All about the high inside fastball

Can we please see an analysis like this in regards to the count? Sequencing’s a start, but batters and pitchers are geared to expect certain pitches/speeds/locations on certain counts.

It appears that some pitchers are more skilled than others at throwing a given pitch for the desired effect on a given count (yes, that is a reference to John Lannan - I believe his ability to outperform his FIP comes from being able to throw any pitch on any count with a greater-than-average consistency) than others, but I’d like to *know* that.

Obviously, I misposted.  For LHH, the numbers are:

0   0   29%  -2.4848828
0   1   14%  -2.0650778
0   2   5%  0.7196444
1   0   12%  -3.1323764
1   1   11%  -1.9516521
1   2   8%  -0.5832457
2   0   4%  -4.4612756
2   1   5%  -4.0800488
2   2   7%  -3.0926844
3   0   1%  -5.1016326
3   1   2%  -6.3334085
3   2   4%  -6.9044217

The conclusions are the same, maybe even more pronounced.

My best guess? On 0-2 and 1-2, hitters are looking to just make contact whereas on favorable counts, they tend to grip-it-and-rip-it, knowing they have wiggle room. Also, I rarely see pitchers go that route—it’s usually high-hard-and-middle or soft-low-and-away. Perhaps there’s a selection bias there, too (i.e. the least effective pitchers try that and fail).

The difference between LHHs and RHHs on 3-2 *seems* to verify the stereotype that RHHs tend to be better high-ball hitters vs. LHHs, but I also wonder if the opposite selection bias is in there too (i.e. the most effective pitchers are able to do this and succeed).

Thanks for crunching the extra data!

Out of curiosity, do walks that result on taken 3-ball counts show up in that data? I would tend to think not, but I would further wonder if the

Great work - for starters.  Any work done on the effect of pitch sequencing has to control for the pitcher and the count and even the batter if possible.  But at the very least, the count and pitcher.

Basically, it has to be done like the “delta method” for aging and other similar-type studies.  For example, with pitcher A on the mound, on a 1-2 count (preferably with the last pitch being the same result, ball or strike), what is the difference between the run value of a fastball after a fastball and a fastball after an off-speed pitch.  Then you add up all those “deltas”.  That is really the only the way to do it.  If you just aggregate data for all pitchers, you run into huge selective sampling issues and you basically get results that have little meaning.

And like I said, unless you control for the batter, you might run into trouble as well.  For example, let’s say that weak batters see a lot more fastballs (fb) in general and strong batters see more off-speed pitches (os).  Your os-fb sequence might be populated by better hitters than your fb-fb sequence, for obvious reasons.

And I would not be surprised if conventional wisdom about throwing down and away (or just an off-speed pitch) after a high hard one is not correct or that changing pitches is not an effective strategy.  Anything that is predictable can be and probably is exploited by batters. I have always thought that going with an off-speed pitch an inordinate percentage of the time (more than would be random) is a ridiculous strategy, even if you are changing the batter’s “eye level” or whatever you want to call it.  I mean if Tim McCarver (or basically everyone in the ballpark or watching on TV) thinks that after a high hard one, the pitcher is going to throw a breaking pitch like 80% of the time, don’t you think that the batter would know that too and would be looking for it?  Depending upon the count and the pitcher of course, the next pitch has to be as likely to be high and hard again as it is low and away or offspeed.  When I say “as likely” I don’t mean 50/50 or all pitches with equal likelihood. I simply mean whatever the count, the pitcher’s repertoire, the batter’s strengths and weaknesses, and the game situation would dictate.

BTW, does rv100 include the value of a ball or a strike as well as the value of a ball in play?  IOW, if a pitch is not put into play, the rv100 value is the average value of the after count minus the before count (times 100)?

Thanks MGL and dks.  I had already planned to do a full article soon looking solely at pitch sequencing, so I’ll employ all of your suggestions. 

However, I believe rv100, or at least the one I’m using, already takes into account the count.  You take the linear weights of each event - the expected linear weights of that count to get the net run value.  I assume that’s a powerful enough adjustment to make up for the difference’s in count.

As for adjusting for pitch and hitter, that makes sense.  I’m not sure why I didn’t do that in the original article (probably just being lazy, as it takes a bit more work).

As for giving the weights of pitches in the strike zone, I’m just not sure that’s relevant, at least for the purposes of pitch sequencing.  I’m already adjusting for count, so that doesn’t really matter in terms of strikes and balls.  I’m probably wrong though.  As for aesthetic purposes, that works, but I’m not sure I want to doctor the numbers too much.

Sure, Davor. 

Like I said, I plan to devote an entire article to pitch sequencing, so I’ll separate by whether the previous pitch was a strike or not, and probably more interesting, what kind of strike was it (slider down in away, curveball in the dirt, fastball down the middle).