There are two statistics that can be used to verify if a pitcher's stats are genuine or a result of luck or misfortune: Batting average on balls in play and strand rate. Most know that the league average BABIP is around .300, and that a pitcher's BABIP tends to regress to that benchmark if his is above or below it.
I'd guess that, however, the majority of fantasy players do not know or have never heard of strand rate.
Strand rate, or left on base percentage (LOB%), is the percentage of baserunners allowed that do not score a run. The formula is
(H + BB + HBP - R) / (H + BB + HBP - 1.4 x HR).
It may sound like a pitcher's LOB% is a function of his skill or ability to get out of jams, but most of the time, it isn't. (See graph below.)
Among all pitchers with 300 batters faced in back-to-back years, the r-squared of LOB% from one year to the next is only 0.048.
The reason why this issue is so important is that LOB% is highly correlated with ERA. Using the same group as before, the correlation of LOB% and ERA is -0.76. The higher the strand rate, the lower the ERA. You can see the significant relationship here.
I then grouped players based on their LOB%. I rounded a player's LOB% to the nearest 0.5 percent, and added up the stats of all players in each group with a LOB% between 65 percent and 80 percent (anything outside of that had less than 50 players in the group).
I found the combined ERA of each group. The correlation of the group LOB% and the group ERA? Almost a perfect relationship: minus-0.995.
The quick-and-dirty way to predict a player's ERA based on LOB% is
-14.8 x LOB% + 14.5, which obviously has a 0.76 correlation with actual ERA.
(I also tested the correlation of LOB% with the difference of ERA and FIP, which is an ERA estimator that only uses strikeouts, walks, and home runs allowed to estimate ERA and assumes an average LOB%. This turned out to be -0.77, which is almost the same as LOB% with ERA. The amount of luck associated with a player's statistics correlates with LOB% slightly more than ERA itself!)
Putting two and two together, we can make these two assumptions:
1. Pitchers with a high LOB% will see theirs decrease the next year, and vice versa;
2. Pitchers with a high LOB% will likely have a low ERA, and vice versa.
The league average strand rate hovers around 72 percent. Any deviation from that number can be caused by random fluctuation or pure chance.
(Star pitchers and most relievers, however, may be able to control their strand rates—Johan Santana's career LOB% is over 77 percent, and Octavio Dotel, to name an average middle reliever, has a career LOB% of 75.5 percent.)
Justin Verlander's 2008 campaign can be described as a disappointment. So can Javier Vazquez's. But the truth lies in LOB%: Verlander's was 65.4 percent, and Vazquez's was 68.3 percent; each had a LOB% just under 75 percent in 2007, and both players' FIPs were 0.65 points less than their actual ERA. They are due for a turnaround.
As is Nate Robertson. Robertson's LOB% was 64.3 percent, seven percent less than his 2007 rate. As expected, his ERA was exceptionally high, at 6.35, but his FIP was only 4.99. His ERA should fall to near 4.76, his ERA in 2007.
Ian Snell is another example of someone negatively affected by a low LOB%. His strand rate last year was 69.3 percent, almost six points less than a year before, and his FIP was 0.85 lower than his actual ERA.
Just a year ago, he had a 3.76 ERA—why wouldn't you take a stab at him in the later rounds of your draft?
On the other end, Daisuke Matsuzaka had a strand rate over 80 percent after a 73.9 LOB% in 2007. His FIP was more than a full point higher than his ERA—expect a decline from him.
Joe Saunders' LOB% was 75.7 percent, three more percentage points than his 2007 strand rate; his FIP was 0.96 points higher than his ERA, and—get this—was a tenth of a point higher than his 2007 FIP (even though his 2008 ERA was a point lower than the year before).
Take the time to look through strand rates to find late sleepers. You may just find the next Cliff Lee—his 2007 strand rate was 62.5 percent.