THE BOOK--Playing The Percentages In Baseball

I honestly don’t get all of this:

The run value I provide has been altered to account for the position differences.  And I put it against 500 “BIP”.  I use the BIP per game:
pos BIP adj
2 0 12.5
3 2 -12.5
4 4 2.5
5 3.5 2.5
6 4.5 7.5
7 2.5 -7.5
8 3.5 2.5
9 2.5 -7.5

His “original” UZR is actually +121 (!!) runs in 585 G, or 33.5.

And your numbers include positional adjustments?  For example, if a SS were +10 UZR (compared to all SS) after regressing, you are adding some number of runs because he is a SS, as compared to say, a LF’er who was +10 in UZR (compared to all LF) after regressing?

And your lists are per 150 games?  per 500 BIP?

I know that your tried to get this out quickly and we all appreciate the data, but I am lost with the way it is presented.  Perhaps you can redo the original post with some clearer explanation.  Unless I am the only one or am in a small minority (who does not understand what you are doing with the numbers).

Numbers are per 500 BIP.

UZR is runs above average for that position, with the average SS being far different from the average 1B.

The adjustment presumes what the average player at each position would do compared to Willie Bloomquist.  The average SS would be +7.5 runs compared to Willie and the average LF would be -7.5 runs, etc.  The adjustments are per 162 G.

I’ll illuminate more on Monday.

Of course.  My bad.  I see “regRuns500.” Plus, I see that Pujols is minus so you obviously did the positional adjustments.

Amazing numbers.  Most people would not believe some of the names on the list and would not believe how much of a run/win difference there is among players on defense.

All people have to is to add regressed offensive lwts (and base running, which is small) to get a total player value.

Remember it is per 500 BIP.  For a LF, that would mean 200 G.  For a SS that would be 111 G.

I mentioned a similar thing on the TB Rays odds thread, but it is definitely more useful to see each player’s value per game rather than per 500 BIP. That is especially true if you don’t know the BIP per game for the various positions.  How would I know the relative value of various players at different positions if I didn’t know their value per game (or per inning)?

At the top of the thread, I listed the number of BIP per game, for each position.

Aha, I guess I’m just happy not to see Jeter’s name on the negative list.

traced/7 - I’m going to guess that it’s partially because of the positional adjustment and partially because he apparently had a better-than-usual season defensively this year.

Well, yes it is almost impossible for a SS to be one of the worst defenders among all fielders, after you adjust for position. I mean you have lots of terrible corner outfielders who fill that position easily (Griffey, Lee, Burrel, Manny, Dunn, etc.) Jeter as a corner outfielder would presumably be much better than them.  Probably about average actually.

His UZR the last 4 years

05 -10
06 -13
07 -29
08 -4

That is very bad, but not atrociously bad.  I’m not sure there are any atrocious SS right now.  There shouldn’t be anyway. If you have a really bad SS, you simply move him. If you have a really bad outfielder, what do you do?  There are only so many people who can play 1B.

Hopefully I can phrase this question correctly:

Say you are trying to estimate how many runs a team will be above/below average on defense in an upcoming year.  Would the correct method for this be to take the players’ regressed, position adjusted numbers and then reallocate them at the positions they are expected to play including “undo-ing” the appropriate position adjustment?

Jeter is -5 per 500 BIP from 2003-08.  Among the 196 infielders (2B, SS, 3B) with at least 120 games, that puts him tied at #164-172.

Among the 59 SS, he’s tied at #53-55, with Michael Young and Felipe Lopez.  Lopez is a horrible SS, according to UZR, WOWY, and Fans.  Below them are guys with less than 200 Games: Keppinger (-6), Santiago (-6), Brendan Harris (-9), Russ Adams (-10).  Russ Adams was also hated by Jays fans.

***

Here’s step-by-step what I did:
1. Added a year field to MGL’s data

2. Merged all the data into one table (data at this point is broken down by player, year, pos… I have 5825 records)

3. Added a positional adjustment for each record as: ADJ/162*G.  The ADJ figure is listed by position at the top of this thread.  This figure is called pos_adj.

4. Added a BIP estimate for each record as: BIP*G.  The BIP figure is listed by position at the top of this thread.

5. Added a BIPouts estimate for each record as: 0.65*[BIP]+([runs_total]+[pos_adj])/0.8.  runs_total is MGL’s UZR runs.  So, what I’m doing here is adding his UZR runs number to the pos_adj, and dividing it by .8 to get it into outs above Willie Bloomquist.  I then add that to 65% of BIP (not really important what I add it to, I just wanted to use something that could mean that 65% of BIP are outs).

6. Add up all the records per player.  At this point the position and years go away, and we have one record per player.

7. Add 400 BIP of league average performance as regression toward the mean.  Or, similarly, multiply the figure in #6 by BIP/(BIP+400). (*)

8. Present all data as regressed runs per 500 BIP.

***

(*) This I should fix.  The regression should happen a bit earlier.  Otherwise, I am adding a league average component (0) that is the same for Pujols as for Everett.  This is wrong, since we know Pujols plays 1B and Everett SS.  I need to add the ADJ figure, proportionately to how many games he played at each position0.  Won’t matter much, except for guys with few games played.  But, I’ll fix that soon enough.

Ok, I fixed the regression component so that players are regressed to the mean to the pos_adj value.  As a result, all 1B come out as a negative. I’m not too happy about that.

***

One of the things I have to work in is a better way to translate numbers between positions.  For example, we can accept that the average fielder at 1B is worse than the average fielder in LF, PER BIP.  However, what happens if the LF is involved in twice as many plays as the 1B?  Then what?  The translation is not necessarily a linear one, but a multiplicative one.  I still have to work on that.

philston, #10, no need to undo the positional adjustments, I don’t think, since they should all “balance out” anyway. I’m not sure though.

Certainly, if you did everything that Tango did, without the positional adjustments, that would tell you what to expect from each team defensively next year, if you know how many innings/games each player is likely to play of course.

There’s really no need to position-adjust, and then un-adjust for players like Utley and Crawford, who only play one position.  Otherwise, yes, you’d have to do that.  Think of it like park factors.  You don’t need to park-adjust Helton’s numbers to a neutral park, and then readjust it park to Coors.