Okay, it has been a while since my last post on here, but things have been kind of crazy for me.
First, let me say congratulations to the St. Louis Cardinals and their fine fans. While I think it is safe to say that they probably weren’t the best team in baseball this year, they were the best team when it mattered most.
Here’s my latest article on the Orioles Hangout.
Yet another outstanding article from the fine folks at THT can be found here. And yes, it is on my favorite subject. I can hardly wait for the THT Annual. Be sure to order your copy today.
One thing that has taken a lot of my time is that I had to rebuild my stats database after my hard drive crashed on me. I now have the pitching stats for every pitcher from 2002 through 2006. Included in these stats is batted ball data. Having this nice database of information gives me the ability to perform a lot of studies over the winter. I’m open to any ideas that you may have. Just remember that I’m not a mathematician, so I may not be as thorough or as clean with my analysis as the fine folks at sites such as BP and THT.
One of the first studies I wanted to do was one on the consistency on specific stats such as ERA, various theoretical ERAs (think DIPS and FIP ERA), K-Rate, and so on. I can now do that. To do it I first had to narrow down all the instances in which a pitcher appeared in back to back seasons. Rodrigo Lopez, for instance, ended up with four pairs of seasons (2002-2003 / 2003-2004 / 2004-2005 / 2005-2006). I lined up each of the Year 1 and Year 2 ERAs, and then performed a simple correlation test on Excel.
I wanted to compare the correlation of ERA to those of various theoretical ERAs. I used three versions each of four fairly common theoretical ERAs – DIPS, Component ERA (ERC), FIP, and XERA. ERC was invented by Bill James, XERA by Ron Shandler and the folks at Baseball HQ. This site gives the basic formula for each. FIP ERA is sort of the cousin to DIPS.
I actually use three different versions of each theoretical ERA. The first version (ERC1, XERA1, FIP1, DIPS1) is based on the standard non-adjusted stat. I also use DIPS 3.0 instead of either of the original two versions created by Voros McCracken (so my numbers would be different from those posted on ESPN. For version 2 of each (ERC2, XERA2, FIP2, DIPS2) I adjusted the hits and innings pitched totals using batted ball data. For version 3 of each (ERC3, XERA3, FIP3, DIPS3) I made the same adjustments as version 2, but took it a step further by normalizing the number of infield flyballs and line drives for each pitcher to the overall MLB average for each stat for the 2002-2006 seasons. The next trick was to filter out pitchers based on innings pitched totals as I went along.
This list shows the correlation for each stat for all pitchers. The totals for the 1,848 pitchers were 156,276 IP in Year 1, and 152,186 IP in Year 2:
ERA .068
ERC1 .075
ERC2 .084
ERC3 .204
DIPS1 .199
DIPS2 .185
DIPS3 .265
XERA1 .118
XERA2 .120
XERA3 .271
FIP1 .139
FIP2 .202
FIP3 .251
The highest correlation for any of these was only.271, which isn’t very high. It is telling that ERA was so low when compared to anything other than ERC1 or ERC2. Note that XERA3 had the highest correlation.
This list shows the correlation of each stat for all pitchers with at least 25 IP in each season. The totals for the 1,250 pitchers were 137,732 IP in Year 1, and 135,054 IP in Year 2:
ERA .263
ERC1 .301
ERC2 .482
ERC3 .578
DIPS1 .549
DIPS2 .562
DIPS3 .593
XERA1 .321
XERA2 .503
XERA3 .623
FIP1 .402
FIP2 .576
FIP3 .587
The correlation for ERA became a little better this time, even though it was still less than it was for XERA3 in the first chart. Once again, XERA3 had the highest correlation – with a very respectable .623.
One last list – this one showing all pitchers with at least 75 IP in each season. The totals for the 557 pitchers were 90,998 IP in Year 1, and 89,583 IP in Year 2:
ERA .381
ERC1 .430
ERC2 .595
ERC3 .680
DIPS1 .665
DIPS2 .681
DIPS3 .688
XERA1 .452
XERA2 .601
XERA3 .706
FIP1 .553
FIP2 .687
FIP3 .688
Once again we see improvement for ERA – even though it was still lower than every single theoretical ERA. XERA3 was also the king once again, even though ERC3. DIPS1, DIPS2, DIPS3, FIP1, and FIP2 also did quite well, and weren’t far behind XERA3.
So now we have an idea of the consistency of each theoretical ERA. In the next installment I plan to evaluate the success rate of each at predicting whether Year 2’s ERA will go up or down. We’ll also look at whether the difference can be used to predict Year 2’s ERA.
After I had begun typing this article I stumbled upon this posting on another site. This guy essentially argues that the theoretical stats are too busy and that we should just focus on K:BB ratio (or Command Rate, or whatever else you might want to call it) as they are more consistent. So I decided to do a quick correlation test. Using pitchers with at least 75 IP in each season, K:BB had a correlation of .598. Pretty good, but eight of the theoretical stats were higher. Having said that, I think he makes a pretty good argument, and I plan to study it further.
One final set of correlations to mention. This list shows the correlations of various rate stats. The columns are 25 IP, and 75 IP.
H/9 .392 .471
HR/9 .207 .356
HB/9 .330 .408
BB/9 .521 .653
K/9 .720 .768
GB% .756 .821
FB% .722 .734
IFFB% .187 .241
LD% .088 .066
HR/OFFB% .083 .199
BIP% .735 .779
Most of the above is not a surprise, but there is one big surprise – at least to me. I was floored that the Year 1 to Year 2 correlation for H/9 was higher than the Year 1 to Year 2 correlation for HR/9 – and it wasn’t really even close.
I was concerned that my methodology may have been wrong, but the data available in the 2006 THT Annual was consistent with my numbers, so I’m pretty confident they are right.
I’ve seen it written many times that pitchers have a lot of control over whether a batter hits a home run, but not nearly as much control over whether a batted ball becomes a hit. The above tells me we’re either giving pitchers too much or too little credit (depending on your point of view). One thought that crossed my mind is that home run data was skewing the data, so I ran correlations for (H-HR)/9. I came up with a correlation of .447 – not much less than H/9, but still higher than HR/9.
Interesting.
