Nelson's Accuracy

[Thunderheads]

Cool article from Gil over at the UEFL

http://bleacherreport.com/articles/893477-alcs-game-5-rangers-tigers-plate-umpire-falls-two-pitches-short-of-perfect-game

[Umpire in Chief]

Very nice

[mstaylor]

Well written article, and very, very true.

[Umpy]

I disagree that Batter's can't bail you out on strike calls. There have been more than a few pitches where I thanked the lord the kid swung, because I wasn't sure if it was going to nick the edge of the zone or not.

Also, I thought we all thought the computerized strike zone trackers aren't accurate enough, why are we using it as a benchmark?

[Thunderheads]

I disagree that Batter's can't bail you out on strike calls. There have been more than a few pitches where I thanked the lord the kid swung, because I wasn't sure if it was going to nick the edge of the zone or not.

If it's that close ............ steeeeeeeeeeerike!

Also, I thought we all thought the computerized strike zone trackers aren't accurate enough, why are we using it as a benchmark?

We aren't ...it was just an article

[Umpy]

Sorry, I thought we were meant to discuss the article.

[Thunderheads]

Sorry, I thought we were meant to discuss the article.

No, sorry, came across wrong ....

What I was saying is that the posting reason wasn't basically a 'support for benchmarking' using the computer ....

That said, re-reading your reply, ...you make an interesting point ;)

[Lindsay]

The basic premise is that pitch f/x is accurate to within one-half inch, per their CEO. As we're all aware of, the 2D plot generated is normalized pitch location as the ball crosses the front edge of home plate, and doesn't take into account the depth and shape of the strike zone. The plot, by virtue of normalization, does consider instantaneous batter stance as the pitch is captured. Per practicality, the generated plot does not take into consideration pitch count, pitch type, break, target accuracy, etc. In terms of a "rules book strike zone," the generated plot is the closest you'll get to an accurate location as it relates to balls/strikes when the pitch is at the front of home plate.

As such, any plotted pitch that is captured within the rules book strike zone (px: -.953, .953, norm_ht: -1.000, 1.000) as it crosses the front edge of home plate must, by rule, be a strike. Any called strike within the rules book strike zone is "correct," while any ball within the rules book strike zone is "incorrect." Any called strike outside of the rules book strike zone may be incorrect, while any ball outside the rules book strike zone is "correct." Margin of error allows for |.953|+ 0.4165, which is ordinarily capitulated to (0.900, 1.000). Vertical margin of error is variable and calculated for each pitch.

Therefore, a plot which shows an umpire missing a number of pitches located within the strike zone and outside of the margin of error (noted as an incorrect ball call) is accurate. A plot which shows an umpire missing a number of strikes outside of the strike zone and outside of the margin of error (incorrect called strike) may be accurate.

Nelson's plot (Game 5 ALCS): pfx (54-56 Called Strikes, 93-93 Balls = 98.7%)

Nelson's two missed pitches were both fastballs, both off the RH batter's side of home plate, and both with a measured break toward the outside (such that the pitches would not have tailed back into the zone past the front edge of home plate). They both were outside the margin of error, and conclusively (100%) categorized as "incorrect called strike." A performance two pitches short of a perfect game is outstanding. Used here, pitch f/x is valuable, objective, and most importantly, accurate. Pitch f/x can be an umpire's harshest critic, and when that toughest of judges has only two incorrect calls out of 149, it's a good day.

On the other hand, as pitch f/x is an umpire's harshest critic, it will give performances a <90.0% rating from time to time.

Welke (Game 1 ALCS): pfx (50-53 Called Strikes, 87-102 Balls = <90.0%)

By rule, math, and science alone, all 15 ball calls were incorrect. Two of the three called strikes were incorrect. The third is borderline. However, pitch f/x doesn't take into account the aforementioned variables, as well as heat mapping. Heat mapping shows an umpire's tendencies in ruling balls and strikes by pitch location, and is unsupported by the rules. For instance, if heat mapping were to be applied to Welke's plot, though he would still be 50-53 in called strikes, he would be 96-102 in balls for 94.2% overall. His ZE percentage might even have come out to around 94%.

Again, heat mapping deviates from the rules. Nelson's two pitch short of a perfect game plot does not. It is an incredible accomplishment for an umpire call a 100% rules book strike zone, missing only two pitches. Whereas Welke's plot needs manipulation to bring him up to 94.2%, Nelson's needs no work at all to show his near-perfect performance. That in itself is worth noting.

[carolinablue]

I'm glad I dwell in the land of one ball off the edge, the neighborhood play, and no super slow motion.

[Umpy]

So the pitch tracker does track more than just the flat front plane of the strike zone?

[UmpJeff]

I'm glad I dwell in the land of one ball off the edge, the neighborhood play, and no super slow motion.

Me, too! However, around the bases, they need to be tagged for me. :yippie:

[carolinablue]

I'm glad I dwell in the land of one ball off the edge, the neighborhood play, and no super slow motion.

Me, too! However, around the bases, they need to be tagged for me.

Oh, trust me, he got it with the toe.

[Lindsay]

So the pitch tracker does track more than just the flat front plane of the strike zone?

Yes, it tracks start_speed, end_speed, sz_top (instantaneous midpoint), sz_bot (instantaneous knee-hollow), pfx_x (horizontal movement), pfx_z (vertical movement), px (plot's x coordinate), pz (plot's y coordinate), xo (initial x), yo (usually 50), zo (initial y), vxo/vyo/vzo (initial velocity x, y, and z), ax/ay/az (initial acceleration x, y, z), break_y (max. "bend"), break_angle (the angle the pitch crosses the front edge of HP with), break_length (distance bw release point and pz), type_confidence (algorithmic most probable pitch type), release point, speed, rotation, spin, norm_ht (calculated y var for normalized ht plot).

So even with just that one data point on the front edge plot, you can use the others to calculate a pitch's most probable trajectory into the catcher's glove, and determine the likelihood of a pitch entering the strike zone.