THE BOOK--Playing The Percentages In Baseball

I could not find a reference on the internet to Polk’s original system. The web site linked to above describes their own system, which, according to them, is based on Polk’s, with some “tweaks.”

By and large, it is the usual B.S.

Of course it depends on what you are trying to do with such a system.  The usual dichotomy which applies to this kind of evaluation, is value, as in MVP/retrospective, and talent.  The former rewards or detracts performance based on things that are not in a player’s control or mostly not in a player’s control (or are partly in the player’s control and partly not), while the latter is intended to only quantify, and quantify properly, that which we think is solely in the player’s control.

The latter is designed to be able to estimate who would be the better player going forward, although, even then, sample size is a huge consideration.

The latter is also supposed to be as context neutral as possible while the former is hugely driven by context.

Most of the readers herein realize and understand the difference between an MVP type of stat and a talent type of stat.

Perhaps most importantly, when creating any metric which is supposed to describe someone’s talent OR value, you must make sure that each aspect of performance is valued correctly in relation to one another.

IOW, you cannot award someone 6 units of credit for a HR and 1 unit or credit for a single, or 5 units of credit for a good defensive play and 1 unit of credit for a hit. Or -2 units for a K and -1 for a batted ball out. Etc.

Looking at the list in this link (the “about” button at the top of the page),

http://royals.kansascity.com/about/

in the pitching section, points are awarded from -3 to +6.  +6, the most number of points awarded, comes from getting a win.

You probably want to stop reading after that.  If you do, I’ll point out some of the other really silly ones:

-3 points (the worst you can get for a pitcher) for a walk or HP at the start of an inning.

On offense, a SH or SF gets +2 points, the same as a double (and “heads up base running).

A walk and a single get the same number of points, +1, which is 1/4 of the number of points that a HR gets.

As Tango says, a called 3rd strike gets you -3 points.  That would be a lot of negative points for a lot of good hitters!

Again, how to criticize such a “system” depends on what it is being used for. In this case, whether it is used to determine who is going to be the more valuable player going forward, who was the more valuable player in retrospect (which is a whole other can of worms), or even as a teaching and development tool, it is pretty bad - really bad, actually. And the web site that Tango links to is an advocate of their system for MLB players.

What is amazing to me is how (apparently), “One of the greatest baseball coaches in NCAA history,” can have such a limited analytical understanding of the game…

He may be trying to do other things than an analysis of factors of the game.  He was a college coach, so there may be a lot of teaching and psychological preparation in his methods.  He may give lots of points for SH not because he thinks that it is a great play that should be done as often as possible, but because he wants the players to be emotionally invested in the team and their teammates.  We don’t know that he actually bunts more than MGL would.
If Barry Bonds, or another batter who knew the strike zone as well as Bonds did, took a called 3rd strike, there is a significant chance that the ump blew the call.  But few MLB players know the zone that well, and many fewer college players would.  He may be trying to teach them not to guess at the pitch, but to swing when it is close.
Yeah, Polk’s Points are fairly useless if you want to show what contributes most to winning games and who is the best player.  But as a system to focus the player’s concentration and make them better teammates, it may be useful.
(My “favorite”; SB +1, PK -3, CS no value listed??)

The only part of MGL/1 I disagree with is this:

IOW, you cannot award someone 6 units of credit for a HR and 1 unit or credit for a single, or 5 units of credit for a good defensive play and 1 unit of credit for a hit. Or -2 units for a K and -1 for a batted ball out. Etc.

In fact, you can very well do that.  For example, say that you are an “extreme” believer in FIP.  While FIP takes an agnostic view of batted balls, an extreme believer in FIP would take the view that batted balls are irrelevant with regards to the talent of the pitcher.

So, FIP gives 13 units for a HR, 3 for a walk, and ZERO for the other hits.

This is what I was getting at when I talked about inference.

Even for a hitter, you can make the case that a walk and a single are equal in impact, with respect to talent, because a walk has less noise to it than a single, and so, after you regress, you end up that having 10 more walks than average is worth the same as having 10 more singles than average, because you regress the walks less.

So, that’s why I can kinda see what Polk is TRYING to do here.  If you think about what kind of inference he’s trying to make based on the outcomes, then the basic idea is ok.  Of course, we have to justify the weights.  I can justify it for FIP and I can justify it for regression (such that BB=1B).

If Polk wants to say that a guy that has lots of SH is a “good” player because maybe it tells us something about his true talent, something we can infer, then that’s worth investigating.

My first thought: let’s assume a player has a x-2 count. If the next pitch is in the strike zone and the player doesn’t swing, he just recorded an out. However, if he swings, he has some probability of getting a hit p1. Furthermore there is some probability p2 < 1-p1 of getting a foul, which resets the count at x-2. This increases the probability of getting a hit to p1/(1-p2) (using a geometric series and assuming the pitcher never throws outside the strike zone).

Of course, that is overly simplified. The x in the x-2 count is potentially important, plus there is error in strike-ball calls. Even if the pitch is in the strike zone, if the umpire calls it a ball the batter wins and gets to see another pitch with a x+1-2 count, or walks. Hence there is some value to not swinging.

Continuing…

Personally this seems something that would be more important to in-game strategy. Calculate the probability of getting a hit if you swing, and the probability of the pitch being a ball if not, and proceed accordingly. So for each pitch you could calculate these probabilities using pitch or hit f/x data, and see if the batter did what he should have done in each case. I imagine better hitters make better decisions and we won’t learn much from this (though it could have an impact for player development).

However, for the statistic creation that we are talking about (assigning different point values to different events), it doesn’t seem like it will make much of a difference. A player who looks at a lot of third strikes probably walks (or waits around to see a better pitch) enough of the time to make up for the lost value of not swinging.

Off topic, but is there some reason we cannot post beyond a certain length?

There’s a lot I’d like to talk about in this thread, but I’ll start with the swinging third strike weighting. 

While I like (or, at least, am indifferent to) the framework involved, and think it’s mostly a question of assigning the proper values, the difference between the swinging third strike and the looking third strike doesn’t sit well with me (except, perhaps, on a 3-2 count). 

This seems like it should be a ratio (like bluffing in poker), because if a batter is always swinging on two strike counts then the pitcher will (should) be throwing a lower percentage of pitches in the strike zone (and vice versa).  The batter wants to take enough pitches to keep the pitcher honest, but not so many that he is counting on his judgement of the strike zone to match the ump’s all of the time. 

All of that is assuming the pitcher has the ability and inclination to locate precisely enough to make this a game theoretic discussion . . . perhaps in college ball it doesn’t matter as much.

With 2 strikes, there is not much game theory going on. The batter simply has to make a judgment on the pitch.  If he thinks that there is a certain chance that it is a strike, then he swings.

You can penalize a batter for a called third strike if he were a perfect judgment machine. But no batter is. All batters must take a certain percentage of called third strikes. If they don’t, then they are swinging at way too many bad pitches. That is why it is patently ridiculous to deduct 3 points for a called third strike. From a teaching standpoint, all you are doing is encouraging batters to swing too much.

Nothing is all bad. This “system” is not all bad. Overall, though, it is terrible. I don’t want to waste my time pointing out the things that are not too bad. It is not worth it in my opinion…

” If he thinks that there is a certain chance that it is a strike, then he swings.”

I should also say that the expected (by the batter) quality of the contact also comes into play. For example, if the count is 3-2 and the pitch is borderline, the batter might take the pitch if he thinks that chances of a strike are 60% and his contact is not going to be good, but he might swing if the chances of a strike are 60%, but he thinks he can make good contact. Obviously, the score, inning, base runners, etc., also come into play.

kds, I said in my first post, that it could be a teaching and development tool, but that I think it is terrible (overall) for that too. The called 3rd strike is a good example. These are not Little League players that are taking 40 mph called 3rd strikes right down the middle because they are afraid to swing. Batters already probably swing too much with 2 strikes and 2 or 3 balls. You don’t want to be encouraging them to swing too much. Automatically giving them -3 points for any called third strike is just a ridiculous system, whether it is for evaluation or teaching purposes, and, as I said, evinces a limited, shallow, and simplistic understanding of hitting strategy.

Again, if other people want to point out the good things in his system, fine. Overall, I give it a D or D- grade.

And don’t forget, this web site uses it to evaluate (not teach) MLB players. For that, I give it a solid F…

The other side of the coin of the called 3rd strike is drawing the walk, especially on a close pitch.

If the reward in Polk’s system is similarly ‘out of proportion’ for the walk as the debit is for the called strikeout, then I’m not sure I have a problem with it.

Usually, that’s not the case with traditional baseball people.

Good piece, Mr. Tango. Am a steady poster at Judging the Royals and would suggest that moving the decimal point two spaces left on Polk Points totals gives a surprisingly close agreement with fWAR. After Friday’s Royals’ game, Moose had fWAR of 1.0, Polk WAZ (wins above zero) of 0.84. Alex Gordon was f0.9, P0.83. Alcides Escobar f0.9, P1.04. Biggest difference among starting position players was with Eric Hosmer, f0.2, P0.95. This likely arises from points for HRs, RBIs, and outstanding defensive plays.

For those interested in the Polk System and its origins 35 years ago or so, this is from Judging the Royals, a site somewhat based on Polk as a method for evaluating plays and players within a game:

http://royals.kansascity.com/about/

Thanks.