THE BOOK--Playing The Percentages In Baseball

Don’t know if you meant this to happen but if you look at the overall average for each position, its lining up with the positional adjustment.  I know you didn’t mean for this to happen so you might have to normalize each position but it lines up really nicely.  Kind of cool.

Bryan, great insight!

I guess MLB managers have already made their choices clear based on who is playing there.  It’s so obvious once you think about it.

Let me think about that while I compare to the results:
http://www.tangotiger.net/tools/index4.php

And to what I have been using.

You guys can stop voting now if you like.  I’ve got what I need.

This was the weights that the Fans Scouting Report currently uses:

In Fi Sp Ha Re St Ac Pos
4 1 1 4 4 4 4 C

4 4 1 4 2 1 1 1B
4 4 2 4 2 1 1 2B
4 4 2 2 4 2 1 SS
2 2 1 2 2 4 2 3B

4 8 8 4 2 1 1 LF
4 8 8 4 2 2 1 CF
4 8 8 4 2 2 2 RF

Basically, I put 1, 2, 4, 8 as I felt was correct.  I don’t think there’s much agreement between my guesses and what you guys are saying.

I’m going to post the positional averages for the scouting values tomorrow to see if that helps.

Just curious why you have your weighting options as 1,2,4,8 when you have five choices?  Or maybe you just imagined four choices when you created this?

Looking forward to tomorrow’s data and analysis.
vr, Xei

What I did a long, long time ago was run the FSR through a factor analysis, and came up with this:

I called them “Field,” “Range” and “Arm.”

Colin: totally forgot about that.  Is it just me, or has there been so much good work put out these last five years, that it’s overwhelming to remember?

Xei: the 1,2,4,8 was something I’ve been doing for 8 years.  The 1-5 thing was just put out yesterday.  The numbers do not have a 1:1 relationship.

Here are the weighted (by innings) Fan averages for the 7 traits by position (the talent numbers):

2 52 38 33 52 53 54 51
3 55 42 35 58 50 48 52
4 56 57 55 52 55 48 57
5 56 51 46 54 56 61 54
6 63 63 62 60 65 65 59
7 41 49 56 42 42 44 46
8 59 71 75 62 54 49 50
9 53 56 58 54 58 64 60

Easiest is to compare 2B to 3B: the third number (Speed) is much higher at 2B than 3B.  The second to last number (Arm Strength) is much higher at 3B than 2B.

That is, we are finding players at those positions based on those traits.

If we look at level of importance (our current poll), for 2B/3B, we get these 7 numbers:
4 3.5 3.9 3.0 4.1 3.8 2.5 3.5
5 4.5 3.6 2.0 4.1 3.8 4.4 4.1

We see that the third number (speed), fans think is much more important at 2B than 3B.  And for the arm strength (second to last number), it’s much more important at 3B than 2B.

So, this is what Bryan’s insight would have predicted.  Rather than ask fans what do they think is important, simply look at what kind of players are at those positions.

Anyway, I’m going to look at both sets and see if there is any disagreement, and then compare to my initial thoughts.  I’m going to re-read Colin’s article, and I think there was someone else who also did something along these lines at THT.

Thanks again guys…

Using the Bryan method (comparing tools of players to fans’ expectation of level of importance), we see the following results by position:

catchers
- strong agreement on 6 of 7 traits
- some disagreement as to reaction/instincts (my original standards had the catcher number here close to the fans’ expectation, rather than the talent level found)

firstbasemen
- disagreement on the 3 throwing categories; possible that fans did not value as position-independent
- my numbers closer to the fan level of importance

secondbasemen
- fairly strong agreement across the board in the Bryan numbers
- my numbers not that good

thirdbasemen
- strong agreement on the Bryan numbers and my original numbers

shortstops
- disagreement on the speed number, otherwise very strong
- my numbers not that good

leftfield
- strong agreement across the board
- my numbers not that good

centerfield
- disagreement in alot of places between all three sets of numbers

rightfield
- strong agreement in the Bryan numbers
- this one, I think I’m right… the fans and the teams, have high numbers in the arm strength category… but, you really need to be able to run… it doesn’t help much if you have a strong arm (Hawpe), but can’t cover the ground

Anyway, I’m going to redo my numbers, taking into consideration the tools numbers, and the importance numbers.

Again guys, really appreciate the involvement.  Here are my proposed weights.  They can be read relatively both horizontally (within position) or vertically (across positions):

Here’s how I did it.  First, I used Colin’s splitting of the 7 traits into 3 categories:
catch: Instincts + Hands
range: first step + speed
throw: release + strength + accuracy

And, I gave each of the three groups equal weight.  I semi-arbitrarily assigned 216 points, of which I split equally into three groups of 72, 72, 72.  So, all the catch numbers for the 8 positions will total 72, and so on.

The next constraint was that for each position, the sum equalled to seven times these numbers:
6 C

5 SS

4 2B
4 3B
4 CF

3 LF
3 RF

2 1B

You’ll notice that those numbers are similar to the number of plays per game, as well as following the fielding spectrum.

Then, I went to the fans’ level of importance, and started breaking up the numbers, and tried to adhere to the above constraints.

It all came out pretty close, and I’m reasonably happy with how things turned out.

Thoughts?

I updated the 2009 and 2010 reports based on the new weights.  I doubt you would notice any difference.

Tom, might be a dumb question but I want to understand what you are doing here.  Could you please explain what it is you have updated based on the “new weights”.  And maybe explain your process a little bit.

ie - Fans Evaluate -> Results -> Apply weightings -> Convert to UZR type #?

Hah, actually I can’t really take credit for anything except tripping your subconscious.  My confusing comment actually was talking about something else.  I like this method you are talking about (that you came up with).

I’ll try to explain my previous comment a little better.  If you take the results from the survey in this post and look at the average rating at each position, there is a good linear fit between this and the positional adjustment using in wins above replacement.  Strictly looking at the poll you gave us, we didn’t do a good job when you said “Don’t compare position to position.” because on average we said skills at SS are more important than skills in LF. 

I was just amazed that we were able to replicate the value of defensive positions so well by rating the importance of skills.

Xei: take the weights I posted, and apply it to Ellsbury at LF and CF, and tell me what his weighted average is at each position:

http://www.tangotiger.net/scout/index4.php?teamid=111&team=Boston%20Red%20Sox

J.Ellsbury
LF: 57.48
CF: 60.78

For each of the seven categories I summed up the (X * (W/T)) where X was the value listed for the player in that particular category, W was the weight for that position in that category, and T was the total of all weights for that position.  I am not sure if this is the correct methodology.  How did I do?
vr, Xei

Looks like you did fine.

So is a rating of 50 no longer the average rating for all these positions?  The weighting adjusts the numbers.  In the case of all three outfield positions you are going to see most if not all of the numbers adjusted up quite a bit.
vr, Xei

So we can use all of this to tell us if a player is at a position which is not ideal.  Is that valid?

If all of this data was perfect we could compare his skill ratings weighted by positional requirements to data from something like UZR to see how much “experience” (the difference between the two) affects a player’s outcomes.

"So we can use all of this to tell us if a player is at a position which is not ideal.  Is that valid?”

That’s been one of my favorite applications of this data in the past.  Would love to read articles based on this approach.

50 is the unweighted average for the whole league.

When I convert to runs (which you’ll see soon), I used the weighted average (weighted by IP).

Bryan: right, that’s the theory, to the extent that fans are evaluating independent of position.

Right, that’s my favorite part too.  For example, in 2008, players most similar to Jeter were playing 3B or RF:
http://www.tangotiger.net/scouting/sim2008_5406.html