I'm Reminded Of 2005

it’s hard to argue against a record-setting team. But again, going by the feel-test, that team never felt like a clear cut favorite, and prove such feeling when it didn’t put up much of a fight against the Yankees in the ALCS (4-1, no?).

I also remember that during a Neyer chat while the A’s were up 2-0 and looked like they were going to march onto the ALCS, someone asked him which of the A’s and the Mariners should be favored, and Neyer saying something to the effect of “it’ll be a toss-up, but the A’s are more talented.” I hope I’m being annoying with all this feel-out memories, but the point is that that A’s team felt/looked the part, and put up the numbers.

Also, I’m not the biggest stat head, but would those stats you listed suggest that we should expect some sort of regression this season similar to the regression predicted for Cahill before he became a strickout pitcher?

Not so well now. We need to do much better than eyeball a BABIP of .279 based on our experiences of the last decade or two and say “that doesn’t sound lucky to me.” On this planet, rigorous drug testing has been set up and ERAs have declined down to where they are now approaching their levels during the first 15 or so years of the DH era. Or if you prefer to speak in euphemisms we can find some odd change like ball composition that they’ve just coincidentally enacted when they started cracking down on steroids. Whatever floats your boat.

My ERA numbers for the A’s staff are not “cherry picked” — the are the most obvious ERA numbers to use to judge the performance of the young A’s staff.

As for “fielding independent” numbers like xFIP, they remove most of causes variation in ERA, namely the hits. In the name of removing luck they also remove most of the predictive information. If the A’s staff’s talent consists mainly in reducing the number or hits or extra base hits, xFIP won’t reflect that and there will be a persistent disconnect between the two.. Models have far stronger correlations to ERA if they include hits. So unless we actually need to be fielding independent, and we don’t here, I prefer statistics that don’t throw out most of the information. I myself have an ERA model that closely correlates with historical team pitching. It predicts the near-record performance by the A’s starting staff this year that I expected during the first week of the season based on observing their pitching once through the rotation, though obviously my model is subject to the objection that the A’s starters have been getting so few hits more out of luck than skill.

So let’s look at one of those tools for estimating luck, namely BABIP. If you run regression analysis on league-average BABIPs vs. ERAs, you’ll find there is a good long-term correlation between them that can’t be explained by random luck. Pitchers in the 1960s weren’t consistently luckier than pitchers in the 1970s, who in turn weren’t consistently luckier than pitchers in the 1990s. An important part of the relationship between ERA and BABIP must be explained by persistence factors, whether they be fielding or pitching is irrelevant for this purpose. From 1974-86 when the league average ERA was in the 3.70s the league average (Fangraph) BABIP was in the .270s. In the 1960s league average ERAs and BABIPs were both even lower.

Of course in the shorter term there is a great deal random variation for individual pitchers, and some even for teams, but there is no magical “average BABIP” by which to compare everybody. If the A’s starters were really good enough to sustain an ERA of 2.20, we’d expect them to have a BABIP below .260 based not on any random luck but on the long-term persistent relationship between ERA and BABIP. If they are good enough to end the season at 3.10 then we’d expect them to on average end with a BABIP of about .270. So, like I said, the .279 BABIP is slightly on the lucky side of where they have been performing. You can only say a BABIP of .279 is not on the “lucky” side of the chart if you already come in with a preconception that they “should” regress to a mean of about 3.70 or higher.. In other words, it’s a judgment based not on the BABIP but on the preconception that 2.20 or 3.10 is “too low”. But then you don’t need BABIP, you can just say that 2.20 or 3.0 is “too low” and be done with the argument.

Offensive statistics vs. previous March/Aprils are also down again this year.to “an alarmingly low level.”