THE BOOK--Playing The Percentages In Baseball

I dislike G/F ratio myself, but on the grounds that the variables of interest are denominated by AB: GB%, LD% and FB%.  Those are ratios, too, though.  Just normalize them.

How is GB% a ratio? 

I guess the way you are saying it is that A/(A+B) is a ratio.  But, it’s ALSO a rate.  A/B is NOT a rate. If you are going to get semantical like that, then I mean a non-rate.  If I can’t use the word ratio, then give me something else I can use.

Let’s add SO/BB ratio to the list of bad ideas (though one intended to capture an important relationship).  A 9K/3BB pitcher is in no way comparable to a 6K/2BB pitcher.

Guy, totally with you.  Guy was the first one who alerted me to the possibility that K-BB per PA would be better.

And he’s totally right.  A 6/2 K/BB ratio is equivalent to a 10/6 ratio (given the same number of batters faced).  I provided empirical results as proof a year or two ago.

Ratio: comparison of two numbers that are either both ‘true numbers’ (like 5/3) or both ‘like denominate numbers’ (like 5ft/3ft) and no longer has a unit of measurement.

Rate: comparison of two numbers that are ‘unlike denominate numbers’ (like 5km/3hr) and create a new unit of measurement.

For whatever that is worth to you.

I find GB%, FB%, LD%, K%, BB%, etc to be the most helpful in understanding the pitcher’s value. 
GB/FB or K/BB dont tell the whole picture, but these do.  They are essentially ratios of GB/AB, FB/AB, etc. 

Then we can find a value for each outcome and sum up the percentages to determine the expected value of the AB, for that pitcher.

I’m not so sure I see that much advantage to GB% over GB/FB in determining a pitcher’s value, going forward at least, beyond just the non-linear issue that Tango talks about.  But that’s more or less an issue of how you present the data.  GB/FB can easily be turned into a rate that looks just like GB%.  Once you do that, how much do you gain by adding line drives into the equation?  As far as the difference in the information provided by each, one basically regresses LD rate 100% and the other not at all.  For one season’s worth of data, I’m not convinced looking at the one that doesn’t regress LD rate at all (GB%) is any better than the one that regresses it 100% (GB/FB).  I’d have to see some sort of data that suggests that GB% is a better estimate of GB-inducing talent than GB/FB (or the form of it turned into a rate), particularly for pitchers with large differences in LD rate like the article cites.

If all you care about is how many ground balls were actually allowed in the sample, then I guess that is different.

In order to regress LD “fairly”, I use (GB-FB)/BIP.

This way, you keep the relative value between GB and FB constant.  And you get the denominator right.

It’s so darn simple, too.

You could try analyzing the ratios in logspace (I do that sometimes for work when the ratios vary by more than a factor of about 5).  It’s possible that the GB/FB ratio is a lognormal distribution.  This would take care of the symmetry, since a 1.0 would be equally spaced between 0.5 and 2.0, and it doesn’t matter if you do GB/FB or FB/GB as far as calculating averages or variances.

While LD% is extremely important for pitcher’s success, pitchers who cannot prevent line drives don’t pitch in majors. So, pitchers in majors should be the very top of line drive prevention, and have small difference in their natural ability, enough that GB/FB from LD perspective isn’t a problem. The fact that GB/FB isn’t symmetrical is a problem and should be taken into account in formulas - so, no straight linear equations.