Sometimes for no reason at all I like to look at baseball related things and analyze them. There is really no larger purpose to this article except I had a curiosity and looked to solve it. Essentially my question is what makes a number one starter a number one starter and so on and so forth. Is it a certain ERA, strike outs, being tough to hit. What is it that defines a top pitcher. Well I’m not a big statistical researcher so I don’t like to delve into things too deep without some sort of general idea what I’m looking at so I usually make some sort of assumption, see what that gives me and proceed from there. Well this is my first attempt at answering that question and as usual I went with an assumption. I decided for general purposes to call a #1 starter a player who posts a WAR greater than or equal to 4.0 and for each subsequent spot I drop 1 WAR (ie a 3.0-3.9 WAR player is a #2 starter). I then wanted to know what the ERA, WHIP and K/BB were for each group in 2012. The result can be seen below:
4.0+ WAR: 3.15 ERA, 1.142 WHIP, 3.78 K/BB
3.0-3.9 WAR: 3.43 ERA, 1.193 WHIP, 3.05 K/BB
2.0-2.9 WAR: 3.95 ERA, 1.286 WHIP, 2.71 K/BB
1.0-1.9 WAR: 4.33 ERA, 1.333 WHIP, 2.45 K/BB
0.0-0.9 WAR: 4.85 ERA, 1.434, 1.92 K/BB
-0.1 or lower WAR: 6.14 ERA, 1.588 WHIP, 1.46 K/BB
On its surface that makes sense to me. I always felt that roughly speaking the difference between each rotation spot should be about a half run of ERA. While that isn’t exactly what I have here is it is roughly close. However upon closer look I’m not sure these numbers really tell the whole story. I looked at the percentage of starts made by each group of pitchers:
In an ideal world the first 5 groups would all be even at 20% and the last group would not exist but of course we all know this isn’t an ideal world. Still the low % of starts made by pitchers on the top end suggest I’m setting the bar too high. I decided to take a different approach and order the pitchers by WAR and then split them into 5 groups each of which started approximately 972 games. In case it’s not obvious where that number comes from that is the number of major league baseball games in a year multiplied by 2 since there are 2 starting pitchers divided by 5 since there are 5 rotation spots. I didn’t bother with splitting the data set exactly as I’m just looking for a rough figure here:
#1: 3.23 ERA, 1.162 WHIP, 3.48 K/BB
#2: 3.76 ERA, 1.255 WHIP, 2.88 K/BB
#3: 4.35 ERA, 1.324 WHIP, 2.50 K/BB
#4: 4.28 ERA, 1.344 WHIP, 2.25 K/BB
5: 5.64 ERA, 1.539 WHIP, 1.64 K/BB
Relatively speaking the effect this new classification has on #1 and #4 starters in minimal but the #5 starter group is greatly impacted by the dreadful performance of the negative WAR group. Also the #2 and #3 starters are severely dragged down and it is actually an insignificant amount separating a #3 and a #4 starter. Needless to say there are plenty of problems with this data set.
In the past I have always used the rule of thumb that a 3.25 ERA was the average for a number one starter and that for each additional starter you add half a run. So number two would be 3.75, number 3 4.25, number 4 4.75 and number 5 would be 5.25. I’m not sure how accurate that really is though and my above attempts really don’t do much to enforce the opinion or not. Taking one last look at this for now I decided to ignore all the pitchers who posted a negative WAR and all the pitchers with fewer than 5 starts and focus on the remaining group. This left me with 203 starters totaling 4406 starts or about just over 880 starts per rotation spot. The results can be seen below:
#1: 3.20 ERA, 1.157 WHIP, 3.56 K/BB
#2: 3.74 ERA, 1.254 WHIP, 2.87 K/BB
#3: 4.17 ERA, 1.303 WHIP, 2.60 K/BB
#4: 4.31 ERA, 1.334 WHIP, 2.37 K/BB
#5: 4.96 ERA, 1.447 WHIP, 1.90 K/BB
Once again I am surprised by the relatively little separation between a number 3 and number 4 starter. Also surprisingly in all 3 measurements the values for a number 4 starter as remained relatively constant giving me a fairly good idea of what a number 4 starter really is. Once again the #5 spot is subject to some wild fluctuations based on a handful of horrible starters. My method made an attempt to remove some so the number is not way out of line.
Bottom line I think I need to study all this in more detail and try a few other methods of looking at how to define each rotation spot but I think these three methods give me a good starting place. My original idea of a .5 run step in ERA appears to fit the first 3 rotation spots fairly well but the drop off to the fourth small is less significant while the drop off to the 5th spot is difficult to determine which much accuracy but it can be though of as around .5 run. I want to make an attempt to research this more in-depth at a later date and I’ll make sure to share my findings but as an initial survey I thought this was fairly interesting while not all that informative.