THE BOOK--Playing The Percentages In Baseball

First of all...this looks awesome.

Secondly, it looks like the Runs per Win is not a constant.  I know 10.5 is the rule of thumb conversion, but I am not sure where this actually came from.  Can someone provide a brief derivation of that relationship?

Thanks

A great addition to a great site.

Using Pythag, assuming 735 runs per 162 games for an average team (or the 2008 NL):

735^2/(735^2+735^2) = .500, or 81 wins

This is always true, by the way - Pythag will give you .500 for any team who scores the same runs as they allow.

Add 10 runs scored:

745^2/(745^2+735^2) = .507, or 82 wins

Subtract 10 runs allowed:

735^2/(735^2+725^2) = .507, or 82 wins

The numbers shift around a bit as you change your run environment. Let’s try 600 runs a season, adding ten runs on offense:

610^2/(610^2+600^2) = .508, or 82 wins

Now let’s try 800:

810^2/(810^2+800^2) = .506, or 82 wins

The trouble here is that in a 162 game season, there are discreet win percentages available - Pythag isn’t aware of this.

That’s the basic theory, anyway - results may vary depending on which version of Pythag you use - I kept it simple for the sake of illustration.

Another question…

Brandon Phillips for example has 305 expected outs in 2007.  According to Hardball Times and play by play data, Phillips had 352 plays made plus 6 field errors.  This breaks down into:

282 regular assists
36 DPs started
33 Force outs started
1 other fielders choice

I am trying to figure out how this reconciles to a RangeRuns of 10.9.  Any help would be appreciated.

Thanks.

I believe UZR only includes ground balls for infielders. I could be wrong on this.

Those numbers are groundball only.

Looking over the Retrosheet data, defining a play made as a:

* Ground ball
* Where the batter did not reach on an error
* Where the second baseman was awarded an assist and was not the pivot of a double play
OR
Was awarded a putout without an assist

I come up with 345 plays made. Hrm.

At the speed of I Dream of Jeanie (sorry for dating myself), we blinked, and here we see Manny’s actual salary compared to his WAR-based salary:

http://www.fangraphs.com/statss.aspx?playerid=210&position=OF

I had thought that an I Dream of Jeanie movie was made a few years ago? After searching the net, I discovered that the movie I was thinking of was Bewitched. For some reason my mind had merged these two shows into one.

Is it really the best way to compare--to have actual salary side by side with an expected free agent salary? The salary structure of baseball is a blend of free agent prices combined with non-FA salaries. So, why not create a “deserved” salary which reflects this? What is the chance that a 0 WAR player is a FA, or other? What is the chance that a +4 WAR player is a FA, or other?

I’m not sure how to do it the best way, but I can see that posting all deserved salaries on the FA scale will not “add up” to the actual MLB team salaries. It seems misleading.

I don’t know, I think people understand that 0-3 year players get paid nothing, and that 4-6 year players make a fraction of their market value.  I don’t know that people will look at the data and say “wow, Evan Longoria signed for way too little”.  Hopefully, they’ll look at it and say “hey, developing talent from within is remarkably cost effective”.

The next step in the process is to apply the discount levels as I do, meaning that for pre-arb, they get almost nothing of their value, at arb they get a diminishing discount for each service year.

So, if Pujols deserves a “free agent” salary of 30MM one year, but he’s in the 2nd year of his arb eligibility he derseves a “CBA” salary of 18MM, that’s fine.

After all, those WAR conversions are really based on what the free agent $ is worth. 

I suppose one thing David can do (if he has the data) is highlight in bold when the salary the player earned was AS a free agent (or when his eligibility would have come up), so the fan can see that “yup, Pujols gave a discount in his last year of arb-eligibility”.

http://www.fangraphs.com/statss.aspx?playerid=443&position=OF#value

Accd to fangraphs, in 2007 Juan Pierre was actually underpaid. He’s a net -0- above replacement but the 2.5 win adjustment for CF makes him worth $10.7M. Is this some sort of lower bound where the system breaks?

ken griffey jr: paid $73.3M, worth $17.4M. wow.

Accd to fangraphs Juan Pierre was actually underpaid in his abysmal 2007 season:

http://www.fangraphs.com/statss.aspx?playerid=443&position=OF#value

He’s showing a net -0.2 run value for batting & fielding. But his 2.5 win positional value for CF makes him worth $10.7M. Is this a case of the replacement player value losing its meaning at the lower bound? Shouldn’t a player of net zero run value be worth $350K plus a very minor positional adustment?

A player of “net zero value” relative to average is an average player. The positional adjustment of 2.5 is simply to line everything up correctly, because UZR measures performance relative to position, and we want performance relative to league. So if the numbers are right, you’ve got a guy that’s roughly +10 on defense, -7 on offense, or about +3 relative to average on the whole. If a league-average ballplayer is worth $8-$9 million, then Pierre would be worth more than that.

If you want to argue that wRAA or UZR are incorrect about Pierre, feel free. Bring evidence. But wRAA/UZR aren’t wrong simply because you don’t like Juan Pierre. (I don’t like Juan Pierre either.)

I guess it’s the replacement level add back of 24.3 runs that I don’t follow. So a true replacement level player would net -24 runs and end up back at zero “value wins.”

Just curious, where did Dave get the salary data from?

big/17: Right, you have to understand what “0” means in the calculations.  Sometimes it means “average” and sometimes it means “zero”.

***

Salary, plus all, data is available here:
http://baseball-databank.org/

So are we confident about this “defensive replacement level” being slightly worse than the average MLB center fielder but 7.5 runs better than the average corner?

I don’t have any data to back this up, but it would seem to me that there are many above-average defensive players freely available.  But maybe those players are below replacement level bats?  Or are offense, defense, and position all considered when we assign +20 or whatever for replacement level?

There are replacement-level PLAYERS, and not anything else.

And yes, if you are 30 runs below average or worse, as a player (meaning off+def+pos), you should be out of MLB.  If you are at -15 or better, you are in.  Everyone between -15 and -30 is all the players that are bouncing between the pros and minors.

It’s as simple as it sounds.

Thanks.  I’m sure it’s been said hundreds of times before, but I needed to hear it again apparently.

Does anyone have any thoughts on the B Phillips discussion from #4-7?

Thanks again.

Anyone?

Only MGL can answer you.

MGL, he is asking how in 2007 Brandon Phillips can have 304 expected outs (according to bUZR at Fangraphs), be close to league average in that regard, yet have 433 assists.  If you take out some of the 113 DP (where he was the pivot man), and some OF assists as the relay guy, you should end up with around 360 assists where he was the main fielder.  That should be the “actual” outs.  If he was league average, we’d expect his “expected” outs to also be around there.

Exactly the question I am trying to figure out - thanks for the help.

This is a direct request of MGL - can you please help us get to the bottom of the question outlined in post #25?  Thanks in advance.

Tango,

Just thought I’d let you know I’ve posted my WAR calculations from 1955-2008.  Here’s the Hawk:
http://www.baseballprojection.com/war/d/dawsa001.htm

Seriously, why will no one answer this question?

birdo: I can only surmise that MGL is not auditing this thread.  If you go back to the main thread, you will see mgl’s byline on several threads.  His name is hyperlinked with his email address.  Feel free to email him directly, and ask him to audit this thread.  MGL is usually more helpful than I am, which says alot for him.

I’d have to look at the numbers more closely.  Where is the data?  What year(s) are we talking about?  About retrosheet, BTW: for example, if a fielder fields a ball and does not get anyone out, it is listed as a fielder’s choice no matter what.  I call that an error.  I don’t know how many there are.  It would be coded on retrosheet as a 19 (FC) but with no out made on the play.

MGL, the question is in post #4.  Fangraphs is showing Brandon Phillips with 305 expected outs in 2007, and a range runs of +11 runs.  That would mean that he must have gotten around 320 or so real outs, more or less.

However, his actual plays made (on groundballs only) is somewhere around the 350 mark or so. 

The question therefore is where is the gap?  Are there some 30 or so plays that you are simply not accounting for?  He had around that many forceouts, as an example.

The number of expected outs may simply be lacking expected DP outs since UZR’s DP rating is calculated separatedly from the range run rating.  The amount of Expected DPs + Expected outs + range runs is very close to matching Actual outs + Actual DP outs, within plus or minus 2.

Peter, all outs, including DP outs (1 out only), are counted in the basic UZR.

MGL - I know from your explanation in the other thread that you count them, I was just wondering whether they are in the number that is being reported on fangraphs as ExpO (305 for Phillips in 2007).

MGL - any luck on looking into those 2007 Phillips numbers more closely?  I really appreciate it. Thanks.