raw numbers make it look. Sam Dyson, for instance, had a recorded fastball of 93.9 mph in his appearance yesterday, which isn’t that far off the 95.2 that we had for him a year ago. Except 93.9 in Statcast velocity is more like 92.9 in PITCHF/x velocity, which is where the 95.2 came from. And indeed, Baseball Savant — which only uses Statcast data — has Dyson down over two ticks from where they had him last year. So, yeah, it’s just one outing in early April, but Dyson is a guy to keep an eye on, given that he was throwing diminished stuff and got rocked yesterday.
The overall story, though, is that you should indeed expect to see higher velocity numbers from Statcast this year than you’ve expected from PITCHF/x over the last decade, on the magnitude of about one mph. If a guy is up in that range from a year ago, it’s the same stuff. If he’s up a few ticks, he’s probably up one tick. And if he’s down relative to last year, well, then you hope it’s just early and he gets his stuff back, and that you hope your team didn’t recently give him a $206 million contract for his decline years.
Dave is the Managing Editor of FanGraphs.
Does anyone know what the change is? Was the math just wrong before, or are they averaging the speed from release to plate now or something? What would make it go up?
“suggesting that MLB was changing the point at which velocity was measured this year”
Yes, this. Physics says that a ball will lose velocity as it travels, so an earlier measurement point will give a slightly faster number.
I think Dave Cameron wrote an article to address this very question recently. IIRC, it was published on, oh, April 4, 2017, in the early afternoon EDT.
This was talked about pretty extensively way before April 4. I love fangraphs but is this the only baseball blog anyone reads?
Whoosh!
frivo, the comment was a shot at the original poster who clearly didnt read the article he commented on.
Frivo *is* the original poster.
STAAAAAAAAAAHP
I was just wondering, does anyone know if there’s been any apparent change in velocity this year over last year?
Would this change impact all pitchers equally or would reported velocity changes be different for different types of pitchers?
Pitchers who throw with good extension would see less of a spike, especially if they are realtively tall to begin with because their actual release point will be closer to the 55 ft from home plate than a short guy with poor extension whose readings should show a more drastic shift towards more velocity.
If it measures it “immediately after the ball leaves the pitcher’s hand” then extension shouldn’t matter in the reading, should it? The measuring point should vary (pretty slightly) around the 55 ft mark. Or wherever.
A player with poor extension would be further from the 55 ft. cutoff upon release, whereas a player with good extension would be closer. Thus, the poor extension pitchers should theoretically see a bigger spike.
But it doesn’t measure at the 55 ft mark. It now measures at its fastest point, “immediately after the ball leaves the pitcher’s hand”, whether that’s at 55’3” or 54’9” or wherever exactly. PitchF/X was at 55 ft, now it’s Statcast.
Think a little harder Richie, they are saying it would have an effect because if one guy releases near 55′ his new number will match last year’s number, but a short extension guy who releases at 56′ will now read faster because last year his pitch traveled a foot before the reading.
OK, now I see what they’re saying in this thread. Other than we are talking about an inch or three here, rather than a foot. And from the (anecdotal) data we’re given in the article, it doesn’t initially appear like there’s much of an extension factor.
oh i dunno an inch or three would represent a pretty big extension factor for me.
You are correct more or less. A few feet of extension would not make much of a difference. I could down-vote you, or acknowledge that you make a reasonable point.
Maybe also differences between parks in terms of air density — pressure, temperature, humidity — since we’re really talking about the impact of drag on velocity for the span of say 2-4 feet. Dry, warm Arizona should have less drag than say cold, damp Minnesota.
Carter Capps releases the ball halfway to home plate, so I wouldnt expect the two measurements to be much different for him.
This would apply to pitchers who are throwing objects dealing with friction/drag… so those guys.
Is there anyway to adjust that into past velocities on the player and leaderboard pages to match the conversion?
IIRC there is release point data. All you’d need is this data and then an equation for moving objects with friction. You probably wouldn’t get it done perfectly but pretty close, since we are only talking about a couple of feet maybe.
Seems like if you use Baseball Savant to compare 2016 and 2017 it should be accurate?
League average 4 seam velo:
2017 – 93.26 MPH
2016 – 93.12 MPH
So does this mean that measurements of pitchers who release the ball closer to the plate will be less affected?
This might have interesting effects that we aren’t really anticipating.
At least some of the ultimate velocity difference between a fastball and a changeup or other offspeed pitch doesn’t come from arm speed, but from the different spin imparted on the ball. That difference from spin, however, probably starts at 0 at the point where the ball is still in the pitcher’s hand and reaches its maximum as or after the ball crosses the plate, increasing as the ball travels and different spin rates and directions have varying effects on different types of pitches.
By measuring velocity at the pitch’s fastest point, before the effects of spin can play any part, we might be losing a lot of relevant information.
It could be that the overwhelming majority of the difference in velocity between pitches is just arm speed, and we won’t really notice this, but I suspect there will be something here, even if it’s small.
This should be testable. If the new measuring system has changeups picking up more additional velocity compared to last year than fastballs, it’s probably this.
Adding to that, velocity at a fixed point makes all pitchers equal. “Time to plate” from 55ft is just a factor of velocity. Now “Time to plate” involves release point and velocity, so two “96 mph” fastballs will arrive at potentially fairly different times.
How did PITCHf/x used to measure at 55 feet from the plate? Most pitchers release the ball closer than 55 feet from home. Am I missing something?
Will FG player pages keep using PitchFX? Will they be updated with a new section for Statcast?
PITCHf/x reported the pitch speed at 50 ft from home plate (not 55 ft). If now the pitch speed is being reported at 55 ft, it will be higher by an amount I estimate to be about 0.9%. Thus, a pitch with a speed of 95 mph at 50 ft will have a speed of 95.9 mph at 55 ft. Similarly, 80 mph at 50 ft will be 80.7 mph at 55 ft.
Is Pitch F/X data still being collected, or are we only left with Statcast now?
Could this impact pitch movement measurements? Gausman’s fastball recorded an insane amount of horizontal movement yesterday. In 2016 he didnt throw a single fastball that had more than 12 inches of arm side movement. In yesterday’s start almost all of his fastballs recorded over 12 inches, with a decent amount being over 15 inches. Britton, Brach and Givens also all had major upticks in horizontal movement on their fastball.
PITCHf/x always reported the movement starting at 40 ft from home plate. If Statcast/Trackman is now being used, then the movement would have to be calculated at the same distance to be comparable. From what I know about Trackman, the method of calculating movement is different from that used by PITCHf/x. In the past, I have been critical of the latter for not removing the effect of drag. I think Trackman does that much better. So, the numbers may turn out to be different, especially for vertical movement (see http://baseball.physics.illinois.edu/Movement.pdf) which is systematically too positive for PITCHf/x by 1-2 inches.
What’s the difference between Trackman movement and PitchFX movement methods?
Read the article I linked. PITCHf/x does what I described there. Trackman directly determines deviation of trajectory from straight line and uses an approximate procedure to remove the effect of gravity. Their technique is roughly equivalent to the method I described in the article.
I should clarify my comments about movement and the comparison between PITCHf/x and Trackman. The movement in the Gameday data used to be provided from PITCHf/x and now is provided by Trackman. Both are calculated in exactly the same way, from the constant acceleration fit to the trajectory. In particular, both are calculated from y=40 ft. Neither remove the effect of drag, the criticism in the article that I linked to in an earlier comment. As I said yesterday on Twitter, calculating things in the same way does not guarantee that the results will be the same since the two measurements use completely different measurement systems (cameras vs. radar). In a perfect world, with no measurement error, the two would agree exactly. For people with access to Trackman data from 2015 or 2016, you can do a pitch-by-pitch comparison (the Trackman data are not public, so not so many people can do this comparison).
In my earlier comment, I talked about Trackman have a different way to measure movement. That is true, but those numbers are only part of the internal Trackman data (not publicly available) and not part of the Gameday data. I apologize if I caused confustion.
I ran a quick test on change of fastball speed so far. Median change from ’16 to ’17 is +0.4 mph, Average at +0.32 (min 10 pitches in ’17). If I include all fastballs (183) the numbers are: Median: +0.4, Average: +0.28. With aging factored in, probably a +.5 mph bump from 2016.
Better to tell us the mean ratio (new/old) rather than mean change (new-old), as per my earlier comment.
Are the percentages of innings pitched by starters/relievers so far in 2017 similar to the percentages pitched in your 2016 sample? I would assume if you had two sample wear it was a 70/30 split the average velocity would be lower than a 60/40 split. The more innings pitched by relievers I would assume would mean a higher average velocity.
Does this also change pitch movement measurements?
http://tangotiger.com/index.php/site/article/pitch-velocity-new-measurement-process-new-data-points
>So, yeah, it’s just one outing in early April, but Dyson is a guy to keep an eye on, given that he was throwing diminished stuff and got rocked yesterday.
I feel like 0.2 IP isn’t really a large enough sample size to warrant concern.
Can someone dig up Bob Feller’s quarter horse so we can set a baseline?