THE BOOK--Playing The Percentages In Baseball

Nice stuff!  I also commend Derek for his clear writing style and explanations.  I hate it when I have to read things 2 or 3 (or 10 in some cases) times to understand them, which is often the case with these types of articles.  Not so with this one.

I had a couple of thoughts as I read through both articles.  I wonder how changes in pitching approaches impacts a hitter’s numbers (in these categories).  That would especially be true for rookies and perhaps even for older players who have skills that are diminished.  Or even for mid-age players who have changed their approaches by design.

Also this sentence is a little troubling and seems to be one of the flaws of “fantasy thinking.”

You can use it throughout next season to see which players are changing their approach or getting better in terms of their plate skills.

Even though these categories have a high “inter-class correlation,” Derek (like Pizza sometimes does) fails to tell us the average or range of sample sizes used for the correlation.  Remember these kinds of correlations are meaningless numbers without some kind of sample size given (for any skill or statistic, regardless of the luck factor involved, as long as there is some skill involved, the inter-class or other correlation will approach zero for small samples and 1 for large samples).  I assume, though, that they represent around a year’s worth of data for each player in the sample.

Anyway, one has to be careful about inferring anything about “true talent” when looking at changes, patterns, trends, etc., in any statistic.  That is especially true when tracking a statistic often throughout one season.

The words “getting better” or “changing their approach” imply a change in true talent.  If you are tracking these statistics throughout a season and you see them changing from day to day, or even week to week, even by a lot, it is unlikely that any given player is “getting better or worse” in true talent, or “changing their approach.” It is much more likely that you are seeing random fluctuations associated with any binomial or multi-nomial where you have a true “p” or series of “ps”.  At what point can you start to “conclude” that a player has changed his approach or is getting better or worse?  Of course there is no magic number for that.  It is a sliding scale.

It would be lovely if every analyst would agree not to use the word “trend”.

Thanks for the feedback, guys!

I can’t believe I left out the sample size info.  Sorry about that.  I used all hitters with at least 100 PA (player needed 100 PA in both seasons for inter-class correlations) and used data from 2006 and 2007.  I thought 100 would be sufficient given the work Pizza Cutter did here ( http://www.mvn.com/mlb-stats/2007/11/14/525600-minutes-how-do-you-measure-a-player-in-a-year/ ) showing that these pitch-by-pitch stats stabilize very quickly.

This gave me 846 player seasons for the K-rate correlations and 747 player seasons for the inter-class correlations.

As far as the comment “You can use it throughout next season to see which players are changing their approach or getting better in terms of their plate skills” goes, perhaps I should have clarified a little better.

I didn’t mean that we should be checking it every day to see how the player changes his approach throughout the season (though I see how it could have been interpreted that way).  What I meant was that, come, say, May, a fantasy owner notices that Player X’s strikeout rate has dropped 10 points from last year, and he’s interested in trading for him.  He can use these stats - taking sample sizes and things of that nature into consideration, of course - to see what has changed and help him to make the decision.

And I definitely think it would be interesting to see what happens to these for older guys with declining skills and younger guys as pitchers start to figure them out.  Perhaps I’ll take a look at that sometime soon.

Anyway, thanks again!

I prefer to take a liberal view of plate discipline. That is, with as much allowance as possible for different batting styles--batting styles being of course grounded in the player’s skill set.

So, I prefer to start from the point of view of the batter instead of the pitch, and ask the question, “When this batter chose to swing, how often was the pitch a strike?”. This is not one of the stats used by Carty or Pizza. The batters who do well are not all ‘bat on the shoulder’ types, and include some guys who like to take their cuts, like Hawpe, Jeter, and Ordonez.

The formula, using the Fangraphs format, is ZSw% * Z% / Sw%

The natural companion question would be, “When the batter chose to take, how often was the pitch a ball?”. But I don’t pay much attention to that, because it penalizes a batter for taking strikes. Batters take strikes all the time, and for good reasons. Until there is a way to distinguish ‘good takes’ from ‘bad takes’, it’s probably better to refrain from judgement.

Hey David,
Thanks for the comment.  Obviously there will be issues with any framework until we can truly isolate cold zones, but the issue I see with your idea is that the batting style of many hitters is to intentionally swing at balls out of the zone.  They think they can hit them, and those who are right shouldn’t be considered at fault.  If you know you can make contact when you swing at a particular pitch, regardless of location, I don’t think it’s a mistake (ignoring what happens once the ball is put in play) if the pitch is out of the zone.

Vlad, for example, makes contact with nearly 70% of OOZ pitches he swings at.  There are some hitters (Jack Cust, Mark Reynolds, Russell Branyan, for example) who struggle to make 70% contact when they swing at pitches in the zone!  For what it’s worth, the formula you gave produced just an 0.04 correlation with K%.

I definitely agree with you that batters have different styles and that the methodology I used isn’t perfect; I just think it’s the next logical step.

A hitter is credited with a correct decision if he is able to foul off a pitch outside the zone, btu an incorrect decision if he swings and misses. But didn’t the fould ball turn a ball into a strike? That puts the hitter at a disadvantage compared to not swinging. And, does someone like Vlad get as many hits on balls in play from his OOZ swings?

For each pitch, we should classify by count, pitch type and pitch location (xy as well is in or out). The batter must decide to swing or take. There is a mean value for swinging, and another for taking. Use the lw100 formula.

A disciplined hitter would be swinging when there is enough benefit, and taking when a swing is too costly.

----"the batting style of many hitters is to intentionally swing at pitches outside of the zone.”

And that is, almost by definition, poor plate discipline---UNLESS they can hit those pitches well. Vlad can’t--his OPS on batted balls OOZ is around .600, as opposed to his OPS on batted balls in the strike zone of 1.200+ or whatever. That .600 is better than most other hitters can muster, but it’s still helping lose games. A better example might be Ichiro, who seems to show very little difference in his OPS on batted balls in or out of the zone. But even in his case, and the 5% of similar cases, he is still losing out on some additional BB he would get, without any change in his BA or SLG, if he were able to be more selective.

And Brian, your post is exactly what I was trying to rail against--the idea that a batter should be judged as a correct or incorrect decision on a pitch by pitch basis. All batters, and particularly smart batters, are/should be more focused on the entire PA dynamic than on a pitch by pitch Lwts evaluation.

But, there is certainly/likely some benefit to looking at pitch by pitch, but I think it should be quite liberal or generalized, as a concession to the fact that PA outcomes are what matters. So, I advocate the stat I advocated a couple posts ago, as the best way to navigate these waters.

I am quite open to contrary information, or opinions--so, go for it.

David, I didn’t know we were thinking in opposite directions. If not pitch by pitch, how would you model it?

I was trying to think like a batter, even though I was never very good myself. Yes, the batters mindset should be to get the best possible result of the plate appearance, but to do so he must face a sequence of pitches, and make a decision on each. I relize that not all pitches are the same.

I propose a model in which we define a cost and benefit to swinging or taking a pitch, based on count, pitch type and pitch location.

There is little cost to taking a 3-0 strike, but there is huge cost in taking one with a two strike count. Batters who make 97% contact can afford to take a marginal strike in search of a better one to hit, but one with a 70% contact rate has to view strikes as a precious commodity not to be wasted.

So, does the batter swing when it is in his benefit, and not when it would cost him?

Sure, Brian, if you can include enough in the model to make it realistic, then pitch by pitch is fine.

But if all that is done is charge the batter a generic run award (for an avg batter, maybe the difference between swinging at or taking a generic strike is .04 runs, or whatever it is), not much good will come of it.