Since they tend to get the most chances, I chose shortstops.
To do this really simple test I took every shortstop’s Error Rating, and then calculated the number of total Batted Balls in Zone for them while they played shortstop, and how many errors they were attributed. Then I tabulated the data by error rating. Simple, right?
Here’s the result:
| IF Error Rating | E | BIZ | Error Rate |
|---|---|---|---|
| 4 | 1 | 11 | 9.09% |
| 5 | 4 | 142 | 2.82% |
| 6 | 21 | 416 | 5.05% |
| 7 | 36 | 1018 | 3.54% |
| 8 | 154 | 4113 | 3.74% |
| 9 | 150 | 5274 | 2.84% |
| 10 | 190 | 8163 | 2.33% |
| 11 | 74 | 3454 | 2.14% |
| League | 630 | 22591 | 2.79% |
First and foremost, we should immediately kill the “4” and “5” Error data due to sample size. The data for “6” Error should be looked at askew for similar reasons, but at 416 BIP it’s worth at least noting. Even the “7” is not highly represented…and as such represents a selection bias among BBA GMs. If a guy has less than an “8” error, he doesn’t play much SS—which makes this analysis a bit fraught.
If we take this data straight as it is, though, we see that BBA shortstops average out to a 2.79% error rate. If we trust that number, we’d say that the average SS Error Rating is somewhere in the low 9s. Raising to a “10” improves our error rate by about half a percent. Not piddly. Raising to an “11” adds another two-tenths.
The more interesting question, then, is what happens in those lower ranges. How bad is an “8” or a “7” or … and then the more interesting question is how many extra errors does that actually play out to. The last is a little easier to study…at least at the top end, because we can see that our most often played shortstops seem to get 600-700 BIZ. So let’s say 650. A half percent of 650 (the gap between 9 and 10) is 3.25 extra errors. You can do your own math down the line, but if we say three errors are like adding the runs of three additional singles, linear weights would say that value is maybe 1-2 runs.
ISSUES
The more interesting question is down the line, where the “8” and “7” Error Ratings lie, and where the data is a little weird.
I should note there are other issues with this data, too, not the least of which is that there will be variance within the Error Ratings. For example, though, I note that “8 Error” SS appear to create a higher rate of errors than “7” Error Rated SS. The difference is fairly small, though. Does that mean that the average “8” is very close to a “7” and vice versa? Who knows?
But I want to talk about the impact of random variance at these small rates.
For this study, let’s assume the Error Rate we see of “8” is correct. How, then, do “7” Rated players make less errors?
For that, let’s do some mathematical thought experiments. Specifically, let’s say that we think the “real” error rate of guys Rated “7” should be .038. In other words, just a little worse than the guys rated “8.” If we then let 10,000 shortstops with this error rate have 1,018 chances (just like the BBA Shortstops in this band), how often to they make errors at .0354 or less (like BBA shortstops)?
The answer is somewhere around 35% of the time.
That says that there exists a very likely case that the data we see in the “7” zone is an aberration due to sample size. It can go the other way, too. “8” could be a little lower.
So you get to read the data however you want. But one way of looking at the overall span would be to say that the low end error rate is maybe 3.8% and that the high end is 2.1%, meaning that the overall performance span of “normal” BBA shortstops is about 1.7% at the far edges. If we use 650 BIZ as our baseline for playing guys at those edges, then you’d get a variance of around 11 errors.
Is that right?
Dunno.
But Montreal’s Qutuz Mahdi carries an “8” error rating, and he’s made 30 errors on 707 BIZ to date. Phoenix’s Niaz Minhas, with a “9” error rating, has made 22 errors on 635 BIZ. Elvan Masuki ("11"), now with Charlotte, has made 11 errors on 629 BIZ.
If I squint just right, I can see that in these numbers. If anything, maybe that estimate is a little too low. But dealing with randomness at these sample sizes is always a tough thing.
