Buy The Book
A reader asked me to crowdsource this. This is like one of those jellybeans in a jar experiment. We really don’t think much of it, but get a crowd together, and maybe we can figure it out. If you have real data, DON’T post it in this thread. Wait a day or two for that (or send it to me).
Based on the first 161 votes the CROWD has no idea what the answer is . Maybe the question is too hard as it is not something people pay much attention to and so the answers are random unlike a jellybean problem where everyone can make a reasonable estimate
I agree, this is a very hard one, exactly like the jellybean problem.
The mean estimate is somewhere close to 55%. If it’s later shown that that is the correct answer, then that’s a huge vindication. If not, then we can see that there’s a limit to crowdsourcing.
Crowdsourcing will not work if, obviously, if there is a natural bias that the crowd has, which is likely the case here, although I have no idea in what direction it would be. If there is no bias, and there is very little knowledge within the crowd (which seems to be the case here), and that could also very well be the case, then I suppose that the crowd will get still the mean right, even if the spread is high, as it is here, as long as there is some knowledge, even if it is subconscious (e.g. everyone thinks they are taking a wild guess even though somewhere in our memories, we know the answer since we have watched thousands of games).
From what’s recorded in MLBAM Gameday data 2007-2011:
strikes all-fouls foul-tips ratio
0 10.85% 0.35% 3.2%
1 19.42% 0.67% 3.4%
2 23.38% 0.91% 3.9%
I am including foul tips in all fouls. I don’t know what the exact definition of foul tip is for the MLBAM stringers, nor do I know how accurate they are at implementing that definition.
I also would guess that by using multiple choice on this particular question, you will negate most of the effects of crowdsourcing and the ability to average across people.
You’ll also probably see more of that comparison effect that’s reported.
I can take my numbers for post #5 and make a crudely educated guess. Let’s say that foul tips are those foul balls that come off the bat within +/- 5 degrees vertically. If the vast majority of foul balls come off the bat within +/- 50 degrees vertically and are evenly distributed within that range, then 10% of foul balls would be foul tips. Since just over 3% of foul balls are foul tips caught by the catcher, then catchers are getting 1 out of every 3 foul tips.
Right now I can’t think of another good way to guesstimate this from the data.
I will say, though, that “10 DROPPED for every one caught” seems unrealistic. That would mean that nearly 40% of all foul balls are foul tips. That seems way too high.
4/MGL wrote: “(e.g. everyone thinks they are taking a wild guess even though somewhere in our memories, we know the answer since we have watched thousands of games).”
That isn’t really consistent with what we know about memory, though. Something that subtle probably isn’t being encoded into our memories - our memories simply don’t record a lot of the stuff that’s deemed unimportant. (The subconscious is usually used to refer to something that *is* important and recorded but can’t be consciously recalled for reasons of trauma or what have you. Foul tips don’t really fit that criteria, i think.) So even if we’ve watched thousands of games, it’s unlikely that we have any subconscious awareness of how many foul tips are caught.
It’s far more likely that these are totally blind guesses.
But that’s the same jellybean reasoning. It’s a total blind guess, but then you see these results:
(And I’d go for median, not average.)
It seems like the foul tip question is much more in the realm of a blind guess than the jelly beans. They could see the jelly beans; by definition that’s not a blind guess.
With jelly beans you can estimate volumes and so forth, either explicitly or implicitly. The worst jelly-bean guess was only off by 5x, and a couple of the individual guesses were within 5 percent. The guesses for the foul tips are much more widely distributed, which tells you we know less about counting foul tips than we do about counting jelly beans that we can see.
With the foul tip question very few of us had any information that was helpful to even make an educated guess.
The right way to do my guesstimate in #7 would be to vary the offset in a ball-bat collision model and look at the resultant trajectories to see which extreme ball-bat offsets result in trajectories that come near the catcher. That would be a far more robust method of estimation.
"It’s far more likely that these are totally blind guesses.”
Surely there is SOME degree of knowledge within our crowd about this subject, even if it is very small. Even from playing perhaps. Now, how much knowledge is necessary in order for a crowd to arrive at the correct answer, I don’t know. I imagine it is small…
I think the problem here is the range of answers. We’ve done tons of these polls here before, and you’d always get a more bell-curve type of distribution. So, you’d have some confidence that the results would be meaningful.
In this case, the distribution looks very uniform. The four most popular answers are 80%, 67%, 33%, 20%. Basically, the mirrors around 50%. Even 50% was very low.
People believed that it was either lopsided to one side or the other.
It is certainly a very strange result.
Even so, there is a glimmer of hope that the real answer is at around 55%. It’s not that far off from a purely random answer.
I said 2 caught for every one dropped. My initial feeling was that they are equal in number, but that the dropped ones occupy greater importance/memorability in our minds because of their impact.
When it’s a caught 3rd strike (foul tip), it’s just another strike out, not a “oooh, we shoulda had that one”.
I like these caught third situations because I always get to say “good thing we worked on those yesterday.” That joke never gets old. It does, but what the hell.
They get caught quite a bit, and without detection. It could be greater than 2:1 and I would not be surprised. It’s not something that really sticks in the memory.
Perhaps one of the problems here is the ambiguity about what a “dropped” foul tip is (and therefore, what a foul tip is). Any ball the catcher gets a glove on? What about a ball that’s fouled straight back but does not make contact with the glove? The range of answers might be narrower if the question were more clear.
It would also be interesting to see % for individual catchers and/or pitch types in this regard.
I would imagine fastballs have the greatest chance at being caught foul tips since breaking pitches are more likely to be “topped” or “redirected” (like a hockey deflection).
I’d love to see how this compares to similar stats of the past, specifically in regards to changes in catchers mitts. No longer are they the heavy gloves with “big bumpers” but more like a hybrid of mitt and 1st basemen’s glove with “reflex shut” or “praying mantis type action. My sons catchers mitt is like this and it’s amazing. The ball hits the pocket and the glove seemingly slams shut on it’s own without big bumpers for the ball to bounce off of.
This technology also is used by 1B’s in how they’re taught to dig with a backhand motion, letting the glove’s pocket action work, rather than using a forehand motion and risking “flipping it up” with the shovel part of the mitt.
If he didn’t make contact, it’s not dropped, is it?
It’s like a WR in football. He caught it or he dropped it. But if he never made contact, it doesn’t count. Errors in baseball as well, etc.
I’m looking at some data now (1993-2009). I looked at all foul-pitches with 2 strikes. Of those, this is how often the batted fouled out, as a percentage of all foul-pitches:
With…
3 balls: 0.1%
2 balls: 0.3%
1 ball: 1.0%
0 balls: 1.9%
Now, why would this be? Why would a batter at 0-2 foul out on 1.9% of his foul-pitches, while he would foul out only 0.1% of his 3-2 foul-pitches?
Remember, I’m doing foul-out divided by foul-pitches.
Or, is it possible there’s some bias in the data?
Because of the quality of the pitch and swing? On 0-2, he is barely getting a piece of the ball and fouls out on some foul tips and some weak foul balls behind and near home plate. On 3-2 counts, he is hitting harder fouls and more pitches in the center of the plate, so fewer foul tips and more foul balls down the lines and even into the OF seats. He is also looking more fastball on 3-2 counts (and 2-2 as compared to 0-2 and 1-2) and when he gets an off-speed pitch, he is pulling it foul into the seats. I am just guessing…
That is quite a stark contrast though, no? 20 times more likely to foul-out, given that we only look at foul pitches to begin with, at 0-2 than 3-2?
I think there has to be some bias or error in the data. I agree with everything MGL says, but the BABIP difference at 0-2 vs. 3-2 is only about .020 (.285 vs .305). Batters get better/harder swings at 3-2, but not 20x better. I also find it implausible that only 1 foul ball in 1,000 becomes an out at 0-2. Seems like something is wrong with the data.
And don’t you also want the numerator to be PAs that end in a K (and foul), as opposed to all foul outs?
Pitch type and location would be very different 3-2 vs. 0-2. At first blush it doesn’t see like that would be enough to explain the effect, though.
If you’ve got data from 1993-2009, that’s probably from at least two different original data providers, right? Project Scoresheet and MLBAM Gameday, perhaps, if it came from Retrosheet. Assuming that’s the case, do they both show the same effect?
Tango - How are you writing your query or queries for this info?
Tango, either there is something wrong with your query or something wrong with your data source.
I get the following from Gameday for 2007-2011, including some spring training and playoff games.
3 balls: 1409/43575 = 3.2%
2 balls: 1724/62461 = 2.8%
1 ball: 1679/62611 = 2.7%
0 balls: 861/35553 = 2.4%
select p.ball, count(pitch_id) as NUM from pitches p, atbats a
where p.ab_id = a.ab_id and p.type = "X" and p.strike = 2 and a.des like "% foul%"
group by p.ball
select ball, count(pitch_id) as Num from pitches
where des like "Foul%" and strike = 2
group by ball
Mike’s numbers are very close to what I am getting for 2000-2010 regular season from Retrosheet. Mike, unless I am missing something in your code, it doesn’t look like you are including foul tips or foul bunts in your data, just foul flies, liners, and pops caught in the air.
Peter/25, I think you are right. My query did not include foul tips, and I’m not sure about foul bunts without checking, but probably not.
I was using data I dumped from REtro 1993-2009 for other purposes. It’s possible I merged two different categories into 1 that made sense for my original purposes, but then makes no sense here.
I’ll have to take a closer look tonight. Thanks guys…
May 25 10:14
Largest demonstration in Canadian history?
May 25 09:39
What sabermetrics is NOT
May 25 09:31
Do pitcher’s reach back for velocity when needed?
May 25 06:39
Lack of hustle during a game
May 25 02:38
NFLPA lawsuit against collusion
May 25 01:43
Neal Huntington’s best moves
May 24 23:50
Rooting for laundry
May 24 17:04
Firefox, IE, or Chrome?
May 24 12:07
How to beat the shift
May 24 11:11
Incredible story
A huge selective perception problem here—we only pay attention to 2 strike foul tips. But how many 0 and 1 strike foul tips are there? With those counts, we rarely pay attention to the catcher catching it or having it knock off the bottom of his glove and into the dirt.