THE BOOK--Playing The Percentages In Baseball

Having watched about 95% of the games Betancourt has played in since he arrived in the major leagues, I’d say without wavering that there’s no lack of hustle effect going on with Betancourt.  I can’t remember one time ever thinking to myself that he was giving less than 100%.  He’s never been called out by his coaches for anything of the sort, and this is a coaching staff that is legendary for publically criticizing their young players for a lack of effort. 

My best guess is that it’s just a distribution thing that will even out over time.

Thanks David, that’s great to know.  That removes one possible reason.

It would be interesting (and exciting if true) to learn if he’s being positioned suboptimally.  There is always a tendency for a great fielder like Betancourt and Ichiro to be positioned in such a way to overcompensate one way, because their coaches think they can get away with it.  When you have a star at 3B that Beltre obviously is, maybe there is a tendency to move Betancourt a couple of steps closer to 2B.  Maybe when Ichiro had Cameron, they cheated Ichiro closer to the line?

We really need to get Lojacks on these ballplayers.  $5,000 outfits an entire team.

I couldn’t disagree more on your assessment Tango.  Again.

First of all, how can you say that Reyes’ numbers are stable?  Is there something wrong with your eyes?

In 6 out of 7 categories, the fans rated him higher in 06 than in 04.  What are the chances that that occurred by chance (and that they are really rating him the same in 04 as in 06)?  He rose an average of 3.3 points per category from 04 to 06.  What are the chances that that occurred by chance?  I really don’t care (and neither should you) what 05 looks like.  If you want to see whether something improved or not you look at end minus start.  Period.  You must be looking at someone else’s numbers.

I suppose we can argue about the magnitude of the improvement according to the fans, and we can probably argue about what “stable” means (I guess you think that means “a little improvement” although I am not sure that 23 overall points of improvement from 04 to 06 is “a little").  But we CANNOT argue about whether the fans’ rating is higher (and more than a scintilla higher) in 06 than 04.  Nuf said about that, although I have to wait until my head stops swimming.

As far as Betancourt, you are ignoring one very crticial thing.  And that is the very real possbility that there is something about a player (like Betancourt and say, Jeter) that leads the fans to overestimate his total value such that his real value as measured by the metrics is right and that measured by the fans, even sophisticated ones using sophisticated methods, is wrong.  I think that it is not only a real possibility, but one that definitely exists for a certain percentage of players.  I think that there is a small percentage of players whom could be “scouted” by the fans from today until the end of time, and they (the fans) will get it wrong, in both directions.  There are simply some observational biases that are imposible to get rid of, which is one reason why we like to use objective, data-based metrics in all facets of life, especially sports.

And the idea that the fans can capture “potential” (that has not shown up yet) I find really implausible and I will leave it at that, lest someone think that I really hate you (which I don’t of course).  I mean to a large and meaningful degree.  I’m sure there are some players who to some extent might have some raw skills that correlate well with defense and should continue but have some flaws that might be corrected.  But I DON’T think this exists to any significant degree.

IOW, if anyone wants to bet “over” +15 in UZR per 150 next year, I’ll gladly take the under for any amount of money.  I might even go lower than that (+15).  The way you talk about him, you think that +15 is a slam dunk (probability-wise of course) I would think.  Unless you think there is some other problem like positioning, etc. that will continue to suppress his UZR next year.  I would put his over/under next year in UZR at around +8.

Im not sure if this makes a difference but in terms of reyes you also had to factor in matsui comming here and reyes moving from short to second base for a period of time in 2004.  For some reason, I think he was alot better at short then he was at second.

Anyway I always thought he was an above-average fielder since he came to the majors, but I can also say he’s getting better as time passes as well.

An average change of 3 points over a span of two seasons is extremely stable.  Don’t forget, we’re taking a sample of say 30 fans in one year, and 30 fans in another year.  They wouldn’t necessarily be the same fans (assume say a 30% repeat of voters from year x to year x+2).  I guess that’s another thing on my todo list is to show what “stable” is.  That is, what the average change for a player year-to-year, and year-to-year+2.

I said in another thread when discussing Betancourt and how much to trust the Fans (and for how many years) is if I would bet on it.  After all, if we’re not that convinced enough to put our money where our mouth is, then we’re full of it. 

That said, your UZR forecast for Betancourt, outside of scouting data, would probably be in the “-5 to -10” area, probably one of the worst in the league ahead of Berroa and Jeter presumably.  Your “+8” offer on Betancourt is interesting.  It acknowledges a *huge* regression toward the Fan.  That is, if I figure the Fan number represents a “+20”, and UZR represent -5, then a +8 estimate is a 50% regression toward the Fan mean. 

Dispassionately, that’s about right, and that should be our estimate.  And, if Betancourt gets another horrible UZR/ZR year, then that 50% will come down to 33% or whatever, eventually to the point that the regression toward the Fan will be like 5% or 10%.

Betancourt can be this generation’s Robbie Alomar and Griffey Jr, guys who look so incredibly impressive, but the numbers simply don’t back that up.  We’ll see.

One other thought, which is partly linked to what MGL is saying. Is there any evidence that the Fans Scouting report may have a bias against young “star” players who keep on improving, who fans consistently over rate. Also fans may over rate aging verterans who lose a bit of sharpness—eg, Andruw Jones / Torii Hunter unless there is a clear drop in performance ...

MGL sez:

“There are simply some observational biases that are impossible to get rid of.”

To which I reply:

Name them.

One consistent observational bias that I see is the overrating of the spectacular play, and the underrating of the mundane.  For instance, when evaluating second baseman, fans (and Gold Glove voters, it seems) consistently favour the ability to make the full out diving stop and pop-up throw over the ability to turn the double play.  If Tango ran the fan’s survey for second basemen over the last 20 years, I’m pretty sure that Alomar would do exceptionally well.

A blend of the objective and subjective seems to me to be the way to go.  There seems to be little doubt that Betancourt has the tools to be a great shortstop, but whether he will convert the tools to fully formed skills is the issue.  Dave Concepcion might be an example of a shortstop who successfully accomplished that (perhaps not coincidentally immediately after Joe Morgan arrived in Cincinnati). 

When a number of different metrics indicate that a player has been below average at the position over 210 games, it is likely that the metrics are capturing something.  The real issue, it seems to me, is not where Betancourt is, but where he is going.  Having the tools is definitely an important asset, but it is not the only determinant.

There is no question that fans overrate the aging ballplayer who was once good.  I noted that with the 2003 Report, and has been the case since.  The realization does finally come, as it did for Junior this year.  UZR has been down on Junior for a long time.  Here is Junior, by the Fans:

2003 74 64 65 72 67 54 53 64.1
2004 80 56 52 77 73 58 64 65.7
2005 85 55 54 88 75 70 79 72.3
2006 64 16 20 64 61 46 52 46.1

That last column is the simple average of the other 7.  (I put the decimal, not because I believe in the precision, but just so that it stands out from the rest of the numbers.  As well, we don’t want the simple average, but it’ll do for our purposes here.)

The average CF is around a 60 or so.  Clearly, the Fans saw Junior as a somewhat above-average CF from 2003-05, and then, they finally came to an acceptance that he stinks.

Yankee fans had no such issue with Bernie:
2003 25 34 62 55 34 0 22 33.1
2004 26 27 48 46 31 7 30 30.7
2005 10 11 44 32 22 0 16 19.3
2006 23 21 37 46 27 0 12 23.7

They think he’s been horrible for a long time now, as does UZR.

I think one observational bias is that Fans aren’t good at positioning.  For example, when I looked at 3B in 2003 vis-a-vis UZR:
http://www.tangotiger.net/scouting/scout_Results3B.html

The Fans really nailed this one. Of the top 8 fielders in UZR, the Fans were highly favorable to 7 of them. The only player they didn’t appreciate as much as UZR was David Bell. In the bottom of the pile, the Fans and UZR did not like the same two players: Aramis Ramirez, and Hinske.

3B, compared to SS/2B, don’t have much leeway in their positioning.  Fans therefore will pretty much see the 3B from the same spot all the time, making the same kinds of plays all the time.  More or less.  When it comes to evaluating 3B, Fans probably are excellent.  Rookie Ryan Zimmerman was top-of-the-class in 2006, according to the Fans, but not PBP metrics.  I would definitely expect to see him rank way up there.  He also came with very high accolades, with comparison to Brooks Robinson of all players, and of course he went from draft to practically rookie of the year in under a year. 

And, it doesn’t take long to look at Scott Rolen and realize you are looking at a marvelous fielder.  And, with little leeway in positioning at 3B, PBP metrics agree, wholeheartedly.

***

Back to the matter at hand, I’ll be recruiting “superfans” to look in-depth at some of these players.  As luck would have it, a superfan was kind enough to volunteer from Seattle, where our biggest trouble case appears. 

Hopefully, a few others who have opinions on the matter will also volunteer.  If you are not part of the solution…

This is speculation and educated guesses, but here are some biases or at least potential biases that I think exist in fan ratings, even from educated fans who see the players a lot and are using somewhat objective methodology as that employed by Tango in the FSR.  Some of them have been mentoned/discussed already.

1) Aging veterans who used to be good but are no longer are probably overrated, especially if there is not much “talk” about how bad they have gotten.

2) This is related to #1, but clearly (IMO) the fans “objective” ratings are strongly influenced by reputation and what they read and hear.

3) I have always said that “smooth” looking fielders tend to be overrated and awkward ones underrated.

4) If the fans observe more than a player’s share of spectacular plays they will tend to overrate them.  And vice versa.

There are probably more, but that is all I can come up with in a few minutes off the top of my head.

Obviously using the “skill component” methodology that the FSR uses tends to minimize these potential biases, as opposed to simply reporting a player being good or bad, but I think they still exist and will influence the fans’ rating of the individual skill components of defense.

MGL’s last point is extremely important to note, and I concur.

The key to the Fans Scouting Report (FSR… must everything be a darned TLA??… three-letter-acronym...), anyway, the key is the component aspect.  As I’ve always said, people are great in offering specific insights, and relatively terrible at coming up with an overall assessment.  I do agree that someone’s initial overall assessment will impact his individual component assessment.  But, what can you do.  We’re people!

There’s an “overlap” problem with good fielders, too.  Last offseason, I suggested that Russ Adams wasn’t as limited in range as he’d shown in 2005.  Part of the hypothesis was that Orlando Hudson was a ballhog.  This is a great thing for a team, but not-so-great for the adjacent fielders.  Sure enough, even though his error rate went sky-high, Adams’ ZR went up considerably this season.  And, while Hill had an outstanding ZR, his RF was 0.72 behind Hudson’s the previous season (ergo, Hill was getting fewer total balls, and presumably fewer balls outside his “zone”, since his ZR was so high).

==> For reference, among full-time (1000 inning) AL shortstops, only Juan Uribe had a higher ZR than Adams (.844) in 2006.

Betancourt was the “new guy” on the left side of the infield, and also had a rangy 2b on his other side… I have to think that a lot of the “cherry picked” balls weren’t his in 2006.  That said, I did watch a lot of M’s games this year (50+), and didn’t see any obvious evidence of this (as would happen frequently with Hudson).  But then again, I didn’t see anything to suggest that Betancourt wasn’t extremely rangy, either.  He kicked or dropped a few easy balls, but otherwise he was both rangy and reliable.

Your post is a little misleading.  Russ Adams was not the regular SS for the Jays.  John McDonald was.  Adams did have a ZR of .844, but with only 271 innings (96 balls in zone).  McDonald had a ZR of .828 (244 balls in zone).

Your choice of 1000 innings and “AL” is also misleading.  Alex Gonzalez of Boston had 966 innings, with a ZR of .863.  Bobby Crosby was 828, .855.  If you are going to offer a reference baseline comparison, I certainly wouldn’t use 1000 innings as a threshhold!

Looking at the Jays fielding:
http://sportsillustrated.cnn.com/baseball/mlb/teams/stats/2006/blue_jays/0_fielding.html

We see that McDonald had 14 errors on 244 balls in zone (BIZ).  Russ Adams was 10 (!) on 96 BIZ, which prorated to McDonald is 25.  Derek Jeter for example was 15/462.  Pro-rating Adams 10/96 to 462 is 48 errors!

I’m not a fan of low error totals meaning something, but high error totals certainly mean something.

If we look at Jays fans:
http://www.tangotiger.net/scouting/scoutResults2006_TOR.html

They see McDonald and Russ Adams as being night and day.

In 2005:
http://www.tangotiger.net/scouting/scoutResults2005_TOR.html
Jays fans thought that Adams was terrible, horrible, and in 2006, they thought he was even worse.

I doubt the explanation was ball-hogging, and certainly not between 2B/SS.

I have not found much ball hogging effect in the
IF.  In the OF, sure, but not much in the
IF.

I think perhaps the intent of my post was missed.

Fine, Russ Adams wasn’t any hot stuff with the glove this year… that wasn’t even part of the discussion, really… I was pointing out that there was a precipitous increase in his ZR this season.  I used him as an example because…

Betancourt is being considered in this thread as an average-ish fielder based pretty much only on the fact that his ZR was average-ish in 2006 (everyone who’s seen him play defense raves about him).  If one is going to discern that Betancourt is a mediocre fielder based on one season of low ZR, I think that is drawing far too much inference from one number.  I do believe that infielder “overlap” comes into play.  And I do believe that Beltre and Lopez suppressed Betancourt’s ZR slightly this season.

The poor numbers (ZR, UZR, Dewan) for Betancourt was since he came into the league last June (the equivalent of 203 full games).

The regression toward the mean equation is about G=81.  This means that after 203 games, we would regress Betancourt performance metrics by 81/(81+203)=29% toward the league average.  So, if these metrics have him at say -8 per 162, performance-wise, then this means their estimate of his true talent is -8*.71= -6 per 162.

***

As for Adams “increase” in ZR, he was 81 for 96.  One standard deviation, given 96 BIZ, is .040.  His performance of .844 is not terribly reliable, since we’re 95% sure it’s between .764 and .924, or, in essence, the entire range of the worst to best SS in the league. 

You might as well quote me Chris Shelton’s April, or Shane Spencer’s rookie September, in terms of reliability.

I don’t have anything really useful to add, but I do think that Reyes improved quite a bit from last year to this.  Now, I don’t think any of his “components” actually improved, but I saw more consistency and better anticipation.  Probably better positioning too.  The BIS stats back that up, but I thought it before seeing the stats.

Was this random chance?  Did he just happen to be in the right place at the right time, more often than he was in 2005?  Is that what you’re suggesting?  I guess that could be true, but I hope not.  I like to think anticipation and consistency are skills that can be learned.

As for ball hogging in the infield - I don’t think Hudson could have any effect on Adams.  ZR, and UZR, are based on ground balls only.  Hudson was a ballhog on pop flies, but that’s relevant only to his range factor, which we all know is outdated.

There’s just no way a second baseman will steal a significant number of plays in the shortstop zone.  If it ever happens, its due to extreme positioning, and is very rare.

Beltre’s range, though, might have an effect on Betencourt.

re: As for Adams “increase” in ZR, he was 81 for 96.  One standard deviation, given 96 BIZ, is .040.  His performance of .844 is not terribly reliable, since we’re 95% sure it’s between .764 and .924, or, in essence, the entire range of the worst to best SS in the league.

You might as well quote me Chris Shelton’s April, or Shane Spencer’s rookie September, in terms of reliability.
...
Again, you miss my point, and are replying to something tangential to my assertion.  My assertion is that Hudson suppressed Adams’ ZR in 2005, followed by the hypothesis that Betancourt’s ZR may have been suppressed so far by rangy neighbors so far in his career.

What you did was question the .844 number (and in using it again, you’re going back to an imagined statement that Adams is a rangy shortstop, and are attacking that premise, which isn’t part of my argument).  More apropos would be to question the .780 number from 2005 (which is what I attempted to do, although I wasn’t clear in it).  [Sorry, but my computer with Excel died, so I’m going to wing some things.]

So, using the SD from the Everett blog entry (.014), we’re looking at someone with a supposed mu of .780 fielding 96 balls at a .844 clip a delta of almost 5 mu’s.  No matter what assumptions you make about population and sample, that’s extremely unlikely… concluding - as I said at the start - that it’s unlikely that Adams is as “un-rangy” as he appeared to be in 2005. 

We cannot prove that it was Hudson that suppressed his ZR in 2005, but by virtue of the 2006 data, I think it’s clear that he wasn’t AS BAD as his 2005 numbers showed, barring an extreme improbability.  And I will not be surprised in the least to see an increase in YuBet’s ZR in 2007.

P.S. You were missing the point on the higher error totals, too.  The reason I cited those is that getting to balls and making errors drives DOWN a ZR, and yet Adams had a very high ZR in 2006.

96 BIZ implies 1 SD = .04.  Therefore, a .780 fielders is as likely as a .900 fielder to play .840 (1.5 SD).

Well, I’ll have to go and review how you got that, I guess.  My understanding was that you either use the SD from the population (as well as it’s known), or you take sd(sample)/sqrt(n).  Still, it’s not something I work with everyday, so I’m sure there’s something I missed.  Thanks for your patience.

Still, at 1.5 SD, it’s over 93% likely that the sample didn’t come from a population with mean of .780.  While that still falls within a normal 5% confidence interval, it’s still unlikely.

Also - commenting on another reply - I went and looked at some of David Pinto’s PMR stuff on Ball Hogs, and he did look only at balls in the air.  That said, Adams (2005) was very high on his list of “most hogged against” (3rd among all shortstops, IIRC), and my assertion about 2005 was that Hudson “hogged” some balls from Adams - not necessarily GB only.  I haven’t really worked with ZR for over a decade, when I used to both score games as well as “play” with the data, so it’s possible that they removed any of the possible “overlap” plays, but I would be surprised.

Anyway, with overlap, it may not take many balls to move someone on the scale, although it would be a seemingly subtle impact, since a player may be able to get to 400 balls out of 500, for a .800 ZR, but if others rip 10 balls from him over the course of a season, he’d be at 390/490, or .796.  And .004 is a significant portion of 1 SD for a player with 500 BIZ.

Consider what I’m suggesting for a moment, from a sanity-check perspective… is it that unreasonable to assume that there are at least 15 balls to either side of a shortstop that either infielder could field cleanly without much trouble?  If so, if they split 10-5 “against” said infielder on both sides in a given season, and then 10-5 “in favor” of said infielder the next season, that’s the 10 balls, and the .004 ZR delta.  Sure, that doesn’t account for the huge distance Betancourt is below average in ZR, but I don’t think it can be entirely dismissed, either.

Rob,

There’s almost no chance of ball-hogging in zone rating since only zones where a player is expected to make more than 50% of the plays count. Those aren’t really zones where other players can make much of a play, especially not in the infield.

That’s nice in an idealistic world, but here’s the grid, as I’m sure you know (http://www.baseballthinkfactory.org/szymborski/zrgrid.jpg).  Having worked with the raw data that was inbound from reporters for 4 seasons, I know that assuming data to be “clean” in the sense that nobody gets to balls in zones G,M,N is completely unrealistic. 

And again, I think I’m failing to communicate my point… the point isn’t really that Beltre, for example, comes over into Zone H and steals a ball from Betancourt (although I’m sure that happens reasonably often, as with most 3b - who will charge and take “whatever balls they can get” on those plays).  The 10-5 schizm is in the “tweener” zones, such as “G”.  If a medium-speed grounder bounces between Beltre and Betancourt, EITHER ONE can easily field it.  Sure, for a “hard” grounder, it’s a tough play, and maybe even impossible for either.  But for a “medium” grounder, either one can get it, and it’s a matter of personality, randomness, etc. (i.e. NOT fielding range).

From Chris Dial’s great writeup, “‘Outs Made’ equals every ball fielded within the zone that is turned into an out, plus all balls fielded outside the zone turned into an out.” So, these “tweener” balls will count as an out and a BIZ for one of the players (http://www.baseballthinkfactory.org/files/dialed_in/discussion/what_is_zone_rating/), and NOTHING for the other player… that’s why I used the 10-5 example - in one season, Player A would have +10 outs and +10 BIZ, while in the next season, Player B would have +10 outs and +10 BIZ.

Back to my hero, Russ Adams - all I am saying is that in 2005, there were likely several medium-speed grounders up the middle (zones M and N), where either he or Hudson could have gotten them, and Hudson did.

SD = sqrt(p*q/n), where p=success rate, q=1-p

So, with 500 BIZ, 1 SD = .018.  A difference of .004 is 0.25 SD.

As for the significance of 1.5 SD, and calling something that is 7% likely to happen by chance as “unlikely” is not valid.  It’s 7% likely.  It is what it is.

If you’re looking at 200 fielders year-to-year, 5% of them (10 fielders) will be outside the 2 SD range just by luck.  If you find more than 8-12 fielders, then, let’s talk. 

***

You may also be interested in this analysis of interposition correlation of UZR:
http://www.tangotiger.net/Scott/UZRcorr.html

along with the discussion here:
http://www.tangotiger.net/archives/stud0035.shtml