THE BOOK--Playing The Percentages In Baseball

In addition to ISO and PF, there is also simply (TB-H)/PA, or EBF (extra base frequency), which is probably the most useful concept for overall evaluation. If 2 guys each have 100 extra bases in 600 PA, the guy with the fewer ABs will have a better ISO, but has he really contributed more to his team with his power?

Good point, David.  Most of the measures assume that you want to exclude walks from the denominator, but it’s not a given.  The per AB approach appears to have gained acceptance as a natural offshoot of Batting Average “logic”.

Another alternative viewpoint I failed to mention is that if you want a “value of power” measure, you could use outs as the denominator.

I don’t think that looking at who contributed more to his team is the right way to look at it here. I don’t think there’s a lot of utility to looking at a skill in isolation if you’re going to try to tie it into value.

If you have a player that bats 1.000, all singles, and a player that bats .100, all home runs, it’s pretty obvious that the first player is more valuable, and the second player has more power.

When we ask a question of who has the most power, we want to know about a particular hitting skill. If Mike Schmidt can have more extra-base hits than Rob Deer because he can make contact more often that makes him more valuable, but it doesn’t mean he has more power.

In short, the best measure of power - in my estimation - is one that isolates it from other skills, like taking walks, striking out or making contact. Either PF or slugging on contact makes the most sense to me.

Colin’s viewpoint in #3 seems to be that you can separate the skills of “power” and “contact”.  Maybe so, but I’m not convinced that is the case.

If Deer strikes out more because he tries to hit the ball to Mars with each swing, then his lack of contact is a result of his approach, and can’t truly be isolated from the power that results from that same approach.

On the other hand, we can assume that Deer has chosen the optimal approach to maximize his production within the confines of his skills, and then isolating power and contact is more possible.

My “best” stat would be the one that is the most repeatable skill. For instance, ISO can stay constant while a batter is walking more and hitting less, but this may reflect a greater level of respect that player is being given by the pitchers - the underlying skill has stayed the same. As I understand it, ISO is a pretty stable stat, while SLG has BA built into it.

Of course, HITfx values such as ave speed off bat and ave launch angle will tell us a lot more about a players actual power hitting skills.

I think Collin’s point #3 is probably what I want from a power metric. What I would like is a metric that allows me to take two players with the same contact rate, walk rate, strike out rate and answer the question “But who has more power?” Ideally, we would even be able to answer the question “over an above his ability to get on base, how many runs did his power contribute?” It seems to me that the appropriate way to do that is to consider some sort of metric over batted balls, e.g. (TB - H)/(AB - SO). Of course, something that can tell us how many runs over and above getting on base would need to use something like linear weights, not TB.

Patriot makes a good point that separating contact from power as skills is not something we can clearly do. Power hitters strike out more and make contact less frequently on swings. However, just because we can’t always separate these skills doesn’t mean that we can understand the relative contribution of power and contact within a skill set.

An absurd example is sometimes helpful. Let’s say we have a batter who gets 500 PAs, striking out 499 times and hitting a HR. Let’s say he does this every season, so this is his actual level of ability.

If Colin et al, using PF, want to describe this guy as having unparalleled power (and unparalleled inability to make contact), then be my guests. It doesn’t do much for me, though…

I agree with Dave. But let’s go further. There is a another batter who strikes out 499 times and hits a HR every season. In which case, the guy who hits the farther homers wins. There should just be an arm wrestling contest during the AS break.

Re Matt’s comment (#5), I agree completely.  I look forward to using the new hitf/x system and seeing how it correlates with the “usual” power numbers.  I would bet that sometime in the near future, batters will be evaluated for power based on some combination of the hit ball speed and launch angle, since these numbers provide a fielder-independent way of measuring power.

I wonder if one metric or the other might be a better way to look at minor league or young major league players who still have a lot of development time left?

Top ten closest career hitters to Rob Deer by ISO, min. 500 AB:

Jon Nunnally
Rocky Colavito
Gorman Thomas
Henry Rodriguez
Tony Clark
Bob Horner
Danny Tartabull
Steve Balboni
Roy Sievers
Gary Sheffield

By slugging on contact, or TB/(AB-SO):

Ted Williams
Juan Gonzalez
Craig Wilson
Marcus Thames
Richie Sexson
Lance Berkman
Miguel Cabrera
Willie Stargell
Jason Bay
Larry Walker

I’m not saying that Deer was as good a hitter as the second group, but the second group seems to me to be more descriptive of the sort of power Deer had.

For things like sim scores, projections, aging curves, etc., I think that seperating out power on contact from making contact has more value.

Funny, I would have picked the first group as being more similar. Most of those guys in the second group were/are able to hit for average and power - not something Deer was known for.

@rwperu34

I would say just looking at HR/AB would tell you a lot about the player’s power. Game power is something you got or you don’t. The quality of the pitching doesn’t matter so much. Bat speed and a power swing are independent of the pitcher.

There may be other issues like making contact, but the power is still there. I’m pretty sure league wide HR/AB goes up with each level in spite of the fact that the pitching is getting better also.

My first example in post 7 dealt with PF. Here’s one to deal with ISO and XB per out. Another player gets 500 PA and hits one HR while walking 499 times. His high ISO is certainly ‘isolated’ (to 1/600 of his PAs). And ignoring the problem of dividing by zero, his XB per out is high, not because he has great power, but because he draws so many BB.

All of the measures mentioned (extra bases per AB, per H, per batted ball, and per PA) will do a good job for realistic players. One way to select the best construct for general purposes is to look at extremes.

The construct that seems to hold up best is XB per PA.

I agree with Matt/12’s subjective choice of Deer as fitting better in the first group.  It’s not as if that’s some bunch of punch and judy hitters.  Those are all guys for whom power would be considered one of the strengths of their games.

Also, if the measure is power, I’m not sure slugging on contact is what you want to use at all.  A player who hit a single every time up would have a 1.000 SLG on contact despite displaying absolutely no power.  Singles should be removed from the numerator for any power measure, IMO.  The exception is Power Factor since singles bring that down; you could do EB/H, but it would just be equal to PF - 1 in any event.

I don’t disagree with David/14 that per PA is preferable to per out.  I just mentioned it as another metric one could offer if they were trying to satisfy a niche definition of some sort.

"One way to select the best construct for general purposes is to look at extremes”

I couldn’t disagree more. I’m looking for a metric that applies to real baseball, not fictional constructs whose abilities in no way resemble those of major league players. Even when we look at real baseball, looking to extremes is looking to outliers. Constructing your theory around the outliers will not typically produce a good theory.

The test of a power metric, in my opinion, is how well the metric can be incorporated in to a model that predicts run contributions as a function of two parameters, one of which can be taken to measure power. So, I want a function, f(x,y), where x is a metric for something (probably getting on base) and y is a metric for power, that as much as possible gives a value of the player’s total contribution.

Which metric is the best measure of power is an empirical question whose answer is determined by which metric is incorporated in the best model of scoring we can create.

David’s point was that looking at extremes you can expose the holes in an argument.

Our simple exam is how Runs Created is a flawed metric because it grossly overvalues the HR, the higher the run environment.  You can’t see that flaw in typical, or even extreme-typical teams, but it comes into play with extreme-typical players. 

In the same manner, if you are talking about “Power”, you are looking at extreme cases to make it abundantly clear what you are, and are not, measuring.  Patriot used extreme players, like Rob Deer and Mike Schmidt, to expose the argument.  David brought the argument to its extreme endpoint.

The point is NOT to create a metric that works at the extremes.  The point IS to illustrate what the metric is doing by looking at the extremes.

I agree with Tango about the validity of extreme player arguments, up to a point. But in this specific instance, I don’t think it works the way its being used. “Power” isn’t a defined concept like “run scoring” in that it has One True Definition.

If we’re looking at the traditional five scouting tools, hitting for power is separate from hitting for average. I think a useful power metric - one that we can use as a proxy for scouting observations - holds to this division. And I don’t know what other need one has for a pure power metric, as opposed to a simple metric of hitting value like wOBA.

Or look at it this way. Let’s say the question is, “Is a contact hitter or a power hitter better suited to this base-out state?” (For instance, you’d prefer a power hitter in a man on first, two outs situation, a contact hitter with a man on third.)

Using (TB-H)/PA leads to the conclusion that what we’d prefer is a guy that can do both, which is pretty obvious and not particularly helpful.

For me, the value is in trying to establish a profile of the player.  To that end, I need to know how hard, quick, and effective his swing is when he makes contact.  As Alan and others note, that’s described with the eventual HITf/x.  Until then, extrabase hits and/or HR per swing likely serves that purpose.

Colin:  I think trying to isolate power from other offensive value is fine.  But I’m not clear on why you think Slg on contact is any better than ISO in isolating power from BA.  Slg/Con removes strikeouts from consideration, but it confers value on singles in a way that ISO does not.  And looking at your two lists in post 11, I have to say it looks like the second list has a much higher average BA.  I think you’d find Slg/Con correlates more with BA than does ISO.

While a metric need not reflect total value, I do think a good metric reflects value within the dimension it’s measuring.  Ideally, a player with a high power score will provide more XB hits in x opportunities than a player with a low score.  ISO satisfies this reasonably well.  In general, a higher ISO = move value, leaving aside a player’s walks and singles.  (50 doubles and 15 HRs is more valuable than 5 doubles and 30 HRs, but in the normal range a high ISO is good.) PF fails this test.  By excluding not only BBs and Ks, but also outs on balls in play, we have no idea how much power (how many XBs) a given PF implies.  A metric should tell us something useful without first having to be multiplied by some other rate. But PF tells me almost nothing without other information about the player.  To me, that’s not a very useful metric.

I remain puzzled by Colin’s (implied) references to Isolated Power or its cousin (TB-H)/PA as “contact” measures. They’re really not; obviously, contact is a prerequisite for gaining an extra base, but they are leaving out 2/3 of the average player’s base hits!  Hardly a contact measure. 

Through 1995, among players with at least 3500 PA, Deer ranks

If we’re looking at the traditional five scouting tools, hitting for power is separate from hitting for average. I think a useful power metric - one that we can use as a proxy for scouting observations - holds to this division.

This is helpful, as it illustrates the fundamental difference in viewpoint between the two camps.  Colin and the scouts want to look at tools in isolation, and give Rob Deer credit for having the power tool in equal amounts to Schmidt even if he is incapable of deploying it in a game.  The scouts would also rave about Juan Pierre’s speed even if he can’t get on base and make use of it.  This viewpoint is fine in its own right.

The viewpoint I’m coming from is breaking down a player’s value into components.  How much of his value comes from his offense?  From his fielding?  Of his offensive value, how much is due to his getting on base?  His getting extra bases?  And so on and so forth.  Colin apparently doesn’t feel that these types of questions are interesting, and so to him, it is a puzzle why anyone would be interested in ISO or a similar metric.

Sorry, I was working on three separate points, and neglected the middle one about Deer’s ISO and its historical standing. 

Through 1995, among players with at least 3500 PA, Deer ranks 45th in Isolated Power.  I’m sure he’s been moved down the list since then, but I think this seems like a pretty generous ranking.  Obviously it all depends on your definition, but I don’t think it is by any means obvious that Rob Deer’s power should rank higher than 45th best (before league adjustments, park adjustments, etc.) over a 75 year period (~ live ball era).  ISO recognizes that Deer contributed a great deal of value with extra base hits.

I want to take back something I said about power metrics above, viz. “Which metric is the best measure of power is an empirical question whose answer is determined by which metric is incorporated in the best model of scoring we can create.”

There’s another thing that you might want to do, which is project a player’s development. For that purpose, it may turn out that certain metrics are better than others for measuring a player’s likely future performance. It might turn out, for example, (and this is just an example, I have no idea whether it holds empirically) that (TB-H)/PA does less to indicate future power than ISO, perhaps because contact skills can improve independently of “power” and when they do, ISO does a better job of predicting value in the future. I don’t think I’m being terribly clear.

"I remain puzzled by Colin’s (implied) references to Isolated Power or its cousin (TB-H)/PA as “contact” measures. They’re really not; obviously, contact is a prerequisite for gaining an extra base, but they are leaving out 2/3 of the average player’s base hits!”

Patriot: although ISO does exclude singles, it effectively includes the impact of Ks, while Colin’s preferred measure does not.  So in that sense he’s right that ISO includes a contact component.  But I don’t see why we want to treat Ks and outs on BIP as fundamentally different. Why exclude one but not the other?  And Slg/Contact doesn’t really eliminate “contamination” with contact, since it values singles.  I suppose (TB-H)/(AB-K) would give you a pretty pure non-contact power measure.  But to me, you’re then back to the same problem as PF:  a metric with no real relationship to value (and also very non-intuitive).

I don’t think (TB-H)/(AB-K) is all that non-intuitive.

If you called it “ISOBIP” then it would be immediately obvious what you were talking about.