Some PrOPS for JB
Posted by ubelmann on Monday, April 30th, 2007 at 2:56 am
Jason Bartlett's hitting just .221 so far this season, but I'm not too concerned about it. It's been said that Bartlett's been hitting the ball hard with bad luck, but we've all heard that song and dance way too many times in the past. To help cut through the (half-baked) crap, we can (indirectly) ask Baseball Info Solutions whether or not JB's been hitting the ball hard.
Specifically, we can check out his PrOPS. As we find in The Hardball Times glossary:
PrOPS stands for "Predicted OPS." It was developed by J.C. Bradbury amd introduced in this article. PrOPS isn't really a new stat; it's a formula for predicting what a player's OPS is likely to be in the future based on his batted balls, strikeouts, home runs and walks.
Without any further ado:
.309/.367/.393 -- Bartlett actual, 2006
.283/.346/.373 -- Bartlett PrOPS, 2006
.221/.293/.250 -- Bartlett actual, 2007
.301/.364/.380 -- Bartlett PrOPS, 2007
.290/.346/.391 -- Bartlett ZiPS projection, 2007
Sing it with me: "One of these things is not like the others/One of these things just doesn't belong." Over the course of 500-600 PA, even 300-400 PA, actual AVG/OBP/SLG is probably going to be a better measure of a hitters' ability than counting up his line drives and such. But over the course of roughly 70 AB, I'm comfortable saying that Bartlett's been closer to .301/.364/.380 than he has been to .221/.293/.250.
After all, the majority of the difference between his PrOPS and his actual production this year is about four singles and a double or two, so it doesn't take a whole lot of bad luck to explain his results so far. And for fun, let's compare his 2007 PrOPS to my non-random sample of hitters:
.301/.364/.380 -- Bartlett
.283/.356/.380 -- Cap'n Jetes
.266/.286/.397 -- Michael Young
.260/.342/.327 -- Punto
.291/.318/.348 -- Tyner
.267/.316/.320 -- Castillo
Loading ...
I like this type of stat, but am a little confused in its underlying factors, which wasn't clear to me from reading the link. From reading, it appears the factors included are batted balls, strikeouts, home runs and walks.
In comparing actual isolated discipline (OBP-BA) to PrOPS ISO they differ by approximately 10 points. How does batted ball data affect walk rate, or why is his walk rate projected to decline by that amount, while his BA increases by about 80 points? 2006 stats actual to PrOPS differ by about 5 points; a relatively small amount I suppose, but I don't understand how they come up with a difference.
Either way, it appears to be clear that Bartlett has been unlucky and should be hitting lead-off or in the second spot with that .350+ OBP.
In comparing actual isolated discipline (OBP-BA) to PrOPS ISO they differ by approximately 10 points.
Walk rate isn't OBP-BA, since OBP and BA are fractions with different denominators, so you're going to get slightly wacky, but roughly close, answers by subtracting batting average from OBP. Whether or not a ball in play falls for a hit will affect OBP-BA, but not BB/PA.
PrOPS also now takes into account player "speed scores". I still think this is probably the wrong approach (albeit perhaps mildly so) -- he should estimate separate sets of coefficients for different hitter "types" rather than assuming that all outcomes are drawn from the same probability distribution, controlling for what he controls for. I'm not convinced that his explanatory vars adequately capture the salient differences between, say, Nicky Punto and Torii Hunter (two players with similar GB/FB ratios, LD rates and K rates, but very different HR rates), or between Slappy Castillo and the Good Doctor.
if the data really cluster the way I think they do, he's systematically underestimating performance for some types of hitters and systematically overestimating it for other types. I just haven't looked enough at his results to know for sure.
Like I sort of said in the article, if you really want to know how a player hits, just wait 500 PA. At this point in the season, though, I think it's pretty reasonable to suspect that the random variance in a player's actual AVG/OBP/SLG is a bigger concern than the difference between a Joe Mauer line drive and a Jason Tyner line drive.
I’m not convinced that his explanatory vars adequately capture the salient differences between, say, Nicky Punto and Torii Hunter (two players with similar GB/FB ratios, LD rates and K rates, but very different HR rates)...
HR rate is included as a component that goes into PrOPS, so if Hunter and Punto have vastly different HR rates, they should have vastly different isolated slugging according to PrOPS. And they do, as Hunter has a .240 PrISO so far while LNP has a .066 PrISO so far.
I mean, if a big part of your concern is players like Punto and Hunter or Castillo and Morneau being confused for one another, a cursory glance at this year's and last year's data shows you that shouldn't concern you at all.
Like I said, probably not a huge effect. I certainly would agree that the random component probably swamps any systematic biases in the estimates associated with omitted variable bias arising from unmodeled heterogeneity in player types.
the modeling question is: do the included variables adequately capture the functional differences between players of different types (e.g., Punto and Hunter, or Castillo and Morneau; or Punto and Castillo; etc., etc.).
Well, Castillo and Punto are similar in terms of speed, but very different in terms of hitting styles. Do the included vars account for (most of) the differences? That is a testable hypothesis.