How well does a pitcher’s W-L record indicate his value?

It’s obvious to most people that it doesn’t do a very good job at all, but I decided to look at some data to support that claim.

To do this, I looked at the 76 pitchers who had 30 or more starts in 2004. I compared their W-L record to their ERA. While ERA is not the best indicator of a pitcher’s ability, it does have the advantage of being well-known and understood.

The correlation coefficient for these pitchers was -0.36, which means that when a pitcher’s ERA goes down, his winning percentage can be expected to increase at a moderate rate.

But by breaking these pitchers into bins, I noticed that the data trends are really inconsistent. In particular, look at the winning percentages below.

ERA	# Pitchers	AVG(W)	AVG(L)	Winning %
0.00 - 3.00	6	15.7	9.3	0.627
3.00 - 3.50	10	14.6	8.2	0.640
3.50 - 4.00	11	12.0	10.8	0.528
4.00 - 4.50	18	13.0	10.4	0.555
4.50 - 5.00	18	11.8	11.3	0.511
5.00 - 5.50	7	11.8	12.1	0.495
5.50+	6	11.9	12.0	0.500

The winning percentages trend downward, but it goes up, down, up, down, down, and then up. Not exactly a smooth line.

Looking at this table further, W-L record seems to do a good job of differentiating between a pitcher who has an ERA above 3.50 and another with an ERA below 3.50. But it does very little to distinguish between a pitcher with a 3.50 ERA and one with a 5.50 ERA.

In fact, if we remove the first 2 bins (great pitchers) and just look at the 60 pitchers with an ERA above 3.50, the correlation coefficient shrinks in magnitude to -0.13, which is clearly weak.

All of which says that a pitcher’s teammates (offense and bullpen) have a tremendous influence on his W-L record. But everyone except John Kruk already knows this...