THE BOOK--Playing The Percentages In Baseball

Great stuff, MGL.  I missed it—is there a link to Rally’s study?  Are your results similar?

http://lanaheimangelfan.blogspot.com/2008/02/first-basemen-and-saving-errors.html

Also discussed here:
http://www.insidethebook.com/ee/index.php/site/comments/wowy_firstbasemen/

For weighting, I used the actual # of throws but added 100 throws at an average error rate for the “without” part.

Results look reasonably similar.  I had Lee, Olerud, Sexson rated well.  Karros was down at the bottom.  I remember Berkman rating better than Bagwell (they were of course each other’s with and without).  Contrary to Moneyball, Scott Hatteberg is not a scooping machine.

I was looking at more years but with a different datasource (retrosheet).

In this case, the data source should not make any difference as I doubt there are many differences in whether there was a throwing error or not.

Rally, did you use all fielders, including catchers and pitches?  Did you include the back end of GDP’s?  For example, a 643 was treated just like a one-out 4-3, although there were little or no corresponding throwing errors since unless runners advance extra bases, they generally don’t award an error on the back end of a GDP, if the lead runner was out (one of the sillier scoring rules).

BTW, I think this is a revolutionary study, Rally’s that is.  It should be getting more press.  Not that it was hard to do - it wasn’t.  Or that the idea was revolutionary. It is just that people have been talking about the “value of a first baseman’s catching ability on throws” for a long time, and I am not aware of any studies that have attempted to quantify it until Rally’s.  Maybe credit to Tango for giving Rally the WWOY idea (I don’t know).  Anyway, now we know - pretty much.  Again, I don’t know the exact regression, but it looks to me like the skill is worth plus or minus 3 runs a year, skewed a little toward the plus side, which is not uncommon (non-symmetrical talent curve), BTW.  Not a whole lot, but something.  Probably on the same order as baserunning.  Interestingly, baserunning is generally considered (by fans, media, baseball insiders, etc.) to be worth much more than a first baseman’s catching skills (I think) and definitely much more than it actually is.

So pretty much, advice to aspiring first basemen is, be tall.  Sexson, Olerud, Clark, Teix, Berkman, all strike me as tall to very-tall players.  Makes sense that this helps, but interesting to see it in the results at a glance.

I did not include pitchers and catchers, and I didn’t look at the back end of double plays either.

My code looked for groundballs where the 1st assist was recoreded by 2b, 3b, and ss, and the 1st putout by the 1B.

The regression I got was 50% at 975 throws, though you might as well round it to 1000 since that’s about one full season.

Since before we didn’t have any data, I don’t think there was any consensus about how valuable the skill was.  For the most part it was probably ignored, as is most defensive play when salaries are considered, but every now and then you’ll read some hyperbole about a player who saves 50-100 bad throws per year.

50% for one season is pretty “good”.  How did you come up with that?  Did you do an ICC? 

I was going to initially guess 50% for 1000, but when looking at the sample results for players with around 1000 throws, they seemed a little high (suggesting a greater regression).  Then again, your 1000 is probably equivalent to my 1200 or so, since you did not include pitcher and catcher throws and the back end of DP’s.

Here is the Google Docs address for the entire spreadsheet:

http://spreadsheets.google.com/pub?key=p4mB-r5bxU8gIS5QTxMVfDw

1. You have to figure out the error rate per throw, for each player.  I’ll presume the league average is .015, so I’ll do .015 minus your save rate.  So, Sexson is .010 and Stairs is .020.

2. You figure out how much 1 SD is for each player.  For Sexson, 1 SD = sqrt(.010 * .990 / 6004) = .00128

3. Figure a z-score for each player.  Sexson is +.005 / .00128 = 3.89

4. Take the SD of all the z-scores.  Just taking all your guys with 500 throws, I get 1 SD = 1.36

5. r = 1 - (1/1.36)^2 = .46

6. Figure out the average number of throws in your sample.  Your 88 1B averaged 2304 throws.

7. r = throws / (throws + x)
= .46 = 2304 / (2304 + x)
x = 2700

Now, if it wasn’t .015, but .010, then step 7 will give you 912.

So you are saying that the 50% regression point is 2700 throws?  Based on a league average error rate of .015?  If the league average error rate is .010, then the 50% regression point drops to only 912?  And Rally says that it is 975, suggesting a league average error rate of .010?

The league average error rate is actually .01165.  What would be the 50% regression point for that?  Sounds like Rally was right on the money with 975.

It also looks like you want to regress each player to a different mean, depending upon height.  Let me see if I can easily add height to the database in the above Google Docs link.

I changed the doc in the link above to include each player’s full name and height.

The “r” between height and regressed error rate (using 1000 throws for 50% regression) is .158.  For players less than 6’3” it is .239.

For players less than 6 feet, their average errors saved per 1000 is -1.8.  Over 6 foot, it is +.5. 

Under 6’2”, it is -.9 and over 6’2”, it is +.9.

Under 6 foot, it is -1.8, as I already said.
From 6 foot to 6’1”, it is -.8.
6’2” to 6’3”, it is +1.2.
Over 6’3”, it is +.2.

I guess it peaks at 6’2” or 6’3” and then awkwardness starts to cancel out height.  Or maybe the relationship is truly linear and there is too much noise.

I did not look at handedness. I think that Rally found that lefties were better.

Sounds cool then.  Let’s accept 1000 throws as the regression point.

I think lefty 1B must be better if for no other reason that all lefthanded 3B in high school will play 1B in MLB.

I figured the regression with Tango’s process outlined in #10.  Seems like we’re all on the same page.

If you’re a lefthanded 1B, you run to your bag, plant your left foot on 1B, and prepare to receive with your right hand.  If you’re righthanded, it gets a little awkward because you have to turn around at some point to get your right foot on the bag and your left hand in position.

Tango, I think the lefty 3B end up at 1B because you don’t want a lefty thrower at 3B in the big leagues.

vj, I think that is what Tango means.  Players that otherwise have the skills to be infielders but are left-handed either end up at first or in the OF.  So they tend to be better fielders in general.  RH first basemen are not capable of playing any other infield position (well) otherwise they would, so they tend to be bad fielders in general.  More or less, of course.

Righties below 6’2” (85) were -1.1 per 1000.  Taller righties (75) were .5.

Shorter lefties (15) were -.1 Taller lefties (34) were 1.5.

All righties were 0.0 and all lefties were 1.3.  (I don’t know what the total is plus.)

So it looks like both height and being lefty is an advantage.

And the “advantage of being lefty” could partially be the selection bias that Tango mentioned.

I think its worth clarifying that with Rally and MGL’s methodology, perhaps only 2/3 of a first baseman’s value in “catching throws” is being measured.

My queries were quick and dirty, but I assume not way off. I get 4308 “reached on error” throwing errors from retrosheet data for 2000-2007, using criteria similar to MGL’s, with a rate of about .015 throwing errors per opportunity. But I also get 688 throwing errors on pickoff attempts at first. Also I get at least 1118 advancement errors [using the criteria of single plus throwing error by pitcher,catcher,2b,ss,3b on ground balls and bunts fielded by one of those fielders]. A minority of those will be on throws to other bases, but on the other hand I did not try to count “double errors” where there is a throwing error after a fielding error on the same play. And I count 331 errors on dropped throws by 1b on balls in play where an assist was credited, and 130 dropped throw errors by 1b leading to advancement on pickoff attempts.

So besides the 4308 ROE throws used in this analysis, there were in the vicinity of 2000+ other throwing errors/catching errors relevant to the first baseman’s catching ability.

Joe, good points.  I should have included pickoff throws.  I’ll have to add that.

In my STATS database, I didn’t have singles and then errors handy so I did not include them.

I did not include “dropped throw” errors by first basemen only because I already include them in UZR (I have 3 categories in UZR, which are usually combined - one, range, two, ROE errors, and three, other errors where the batter does not reach on an ROE).

But you are right.  Given the data set, at the very least, you want to “bump up” the values that we get by maybe 40 or 50%.

Similar to arm ratings and UZR in the outfield.  I, and other researchers, only look at runner advances and “kills”.  We don’t look at holding runners to singles that might have been doubles (or doubles that might have been triples) because of the OF’ers speed or arm.  IOW, you can probably bump up or down UZR a little (actually add a component based on speed only), and the same for arms, although arms would simply be increasing the magnitude of whatever you already got.

This is one reason why I LOVE vetting these kinds of research projects on this blog!

Actually in the database, it does not identify successful throws where the runner was safe on a single even though the fielder threw to first base.  IOW, it is not scored, for example, single, 6-3.  So I did not want to include errors on those throws when there were no corresponding non-errors.  Again, the errors by first basemen on those throws, when the batter was awarded a single, is already included in the first baseman’s UZR.  If the data would indicate a “successful” throw to first on a single, then yes, we could include all of those in 1B “scoops.”

Yes, ideally, we brainstorm ideas and get the work off the ground in this blog.  And then, other researchers can do more heavy lifting, dot i’s, cross t’s, and are involved in the the rest of research and publishing that is time consuming.

It’s the “honest mess” v “tidy lie”.

That’s why I like Hardball Times.

***

Here’s Dewan’s “Bad Throw Saves” in 2005:
http://actasports.com/sow.php?id=85

Except for Pujols, they are all around +/- 5 plays.

And here’s his “team error” kind of stat:
http://actasports.com/sow.php?id=100

Dewan also privately published his Fielding Bible for 2007, where he counts “Good plays” and “misplays”, similar to what we are talking about here, but includes more.  Anyway, the range for 1B is +/- 15 plays (Pujols on top again).

The other positions (other than C) don’t have such a wide range, suggesting that the “standard” PBP stats do alot to capture their value, unlike 1B.

I think my code included singles + errors.  Hit to short, fielded by = 6, bbtype = G, 1st error =6, error type = T.  An ROE and 1B+E would both meet those filters.  I don’t think I specifically excluded hits.

I did not include pickoff attempts.  I’m not sure if I would be able to include catcher errors.  Most catcher throwing errors occur on stolen base attempts, I’m not sure if I could separate those from throws to first.

Rally, for pickoffs, in 2000-2007 they appear under retrosheet event_type 8, where a throw error is recorded by P or C and pitch_sequence ends in “1”

MGL, of course you don’t want to double count what you are measuring with UZR; my thought really was a caution to anyone doing analysis who might be tempted to apply these results in combination with other metrics, where dropped throws are not being included. And as for advancement errors, I agree you can’t precisely quantify it if opportunities are unavailable. Still, it’s part of the skill, and we may be able to guesstimate its magnitude. Your analogy with outfielders holding batters to singles instead of doubles is an excellent comparison.

Catcher throwing errors to first are coded in Event_Type 8 and have the following Event_Text “PO1(E2)1-2”.  But as far as I know catcher throws to first are not recorded anywhere so you would have the same problem MGL described, no way to count opportunities.  The amount of extra value for first baseman plays saved would not be an extra 40 or 50% because in these extra cases the first baseman is just saving advancement on the bases and not saving an out as well.  I have the runs saved for MGL’s original WOWY study to be .67 per play saved rather than the .75 he stated.

My mistake. That last line should be .80 per play saved rather than the .75 that MGL stated.

Peter,

I hadn’t noticed that, but it does appear as though catcher pickoff throws only appear in the retrosheet pitch sequence when a pickoff or error is the result. So opportunities on catcher pickoffs can’t be measured from that data source.

Thanks.

Tango, #22, what do you mean “privately published?”

#22, I swear it seems like Dewan purposely does not give out the proper information on the Acta web site so as to render his “stats” almost useless!

The runner goes to third, the outfielder attempts to nail him, but comes up short, allowing the batter to get to second. This is not ruled an error, but Baseball Info Solutions categorizes this as a defensive misplay.

Now, how is that a “misplay?” What is “coming up short?” Not nailing the runner?  How is that necessarily a mistake by the OF’er?  Maybe the OF’er should have thrown to second and maybe he shouldn’t have.  Is Dewan claiming that his video scout made the decision that the OF’er should not have thrown to third in this situation?  Was there a cutoff man?  Did the OF’er throw low enough to allow the cutoff man to cut it off if necessary?  Did the third baseman say “Let it go!” when he shouldn’t have?  Did he say “cut!” and the cutoff man did not heed?  Geez!

In the other link where he gives “bad throws saved” the data presented is even worse!  How many bad throws did each first baseman have that were NOT saved!  Maybe Albert had that many saves because he got that many more bad throws.  I don’t see how you can say anything about these numbers unless you know how many equally bad throws were NOT saved.

He does this on almost every one of these “stat of the weeks.” Or at least on some of them.  I honestly think that Dewan is a guy with lots of good ideas but has little idea what to do with them.  It really bugs me that he went through all the trouble of doing his plus/minus system and publishing it, and never went from “plays” to “runs!” Not to mention the fact that he gave no citations to some other similar PBP systems that have been around LONG before his Fielding Bible.

Dewan gave out his Fielding Bible to a select group of people, one of which was whoever participated in The Fielding Bible Awards.  I think his intent is to finally publish it again for the 2008 season in 09.  But, I don’t know.

Yes, I think the video scouts make those kinds of judgement calls, that a fielder should not have thrown to 3B if the runner gets in there without a hope of having been thrown out.  How that’s done, I don’t know.  Maybe they have a stopwatch and say “if ball gets to base at least 1 second after runner, call a missplay”.

There should always be a cutoff man in proper position and the outfielder should always hit the cutoff man.  It is always a bad outfielder play if he throws a ball that can’t be cut off and doesn’t get the out and the trailing runner advances.  It is never a bad outfielder play if the outfielder throws a ball that can be cutoff and the decision is not to cut it off.

It would also be a bad outfielder play if he throws an off line throw that must be cutoff and no out results, when an on line throw would have nailed the runner.  Although that play is much more of a judgement call by the observer than the first type of bad play.

It is never a bad outfielder play if the outfielder throws a ball that can be cutoff and the decision is not to cut it off.

Sorry, that is not true at all (I actually played and managed lots of fairly high level baseball).

If you have little or no chance to get the runner (or late in a game, when the lead runner means little or nothing), you are supposed to throw to second base (or perhaps another base, but not the lead runner’s) in order to avoid having to have the cutoff man cut the ball off, which might not happen on a bad throw or if the fielder covering the base does not say “cut.”

In addition, a good baserunner can sometimes take the extra base on a throw, even if the cutoff man cuts the ball off.

That being said, your point is almost correct.  In that if the outfielder makes a good throw to a “lead” base even if he had little or no chance of getting the runner, it is not that bad of a play and maybe not bad at all.

Generally you teach young ball players to always throw towards the lead runner, making sure that the throw is low enough for the cutoff man to cut it off if there is no play on the runner.

However, as they get experience, they learn (and you teach them) that sometimes throwing to second (or another base) is even better when you can’t get the lead runner anyway, so that there is NOT DOUBT that the batter cannot advance an extra base.  Again, that is ALWAYS true late in a game when those lead runners scoring mean little or nothing.

I never said that the outfielder should always throw to the lead runners base although I can see how you might have inferred that from the language I used.  I know all of the points that you made, having played high level outfield myself.  Although in the systems that I played in the decision of which base to throw to was supposed to be made by the cutoff man and the infielder covering the lead base and signaled to the outfield, the decision was not supposed to be made by the outfielder fielding the ball.

Like a base coach and a baserunner, other than rounding third to home, the cutoff man and the fielder covering the bases can “help” the outfielder by yelling what base to throw to, but in professional ball at least, the ultimate responsibility is 100% in the hands of the outfielder with the ball.

Not to belabor a point, but…

It is true that you did not say that the outfielder should always throw to the lead runner’s base, but you did say:

It is never a bad outfielder play if the outfielder throws a ball that can be cutoff and the decision is not to cut it off.

How can those two statements be reconciled?

If the outfielder should sometimes throw to a base other than the lead base (which has to follow from your statement that the outfielder does not ALWAYS have to throw to the lead base), what if he does not?  It is not a “bad” play, but it is not the best one?  I guess that reconciles the two statements.

If an outfielder makes a play which is not the very best one, you have to dock him points in any kind of “accounting system” or at least award him fewer points than if he makes the “correct” (optimum) play.

So you concede that the optimum play is sometimes to throw to another base other than the lead base.

Therefore if he does throw to the lead base (whether the ball is or can be cut off or not), what do you call that (if not a “bad” play)?  A “good” one but not the “best?” I am not being facetious, I am just curious what your definition of “bad” was.  My only definition was that it is not the “best.”

Certainly the order of “optimum to less than optimum plays” if a runner is clearly not going to be thrown out is:

1) throw to the base in front of the batter-runner to make sure he does not advance.

2) throw to the lead runner’s base, but make sure you throw it low enough for the cutoff man to cut it off.

3) throw to the lead base, but throw it too high, allowing the batter-runner to potentially advance a base.

4) make a horrible throw such that all runners advance at least one extra base.

5) put the ball in your pocket and let everyone score.

I call all but the first one “bad.” You can call any of them anything you want, except you can’t call the first one bad and you can’t call the last three good if you want to do justice to the English language.