ESPN ran two controversial articles yesterday: this one by Stats & Info and this one by Eno Sarris. Both mention that the Braves’ rotation, currently, has posted better results than the Phillies’. It’s a good story, essentially a David-is-beating-Goliath. However, these articles make the mistake of taking present-day results and extrapolating them further down the road, as if they will stay constant.
While the Braves have the Phillies beat in ERA by 12 points as of this writing (prior to the Sunday Night Baseball game on ESPN), the Phillies lead in the “performance” categories: K/9, BB/9, and infield fly rate. The Braves lead in ground ball rate, and not much else. The two teams tie in home run per fly ball rate.
Additionally, the Braves are getting an uncharacteristic performance from Jair Jurrjens, among others. His BB/9 sits at 1.5 currently, but his career average is 3.2. While it certainly could be true that Jurrjens made an adjustment to significantly improve his control, we don’t have enough information to make that conclusion yet, especially given the small sample sizes we are dealing with (in Jurrjens’ case, 29 and two-thirds innings). It does appear that he’s become much more of a ground ball pitcher, but that wouldn’t explain the control nor does it justify an ERA at a buck and a half.
When you hear the phrase “small sample size” you are likely to hear the word “regression” as well. As an example, Jeff Francoeur is off to a great start this year. If we want to predict how the rest of his season will go, we need to ask ourselves a question: do we trust the 138 plate appearances he’s had this year, or the 3,581 he has over his career spanning seven seasons — particularly 1,787 from 2008-10? You, being a smart individual, choose the much larger sample size.
So when people use the phrases “small sample size” and “regression” in tandem, they are using shorthand to say that Francoeur will more likely end up hitting like he did in the larger sample size going forward. That doesn’t mean that it’s impossible that Francoeur vastly outperforms, but given the information we have now, the most likely occurrence is that the career numbers are more representative.
That is what we need to do with the Braves’ and Phillies’ pitchers if we’re actually going to compare them and make conclusions about which staff is better. I cracked open my spreadsheet and ran some regressions.
I selected nine pitchers: five Phillies and four Braves (because Brandon Beachy doesn’t have any legitimate Major League numbers for regression). I gathered their innings pitched, K/9, BB/9, and balls in play (BIP) for two time periods: 2011, and a 2008-10 average. For fly balls, I used their career averages found on FanGraphs. Using this information, I found the standard deviation for the rates in both periods, then used that to regress each pitchers’ K/9, BB/9, and FB%. Finally, I plugged the numbers into the xFIP equation.
The standard deviations for each rate:
|2011 STDEV||2008-10 STDEV|
If you’re not familiar with the standard deviation, check out the link above. The standard deviation tells us how far away from the average the data lies, assuming the data is normally distributed. Take Halladay as an example. The standard deviation of his 2008-10 K/9 is 0.45, with an average at 7.7. That means 68 percent of the time, we should expect Halladay’s K/9 to lie between 7.25 and 8.15.
If you compare the three columns on the right (2008-10), to the three columns on the left (2011), you should notice that they’re quite smaller. What that tells you is that we are more certain of the 2008-10 data because of the much larger sample size. When we regress, the larger sample size will be weighted appropriately.
Currently, Roy Halladay and Cliff Lee each have strikeouts well above their career rates. The regression brings them down quite a bit. Similarly, both Jurrjens and Tim Hudson have walk rates well below their career averages, so the regression brings those back up. Additionally, the players’ fly ball rates were regressed; Joe Blanton and Derek Lowe were the most-regressed.
Here is a graphical look at the nine pitchers’ xFIP, 2011 against the regression. (Click to enlarge.)
The big takeaway from all of this is that all of the pitchers have pitched above their norms in one way or another. However, note that after the regression, the Phillies’ “big four” is… well, the top four, all at 3.52 or below in terms of xFIP. The Braves’ best is Derek Lowe at 3.61. Their best by results, Jurrjens, ends up the worst among the nine.
Have the Braves’ starters had better results than the Phillies? Yes. Have they performed better? Absolutely not. Overall, the Braves have a staff BABIP at .269 while the Phillies are at .287. Last year, the Oakland Athletics had the lowest staff BABIP at .274. If the Braves had a staff of fly ball pitchers and played in a spacious home ballpark like Safeco Field or Petco Park, I may be inclined to believe their BABIP, but neither condition is met. Rather, the Braves have a staff of ground ball pitchers (recall that ground balls become hits more frequently than fly balls) benefiting from an individual BABIP .264 or lower. On the Phillies’ side of things, Roy Oswalt is the only one particularly low at .247. Cole Hamels is at .271, but that is close to his .285 career average. Halladay, Lee, and Joe Blanton are each above .300.
The Phillies’ staff is better, based on what we know about each of the nine pitchers studied. It could be true that one or more of the Braves’ pitchers has significantly improved, but we cannot say that with any certainty at the moment. If you’re the betting type, bet on the Phillies’ rotation being better than the Braves’ going forward.
This incredibly informative article by Sal Baxamusa at Athletics Nation, circa March 2008.