THE BOOK--Playing The Percentages In Baseball

Interesting data, Tango.  Are there enough players who log time at both 3B and 2B in this time frame to compare their performances at the two positions?  Are players who are average at 3B also average when playing 2B?

Using a “matched pair” (meaning the same guy in both pools, with the same weight), there are 4214 balls in zone, with the total being +17 plays at 2B, and -61 plays at 3B.

There’s about 350-400 balls in zone at the two positions per 162 GP, meaning the above represents about 11 full teams (out of 90 in the data).

So, the +17 is a bit over +1 and the -61 is around -6.

That is, the exact same guys playing both positions (2B, 3B) were compared to a higher standard at 3B.  They look like they are MLB average 2B, and much below average 3B.

Everything points to the average fielding 3B being a better fielder than the average fielding 2B:
- replacement level
- players playing at both positions
- your eyes

The onus is on others to refute this.

There’s some great posts made by Joe Arthur here:
http://www.baseballthinkfactory.org/files/newsstand/discussion/tht_smith/

Chris’ objections at the end of that thread are a bit odd, like saying:

Um, how do you think they found the bugs to begin with? By checking. I’m not terribly surprised that in the first couple of years of the work there were some problems, but since 1997, they haven’t noticed any, and two over 18 years seems pretty good quality control. Perfect? Who ever thinks it is perfect? Programmers are notoriously poor at perfection - if they did it right the first time, they’d be out of a job. Are there ever programs written without bugs?

This was in response to Joe stating:

So the double counting for two years years added up to an additional 65 plays made and 66 opportunities. Obviously the plays made were glaringly inconsistent with actual putouts.

If what Joe is reporting is accurate, than the quality control was extremely poor.  This is not a “couple of problems”, but a sheer failure.  This is a problem that simply wouldn’t happen in a real production environment, where the development team is small and fully develop the code.  (Which is the case here, since it’s a one-man job.)

Since 2B gets more opps per game, shouldn’t enough of those 3B move to 2B until the average 2B is better than the average 3B?  That should be the case with all positions, no, other than the fact that some positions may require somewhat unique skills?  This is also not-withstanding the fact that a 3B must have a strong arm and a 2B may not.  In fact, any time a position requires a unique attribute that another position does not require, such as RF as opposed to LF (a good arm), that means that the pool of players is smaller and the average fielding quality should be lower.  So that alone should mean that 3B are worse fielders (in a perfect world) and RF are worse than LF (which they are).

Agreed—bery strong evidence.  I can think of two possible offsetting factors, though I’m not sure how much weight to give either:

1) It’s possible that some or even many of the slick-fielding 3B don’t have the agility and speed to handle the DP pivot well.  You’d have to include DPs to fully measure defensive value.

2) Perhaps most of the bench IFs are mainly middle infielders by training, and so don’t perform well in their occasional work at 3B.  However, if these players were asked to play 3B regularly, they could in fact perform at league average.  This would account for both sets of results you report.

But assuming you’re right, it raises two related questions:

1) Which teams with good-fielding 3B would be better off playing their 3B at 2B?  This would better leverage the player’s defensive skills (more plays), and a replacement level 3B will be a better hitter than a repl level 2B. 

2) Which current bench 3B do we think are better players (O and D combined) than some current starting 2B?

"and RF are worse than LF (which they are).”

MGL:  That seems to contradict Tango’s data above, which suggest RFs are slightly better than LFs.  Do you have data showing the reverse?

Actually, my response was not to that specific sentence, but to the suggestion that Dewan doesn’t know what his system is measuring - that he didn’t (or couldn’t) check.

I also noted specifically that when the system first rolled out, there were problems.  Note that happened in 1990-91, over 16 years ago without a similar event.  Isn’t programming QC in general much better in 2007 than 1990 in the industry?

As for the 3B/2B issue, I think the issue could partly be the defensive spectrum.  You have 2B playing the unfamiliar 3B, but fewer 3B playing at 2B.  Placido Polanco and one of the Bell’s is all I can think of.  You say 4200 BIZ, but what’s that split?  3000/1200?  2100/2100?  How many seasons at 2B and how many at 3B?

MGL: historically (since WWII anyway), it may very well be the case that the average fielding 3B is a worse fielder than the average fielding 2B (we all remember Butch Hobson and Bobby Bonilla and Pedro Guerrero).  And, if you have a fluid system, things get into equilibrium rather quickly.  But, we are talking about people, and we are talking about 30 teams, so things don’t shake out so evenly so quickly.  Today, right now, the average fielding 3B is lights out compared to the average fielding 2B. 

Eventually, teams will catch on.  When they need to replace their 2B, they will look around and notice: “hey, there’s lots of good hitting infielders out there, but their gloves are only so-so, and I can’t hide him at 2B… let me move my Nick Punto back to 2B, so that I can bring in a bat at 3B”.

Until that happens, treat Zimmerman, Crede, Inge, Rolen, Chavez, Beltre, et al with the respect they deserve: don’t compare them to a higher baseline.  Just as you wouldn’t compare the high school SS to the average HS SS, and the high school 2B to the average HS 2B.

No, actually my data agree with Tango.  When a player plays both LF and RF, his UZR in RF is 3.9 runs worse in RF, suggesting that the pool of players in RF is better defensively.  I don’t know why that is the case, other than the fact that managers may tend to put good hitters, bad fielders in LF (although I don’t know if there is a good rationale for this, as they get around the same number of opps).  Or perhaps players with better arms are better overall athletes.  I don’t know.

Let’s see, in 06, the average offensive lwts for LF was 5.4 and RF, 2.0.  In 05, they were the same.  In 04, 6.6 in LF and 3.8 in RF.  In 03, 10.0 and 7.4, and in 02, 7.5 and 8.8.

From 02 to 06, LF was 1.5 runs better per 150 than RF, in offensive lwts.

I posted this on the above thread BTF:

According to the STATS data, which I have (the raw data), in 06 Cabrera fielded and turned into at least one out 320 gb’s, 27 line drives, and 47 pop-ups. So it appears that STATS in counting GB’s and LD’s.

In post #2, I said the multiple positions accounts for 11 teams out of 90 (so around 12%).  The “matching” was based on taking the minimum BIZ of the two positions.

***

If I take the top 40 players by BIZ at each position (over the three year total) and flag them as regulars and everyone else as bench, this is what we get:
- 16 guys are regular 2B who also had time at 3B: they are +3 at 2B and +2 at 3B (forcing the same number of BIZ)

- 10 guys are regular 3B who also had time at 2B: they are -10 at 3B, and -7 at 2B

These two suggest that teams are allowing their better fielders at 2B to play 3B, and moving their worse fielders at 3B to play 2B.  Hard to believe, but it is limited data points.  (This is what would normally happen pre-1930).

Most of the players did not qualify as being a regular at either position, which makes them the utility guy.  Those guys are +5 at 2B and -7 at 3B.

Everything is consistent with the supposition that your better fielders are not only at 3B, but that teams are actively moving in that direction.

Okay, Tango, but how many plays are going each way.  I suspect that there are a lot more 2b => 3B than the other way.

I posted this on the BTF thread:

I’ve never been abole to find a roster spot, in all my years of running computer leagues, a spot for a backup 3B who was primarily there for his defense at third.

The pool probably consists of two types of players: Backup middle infielders who can also play 3rd, and tweeners, the 1b/3b types who have a good enough bat for 3b but not quite enough for 1b, while having a 1B glove.

The first type are probably better fielders at other positions than the good third baseman. For example, I’d rather have Maicer Izturis at SS than Joe Crede, and certainly Chone Figgins in center over Crede, but Crede is a better defensive 3B than either. He has more of the specialized skills (arm, quick reactions, soft hands) needed at third but not the foot speed needed at other spots. You don’t need speed to be a great 3B. How fast did Brooks Robinson run?

I think another contributing factor is how important quick reactions are. They are probably more important at 3rd base than any other position. (I know, I’ve played there and I don’t have ‘em). The catch is you won’t often find these extreme good reactions in the backup pool. If you have the reactions to be one of the top 15-20 defensive 3B in baseball, you most likely have some hitting talent too, and you’ll be a starter.

Chris, among the regulars, I said that 16 2B go to 3B, and 10 3B go to 2B.  Are you asking for something else? Are you asking me to classify the bench players are bench2B and bench3B?

Tango,
how many plays did the 16 2B have at 3B and how many plays did the 10 3B have at 2B?  Are we getting a decent sample at both positions?

Are we extrapolating 30 plays too much?

You definitely don’t have the sample among those guys, since they are regulars at the other position to begin with.  The 2B to 3B has 643 BIZ, and the 3B to 2B has 876.  So, roughly about two full seasons.

But, the more important one is the utility guys, those that aren’t necessarily predisposed to be better at one position or the other (or rather, we hope that there’s an equal number of guys predisposed to be good at each position), and for those guys, we’ve got 2,695 BIZ.  And, to top it off, those are the guys that show the *largest* disparity in their plays above average.

So, the overall numbers that are first quoted are actually more muted than you would think.

So the 10 3B average 87 plays total over three years - 29/yr?  And 16 2B average *14* plays per season?

I think that data is too small to draw any conclusion from.  And could be moreso if there is a single player in either collection that played a lot (Polanco and Bell that I can think of)

That’s why I’m saying to stick to the utility guys, or to the whole group. 

That group has 4214 BIZ, with half the sample made up of these guys:
Role4 Role5 Last minBIZ
0 0 Punto 203
0 1 Wigginton 168
0 0 Aurilia 160
0 1 Figgins 154
0 0 Blum 147
0 0 Freel 139
1 0 Cantu 130
0 1 Sanchez 129
0 0 Hall 124
0 1 Cuddyer 113
1 0 Graffanino 105
0 1 DeRosa 105
0 0 Relaford 98
0 0 Ginter 97
0 1 Nunez 90
1 0 Bellhorn 86
0 0 Rodriguez 85

Role4: 1 means primary 2B
Role5: 1 means primary 3B
If both are 0, then they are utility players.
minBIZ is the lower BIZ at 2B and 3B.

And I am not even suggesting that this is a good enough sample either, on it own.  But couple that with what we see with our eyes, that we are seeing some fantastic fielding 3B at a time where Mark Grudz is considered a great fielding 2B.

If I put those guys I just listed in one group (guys who’ve had the most time at both positions), they come in at +36 at 2B, -23 at 3B, on 2133 BIZ.  That’s a difference of 59 plays per 2133 BIZ.  Divide by 6, and you get a 10 play gap per 162 GP.

If I take the rest of the guys, they are -19 at 2B, and -37 at 3B, or an 18 play gap per 2081 BIZ.  That’s a 3 play gap per 162 GP.

Both show that the average fielding 3B is better to some degree.

At the absolute minimum, I’d say that the average fielding 2B and 3B are equals, and with the average 3B being such a better hitter, it’s no contest.

Of that group of utility guys, I think Wigginton is the only one that came up as a 3rd baseman.  Most of the rest were originally shortstops.

I generally agree with your conclusion that the current 3B are better fielders and better hitters than the 2B. 

I don’t think we can take it any further than that, I don’t think you could take a typical good fielding 3B, say Joe Crede, and expect him to be as good a fielder at second, because of position-specific skills.

Rally, your last paragraph, isn’t that completely the opposite of what Tom is saying?  He is saying that the average third baseman is better than the average second baseman, which means, by definition, that if you move an average 3B to 2B, he will perform better (on the average)…

He’s talking about specific “career 3b” guys, because of the pivot play at 2B.  I generally think it’s overrated, but I can concede the point a bit.  I’m sure Beltre, Chavez, et al would be outstanding anywhere they play.

I think Rally is also referring to the slightly different skill sets required:  range/speed more important at 2B, reflexes and arm at 3B. It would be interesting to look at something like speed scores for the two positions—I’d guess the 2B are faster on average.

The fact that util players are better at 2B (vs. avg.) doesn’t completely settle the issue.  It’s probably true that most Util IF were developed at 2B or SS, have more experience at those positions, and have more range than a lot of 3B.  That wouldn’t necessarily translate into great success when playing 3B—in fact, if they play there only occasionally, it might be a tough adjustment.  So it’s still possible that some 3B would not in fact be as good at 2B as they are at 3B.

You can get the simple averages from here, for all traits:
http://www.tangotiger.net/scouting/pos2006_2B.html
http://www.tangotiger.net/scouting/pos2006_3B.html

As long as we are all in agreement that using the negative of offensive LWTS to determine the average defensive LWTS of the two positions is not a good idea, I’m happy.

If we only had 8 teams in the league, then I can definitely see how players would be much more likely to jockey around for a spot, and each position would be in balance.  If we had 80 MLB teams, then it would look more like the minor leagues, where there isn’t as much reason to move around.  If we had 800 MLB teams, then it’ll look more like college.  If we had 8000 MLB teams, all your best players are at SS and 3B.  There’s no reason to believe that 30 MLB teams makes equilibrium, and Patiot’s historical LWTS graphs certainly shows that there’s ebb and flows throughout the years (see late 40s, where off for CF and 1B was equal).

I’ll agree to that Tango.

Baseball teams sure agree to it - just look at how little they pay to free agent 2B compared to other positions.

I would agree that we can’t use average offensive production for this purpose.  But using replacement level offense at each position (such as Silver’s FAT) to assess the positional defensive value remains a very strong—perhaps the best—approach.  In general, that differential tells us what price (in lost offense) teams are actually paying to acquire average defense at a given position. 

Now, at any given time there may be an imbalance or inefficiency in the game.  Perhaps teams are accepting slightly lower replacement offense at 2B than they need to.  If you feel you have strong evidence for that conclusion, make the adjustment.  But in general, it’s reasonable to assume that players are distributed efficiently. Moreover, whatever the limits of the offensive approach it’s certainly not clear that analyzing position switchers gives you a more accurate estimate, when considering sample size and bias and the limits of defensive data.  Especially when looking at pre-PBP periods, the offensive metrics could very well provide our best estimate.

As long as we don’t use single-year offensive performance levels (be it the average or replacement for that position), I can accept to a degree.  There’s just no justification to force avg2B=avg3B, overall, for every single year, using that year’s performance data.

MGL listed his top 10 AL players, which included Ortiz and Hafner, but excluded Vlad and Manny. 

It would seem to me that the position+fielding adjustment he applied to the DH was much nicer than what he did to the corner OF.

As a rule, I do the following:
good-fielding 1B: 0 wins
average-fielding 1B: -1 wins
poor-fielding 1B: -2 wins
DH: -2 wins

This way, Ortiz and Hafner, whether they play at DH or at 1B, gets the same -2 wins adjustment.

MGL, can I assume that you took the approach that I rail against all the time?