More Velocity and K/9 Charting
After this afternoon’s graph, some of you requested a breakdown by starters and relievers to remove any potential bias among soft-tossing relievers getting inflated strikeout rates due to their situational usage. So, here you go – I broke the group presented in the first graph into greater than 100 IP and less than 100 IP, which is a good enough proxy, and redid the graphs.
Obviously, there are a lot more relievers than starters, but the data is basically the same. As you can see, the slope of the regression line is very similar – there doesn’t appear to be a significant difference between starting and relieving in terms of correlation between fastball velocity and strikeout rate. The r (not squared, which I’ve adjusted based on a couple good comments this afternoon) is .43 for starters and .38 for relievers.
Of course, this is for major league players only. How does velocity interact with strikeout rate in the minors? We’ll look at that tomorrow.
Print This Post
It’s interesting that while the slope of the trend lines are nearly identical, their starting points are different. Granted, it isn’t *that* much of a difference, but it reinforces the notion that relievers many times are just failed starters. IE, the starters can get about 0.5 more K/9 for a given fastball speed than the relievers can do at the same speed.
Very cool, great analysis Dave.
I would like to see the slope presented. I’m not sure they are that similar. The trendline for starters starts above 4 and ends up below 8. The trendline for relievers starts below 4 and ends up well above 8, maybe above 9.
Also, noted the cloud distribution with the relievers, breaking them down farther would be helpful. There are some pretty extreme outliers there (11Ks/85MPH for example)
Agreed that K/BF would be more helpful
Just noticed that both the X and Y scales for both graphs are inconsistent. That’s bad presentation form :0)
Actually, at first glance it looks like you’ve got more of a cloud for the relievers, which would imply little to no correlation between fastball speed and k/9 for them.
To start off my virginal post here, I’d first like to thank you guys for your great work. This site is a goldmine.
In response to Matt, I agree with your point that starters probably have a higher expected k/9 at every mph, but I would tentatively argue that the the spread is greater than .5(k/9)/mph. Dave is treating relief and starting Ks equally. However, data may show that it is easier to accumulate high K/9 stats as a reliever. I would have to defer to studies for actual evidence but my uninformed prediction would be based on factors, such as seeing a lineup fewer times in a single game, changing leagues more often (as a relief pitcher), electing not to conserve pitchers to elongate your start, etc.
If it does not already exist, I would be interested to see a study that tracked k/9 differentials among pitchers who were converted from starters to relievers. I can imagine some difficulties with such a study though…
I agree with Jeff that there is much more of a cloud-effect going on with relievers and it is nowhere near as linear as the graph for starters. For starters, nearly all of them are within 3 K/9 of the expected line, while for relievers there is much more variance as many of them are as far 4 and 5 K/9 away from the expected line. This could be due to a small sample size, which could be addressed by using multiple years worth of data until you reached the 100 IP that the starters have. As has already been pointed out in the other post, situational use likely boosts the RP’s K/9 when compared to the SP’s K/9, but that doesn’t really explain the great amount of pitchers that fall significantly under the line. Those pitchers might be explained as mopup guys and generally poor pitchers who just aren’t good enough to make the rotation, which is why the extreme “underperformers” show up in the RP graph and not the SP graph.
Also, I can’t help but wonder if using something like K% or even Contact% would be a better measure of the effectiveness you’re looking for here. I’ve never liked K/9 because I feel it can be influenced too much by the defense – to use an unrealistic but mathematically simple example:
Pitcher A and B play for different teams (Team A and B, respectively), and both strike out every other batter that comes to the plate
Team A’s defense converts every other ball in play into an out.
Team B’s defense converts every third ball in play into an out
So a 9-inning game for Pitcher A would look like this: (K=strikeout, H=hit, F=fielding put out)
KHKF/KHKF/KHKF/KHKF/KHKF/KHKF/KHKF/KHKF/KHKF (split up into innings)
Getting 18 of his 27 outs from K’s, or a K/9 of 18
Same excercise for Pitcher B:
KHKHK/FKHK/HKFK/HKHKF/KHKHK/FKHK/HKFK/HKHKF/KHKHK
Getting 21 of his 27 outs from K’s, or a K/9 of 21
Despite pitching identical fielding-independent games, the inferior defense of Team B artificially inflates Pitcher B’s K/9, as they supply him with fewer putouts, allowing for him to accumulate more K’s relative to total outs.
First, I want to say the feedback on this site is great. The starters do appear to be somewhat of a tighter bunch with a few less outliers, though obviously velocity is not the only answer.
Terminator – this is going to be a really rough back of the envelope type calculation which someone can fix later, but the difference between the best defense and the worst defense last year was less than 150 runs over the course of the season. If the difference between the average non home run base hit and an out is something like three quarters of a run, that’s 200 extra outs over the course of a season, compared to 4300 normal outs, give or take. 200 outs is probably a high estimate, and that’s a bigger spread than between the worst and the best fielding units in baseball last year. It still leaves the fielding affect at less than 5%. There’s some park effect too from size of foul grounds whatnot, but my guess would be K/9 is still pretty good. It would be a concern if there were any teams with as big of a fielding difference as your teams A and B, but I don’t think it’s the case.
Oh you’re absolutely right that the difference is likely small and this is all close enough for jazz, but my point is merely if we have what is potentially more accurate data readily available, why not use that instead? It seems like Contact% could be a great option to replace K/9
Given that the average LHP throws softer than the average RHP… wouldnt a split of these graphs by handedness be worth looking at?
The strange issue of LHP velocity sorts of skews things – though I dont expect any bias, and I do expect a similar slope…
Terminator is right. K/BF would be a better stat to use instead of K/9.