from Amazon
Tuesday, June 06, 2006
Fielding Position Adjustments
MGL was nice enough to send me his UZR data for 2002-2005, but under heavy penalty of silence. However, we can still try to do some cool stuff with it. We have 2163 total records over the 4-yr period, or about 77 players per position per year. His data is limited to those players with at least 10 games at any one position, 1B to RF.
How much harder is it to play CF, compared to LF? Or 2B? RF compared to 3B? LF to 1B? Let’s take a look.
Note that this is a research-in-progress. I will be updating the main article as I produce more research. Treat this as a lab notebook.
MGL was nice enough to send me his UZR data for 2002-2005, but under heavy penalty of silence. However, we can still try to do some cool stuff with it. We have 2163 total records over the 4-yr period, or about 77 players per position per year. His data is limited to those players with at least 10 games at any one position, 1B to RF.
How much harder is it to play CF, compared to LF? Or 2B? RF compared to 3B? LF to 1B? Let’s take a look.
Our first step is to follow a basic approach, and see how players who played both positions did. The most obvious one is the LF/RF comparison. There were 112 such players that played both positions over the 4-yr period. Taking the minimum number of games played at LF and RF, and we get 4959 games. The weighted UZR as a LF for these players (weighted by the lesser of the GP at LF, RF) is -0.7 runs per 162 GP. As a RF, it’s +1.2 runs per 162 GP.
So, these guys, were a bit below the overall LF average, and a bit above the overall RF average. Put two and two together, and this means that the overall LF average is about 2 runs better than the RF average. Our gang of dual-position players, remember, is smack in the middle of these two groups. It looks like this:
average LF is 1.9 runs better than the average RF
Capish? This equation is consistent with our current findings.
What’s the problem? Well, first off, the guys who play both positions are not necessarily a representative sample of the LF and RF population. They probably have certain traits that make them candidates, to the managers’ eyes, to play either position. We also have an experience issue, where a guy who has tons of experience in RF, might not play too well for 10 or 20 games in LF. And we have an aging issue to consider, especially if the guys in our sample went from being a full-time from one position to full-time at the other position.
We’ll get to that in the future. For now, let’s keep going.
Let’s compare the guys who played both LF and CF. That’s 83 guys, with 3447 games. As a LF, these dual guys were +3.3. As a CF, they were -6.0. Makes perfect sense. First off, guys who play both positions are probably better than the average LF, and that’s what we get. They are +3 runs better than the average LF. And, as we expect, since some of these dual guys are LF, the dual guys would be worse than the average CF, and that’s what we get too. So, we get this equation:
average CF is 9.3 runs better than the average LF
Finally, let’s look at the CF/RF guys. We have 78 of those, with 2649 games. As a RF, these dual guys are -0.8. As a CF, these guys are -9.3. This looks like this:
average CF is 8.5 runs better than the average RF
Put it all together, and we have:
LF = RF + 1.9 (n=4959)
CF = LF + 9.3 (n=3447)
CF = RF + 8.5 (n=2649)
From the CF perspective, the average RF is about 0.8 runs better than the average LF. But, when we look at the LF/RF players, the average RF is about 1.9 runs *worse* than the average LF.
Because the LF/RF sample is much larger, we give more weight to that equation. Let’s make it that the average LF is about 1 run better than the average RF.
So, we get these equations:
LF = RF + 1.0
CF = LF + 8.4
CF = RF + 9.4
Now, all three equations are in-synch.
There’s actually another aspect that we need to correct for: CF get about 30% more plays, and so “per 162 GP” is not the correct denominator. “per 600 balls in play” would make more sense. We’ll fix that up next time.
More to come…
But now, we have to adjust it to “balls in play” instead of “games”. There’s about 3 BIP to the LF, RF, and 4 to the CF. So, let’s adjust everything to 600 BIP.
Restart with LF/RF: We had -0.7 runs per 162 GP as a LF. This becomes -0.9 runs per 600 BIP. As a RF, it’s +1.5 runs per 162 GP. New equation:
LF = RF + 2.4
Restart with LF/CF: As a LF, these dual guys were +3.3 per 162 GP, which is +4.1 per 600 BIP. As a CF, they were -6.0 per 162 GP, or -5.6 per 600 BIP. New equation:
CF = LF + 9.7
Restart with CF/RF: As a RF, these dual guys are now -0.4 per 600 BIP. As a CF, these guys are -9.3 per 600 BIP. New equation:
CF = RF + 8.9
Now we get into the situations where from the CF perspective, the average RF is about +0.8 runs better than the average LF. But the LF/RF comp is +2.4. That makes it about a 1 run advantage for the LF. New equations:
LF = RF + 1.0
CF = LF + 8.8
CF = RF + 9.8
Let’s make the average OF a zero. So, our baseline is:
+6.2 CF
-2.6 LF
-3.6 RF
Next up, the infield…
Repeating our process for the infield, let’s start with the 2B/SS. We’ve got 60 players here. Our dual players are +1.3 per 600 BIP as a 2B, and -3.6 as a SS. Makes perfect sense again. That gives us:
SS = 2B + 4.9
The 3B/SS has 52 players, for +1.0 as a 3B and -0.2 as a SS. Interesting that there is not much of a change. This may be due to the selective sampling issue we discussed. You’re not going to put that slow-footed 3B as SS, but that 3B with good speed can hold his own as SS.
SS = 3B + 1.2
There are 64 2B/3B pairs. This one is the most interesting because of the different skillsets in play at those positions. Of the guys that played at both positions, they were +0.0 as 2B and +2.5 as 3B.
2B = 3B + 2.5
The interesting thing here is that from the persective of the 3B, the 2B baseline is much tougher than the SS baseline. Tweaking here and there, and we get:
SS = 2B + 1.8
SS = 3B + 4.8
2B = 3B + 3.0
Let’s make the average IF zero, giving us a baseline of:
+2.2 SS
+0.4 2B
-2.6 3B
So, far, none of these numbers are really surprising. CF is much better than LF or RF, which themselves are close. SS is a bit better than 2B which is a bit better than 3B.
Now, for the fun part. Let’s convert our positional UZR for CF, LF, RF, into just “OF”, using the above adjustments. Same for the three IF positions.
We get 45 players who played some infield position (2B, SS, 3B) and some OF position (LF, CF, RF). This group, was +0.6 in the infield, and -7.6 in the OF. Therefore, it’s much easier to play the IF than the OF.
We must remember one of the other sampling issues we had, which is who is selected to play dual positions. The likelihood is that it’s IF going to the OF, and not the other way around. Therefore, an experience factor also needs to be considered. That is, perhaps the reason IF don’t do well in the OF is that they need time to get the experience. The opposite is also true, except perhaps we don’t see many players make that switch to begin with. So, our sample may simply be overwhelmed by the experienced IF going to OF, rather than the experienced OF going to IF.
Since we’ve got alot of biases to overcome, we need to go back and try to fix all this. Which we’ll do next time…
***
Updated: June 7 - Using more data
MGL was generous enough to give me all players for the 02-05 time period, as opposed to the minimum 10 game requirement. Without boring you with extensive commentary, let me just present you with the updated data, and a few blurbs:
LF/RF: as LF -1.7, as RF -0.2, gap of 1.5 runs, 18,939 BIP
LF/CF: as LF +3.5, as CF -5.7, gap of 9.2 runs, 14,470 BIP
CF/RF: as CF -9.1, as RF +0.1, gap of 9.2 runs, 11,799 BIP
The CF data gives you the same gap of 9.2 runs. Let’s call the new equations as:
LF = RF + 1.0
CF = LF + 8.7
CF = RF + 9.7
Which gives us a baseline of:
+6.4 CF
-2.2 LF
-3.2 RF
All in all, not much change, but we are a bit more certain with the adjustments.
2B/SS: as 2B +0.4, as SS -3.7, gap of 4.2 runs, 17,715 BIP
SS/3B: as SS -1.0, as 3B +0.3, gap of 1.3 runs, 10,503 BIP
2B/3B: as 2B -1.3, as 3B +0.6, gap of 1.9 runs, 13,037 BIP
Giving us a baseline of:
+1.8 SS
+0.0 2B
-1.8 3B
The 3B comes out a bit better.
Putting it into IF/OF, and we have:
IF/OF: as IF +0.1, as OF -6.7, gap of 6.7 runs, 11,996 BIP
We’ve got a pretty substantial sample size here. Let’s assume that the outs per BIP is .700 at every position. This gives us 1 SD from the mean of sqrt(.7*.3/11996) = .0042 outs, or .0033 runs per BIP. Which is 2 runs per 600 BIP. Our gap here is 6.7 runs, which is more than 3 SD from the mean. So, something fairly significant is happening here. Whether it’s something “real” in that it’s tougher to play OF, or whether the sampling bias is so large, we don’t know yet.
***
Updated: June 7 - Compare 1B to everyone
It should be fairly obvious that the average 1B is worse than the average fielder at any other position. Hopefully, the UZR data will bear that out. The real question is how much worse. I don’t know the answer yet, so let’s see.
pos n BIP gap
4 57 1651 (0.6)
5 102 6421 6.7
6 37 1243 3.9
7 105 6696 5.5
8 26 1972 1.7
9 90 4659 10.8
The first column is the other position the 1B plays. The second column is the number of players that played multiple positions. The third column is the lesser of the BIP for the two positions. You can see that there’s not many 1B/SS duality, but there’s plenty of 1B/LF and 1B/3B. The last column is the one of interest, and it’s the gap between 1B and the other positions.
Let’s focus on the OF. We see that the average 1B is 6 runs worse than the average LF, but is 11 runs worse than the average RF. This gives us an average of the 1B being 8 runs worse than the average LF/RF. Since the average LF/RF is 9 runs worse than the average CF, we’ll infer that the average 1B is 17 runs worse than the average CF. Weirdly, the average 1B is only 2 runs worse than the average CF. However, 40% of the sample is Darin Erstad, which is hardly representative. If we take him out, which is hardly correct to do, the gap shoots to 15.3 runs.
We get into the same sampling issue with the infield. The average 1B is 7 runs worse than the average 3B. But, he’s not so bad when compared to SS, and he’s *better* when compared to the average 2B. Given what we know, and given the sample size/sampling issues, this is another thing that needs to be accounted for.
If we focus on the more reliable pairs (1B/3B, 1B/LF, 1B/RF), we see that they are all pretty even, around a 7 run gap. From the perspective of the 1B, the average 3B, LF, and RF are all pretty even.
If we recall these two baselines, where we set the average IF and average OF position each to zero:
+6.4 CF
-2.2 LF
-3.2 RF
+1.8 SS
+0.0 2B
-1.8 3B
Let’s add 0.5 to each of the OF, and subtract 0.5 from each of the IF, and we get:
+6.9 CF
+1.3 SS
-0.5 2B
-1.7 LF
-2.3 3B
-2.7 RF
Putting in our 1B as about 7.7 runs from the LF/3B/RF gang, by adding 1.3 runs to each of the above 6 positions we get the following:
+8.2 CF
+2.6 SS
+0.8 2B
-0.4 LF
-1.0 3B
-1.4 RF
-8.7 1B
This becomes our first-pass fielding spectrum.
I’m not too happy with this. The internal CF and LF/RF comps does work, and seems fine. The internal SS/2B/3B seem fine as well. It’s the IF/OF comps that just doesn’t seem right. However, when we used the 1B as our centering point, we see that the 1B sees the difficulty of the 3B to be inline with that of the LF/RF.
The problem is likely that the skillset of the 1B is so radically different from the others, and that the selection bias is so strong, that the 1B perspective doesn’t really tell us much.
I’m sure as we keep going through these excercises, and try to adjust things accordingly, that the CF will remain at the top, or perhaps just a shade below the SS. What is also interesting is that hitting-wise, the CF is around a league average hitter. If we combine these two, that the average CF is an average hitter and an above -average fielder, then the average CF is an above-average player. Therefore, when you looking at positional-adjustments, the CF will be too heavily penalized.
***
Updated: June 7 - What about experience?
I treated the 02-05 data as the universe, and looked at how many games they played at each position over that time period. If they played at least 3 times more games at one position than the other in the pair, I considered that position as the “primary” position. I do have the issue of a game with say 160 games at 2B, 40 games at SS, and 10 games at 3B. There’s actually three different pairs here for this player, and in each pair, he will have a primary position, including the SS/3B pair. Not ideal, perhaps. But, let’s go with it anyway.
First up is the LF/RF pair. Of the guys who are primarily LF, they were -2.0 in LF, and -5.9 in RF, for a gap of 3.9 runs (on 2019 BIP). So, it looks like the guys who are asked to shift to RF are those guys who are less than average LF, and they perform even worse in RF.
Of the guys who are primarily RF, they were -0.2 in RF, and -13.9 in LF, for a gap of 13.7 runs (on 2295 BIP)! In both cases, the players were better in their primary position, telling us that experience does matter. So, don’t go around willy-nilly moving your fielders around. But, the switch here is huge.
The sample size here tells us that 1 SD from the mean is almost 5 runs. The issue we’ll find is that this will probably be a problem throughout the position pairs. Right now, the “experience” seems to contribute around 8 runs.
Of the guys who pretty much split their time between the two positions, they were +0.2 as LF and +0.6 as RF. Pretty much a wash, and a good indication that the average LF and average RF are even.
For the LF/CF pairs, for the guys who split their time evenly at the two positions (BIP=10066), the fielder was +3.9 above the average LF and -6.4 relative to the average CF, for a gap of 10.3 runs.
For the primary LF, as a LF, he’s +3.6 above the average LF. Put him in CF, his unnatural position, and you should expect him to be below -6.4 right? He was -3.9, for a gap of 7.5 runs. His experience factor was 2.8 runs, but going the wrong way.
For the primary CF, as a CF, he’s -3.9 compared to the average CF. As a LF, we’d expect him to be pretty good there. He was only +0.8, a gap of 4.7 runs, for an experience factor of +5.6 runs.
Hard to tell about experience here.
For the CF/RF pairs, for the guys who split their time evenly at the two positions (BIP=5864), the fielder was -8.2 in CF, and -1.1 in RF, for a gap of 7.2 runs. For the primary CF, he was +0.5 in CF and -2.5 in RF, a gap of 3.0 runs, meaning the experience factor costs 4.2 runs. For the primary RF, he was +3.1 in RF, but a horrible -14.8 in CF, for a gap of 17.9 runs, or an experience factor of 10.7 runs.
Let’s recap the experience factor. LF/RF says 7.9 runs, LF/CF says 1.4, and CF/RF sas 7.5, for an overall effect of 5.6 runs. This experience factor is rather large. If you have a sample where it’s predominantly one-sided (mostly CF who play LF), then your comparisons will be heavily biased. And, this is among a set of positions that are fairly close in required attributes to play. It’s not like comparing 2B to 3B, or even worse, 2B to LF.
If we discard all the “primary” players, and look only at those players who played the two positions aroudn the same amount, we get these equations:
LF = RF + 0.4
CF = LF + 10.3
CF = RF + 7.2
Once again, we see that the CF is about 9 runs better than a corner OF, consistent with our original findings. However, here we see indications, from the CF perspective, that RF is harder than LF. The LF/RF numbers seem to swing one way or the other, depending on how you look at it. It’s probably fair at this point to simply call them even.
positional adjustment factors:
+6.0 CF
-3.0 LF
-3.0 RF
***
Updated: June 7 - Experience in the infield
Ok, let’s repeat the same thing with infielders.
2B/SS: the even players are +0.9 at 2B and -4.4 at SS, for a gap of 5.3 runs. That’s our baseline.
When primary 2B plays 2B: -1.8. When playing SS: -7.1, for a gap of 5.3 runs. No experience factor.
When primary SS plays SS: +1.5. When playing 2B: +0.2, for a gap of 1.3 runs, but going the wrong way. (We expected this player to be much better at 2B. He wasn’t._ This gives us an experience factor of 6.6 runs.
SS/3B: the evens are +0.1 at SS and +1.4 at 3B, for a tiny gap of +1.3 runs.
When primary SS plays SS: -0.6. when playing 3B: -7.9 runs, for a gap of 7.3 runs, going the wrong way. The experience factor is 8.6 runs.
When primary 3B plays 3B: +8.3. when playing SS: -8.7, for a gap of 16.9 runs! The experience factor is 15.6 runs.
2B/3B: the evens are -0.7 at 2B and +3.1 at 3B for a gap of 3.8 runs.
Primary 2B playing 2B: -0.4. At 3B: -9.5, for a gap of 9.1 runs going the wrong way. Experience factor is 12.9 runs.
Primary 3B playing 3B: +0.6. At 2B: -6.0 for a gap of 6.6 runs. Experience factor is 2.8 runs.
Adding it up again, the experience factor in the IF is 7.8 runs. For the OF we said it was 5.6 runs. We seem to have a pretty consistent pattern here. The experience factor seems to be around 7 runs. This is a huge consideration.
Again discarding the primary players, we have the IF as:
SS = 2B + 5.3
SS = 3B + 1.3
2B = 3B + 3.8
positional adjustment factors:
+3.0 SS
-1.0 2B
-2.0 3B
***
Updated: June 7 - IF, OF experience
Now, let’s get on to the IF/OF comps. Of players who had an even amount of time in both the IF and OF, they were +0.4 in the IF, and -7.6 in the OF, for a gap of 8.1 runs. Once again, clear indication that it’s harder to play OF than IF.
The primary IF, as IF, were -3.1. As OF, they were -6.7, for a gap of 3.6 runs.
The primary OF, as OF, were -0.8 runs. As IF, they were -7.9 runs, for a gap of -7.1 runs, going the other way. These OF had a tough time playing IF.
As well, there were 3 times as many IF to OF than OF to IF. Clearly, managers are thinking that their IF are better fielders, and using them to plug holes in the OF.
Our equation from these guys is:
IF = OF - 3.6
IF = OF + 7.1
This reduces to
IF = OF + 1.7
From these guys who are either primary IF or primary OF, we’d like to say that the IF is 2 runs better than the OF. From the guys who spend an equal amount of time in both, the IF is 8 runs *worse* than the OF.
The two scenarios you are left with are:
+5 CF
+4 SS
+0 2B
-1 3B
-4 LF
-4 RF
OR
+10 CF
+1 LF
+1 RF
-1 SS
-5 2B
-6 3B
The first scenario is the most believable of the two, but also the one that is least supported by the data. The second one is too incredulous to believe.
***
Updated: June 7 - Compare 1B to IF, OF, with experience
There are quite a few players that split their time evenly between 1B and the other IF positions (3098 BIP), as well as 1B and the OF positions (7144 BIP). The difficulty adjustment factor in both cases is 8 runs.
Among players who played 1B and IF evenly, they were +0.3 at 1B and -7.2 at 2B/SS/3B. The 1B/OF is +2.6 at 1B and -5.4 in the OF.
From this perspective, it looks like the IF and OF positions are rather even.
How about the guys who are primarily 1B, but play the other positions? Ouch. This is going to be bad. Get ready. -3.3 at 1B and -13.4 in the IF, for a gap of 10.1 runs. How about -2.4 at 1B and -24.2 in the OF (!), for a gap of 21.8 runs? Looks like it’s easier to expose a 1B’s holes in the OF. Note that these groups are fairly small, with around 800 BIP. 1 SD from the mean is 8 runs per 600 BIP.
The other way, the primary IF going to 1B were -5.1 in the IF and +2.3 at 1B, for a gap of 7.4 runs. The primary OF going to 1B were -8.6 in the OF and -7.2 at 1B, for a gap of only 1.4 runs. All this means that it’s 6 runs tougher to play the IF than OF.
So, we’ve got three different viewpoints here!
Sticking with the first scenario, our baseline including the 1B, and throwing in the C based simply on intuition, and we have:
+8 C
+5 CF
+4 SS
+0 2B
-1 3B
-4 LF
-4 RF
-8 1B
More to come…