Awesome now THIS is awesome. Consider this. Let's say we take players the first three months of the season, minimum 200 PAs. We then split up their PA's into two equal sets, randomly. You can do this with a rand function very easily. If you don't want to do that, just alternate them and throw them into two sets that way. I chose the first method. Then in those two sets, calculate OPS. Keep in mind that these two sets have 100-175 PAs max. Then for the players calculate their true talent level the best you can. A simple way is to weight the three previous years, hence why we wanted to look at the first three months of the season. You can also use a projection system to come up with one. Then take the difference between their "true" OPS and their actual OPS's for each set. If it were truly random like you all (and they) say, then we would expect this correlation to be zero, or very close to it. Since doing well (or poor) in one set of 125 PAs shouldn't have anything to do with the other set of 125 PAs. However, when you look at the data you'll find a correlation that's in the .30-.40 range between the differences. Meaning that the two are certainly related and that high fluctuations in short term true talent level exist. It's even more surprising considering that each set is around 150 PAs. Thus, their hypothesis can be thrown out the window and they are simply wrong. Wouldn't the correlation depend upon how comparable the ~300 PA "sample" OPS is to the "true" talent value? If the difference is great between the "true" value and the overall "sample" value, I would be shocked if there wasn't some correlation between the randomized subsets. Unless the sample is VERY small or the sampling VERY weird, two random subsets of ~300 PA of say, .650 OPS can be expected to be close to .650 as well. If the true OPS is .800, then the difference will be close to .150 for each data set. You should reasonably expect a positive correlation, especially over many iterations, which would statistically suppress outliers. Maybe I'm missing something very important in the analysis, but it seems that the only place a near zero correlation might crop up would be a stretch of PAs where the player produced very near his "true" talent, in which case, what are we all arguing about, again?