Because of variance and the fact that not everything that is measured has the same amount of it. What we are trying to measure with a defensive metric is essentially "how likely is a player to make a certain type of play." Is it 70%, 90%, 99.4%...? Well the recorded measurements are only "Did he make a given play or didn't he?" He never makes 70% of the play. There is variance in the observed measurement. Throw in that on the defensive side, most plays are "routine" and the number of games needed to get a reliable estimate is large. Nobody (reasonable) claims that the data accumulated in smaller samples isn't true. They simply say/know/claim that the amount of data to get a reliable estimate is larger than the amount of data that they have. As an analogy that may make more sense, if we are trying to find the average height of men in St. Paul and we measure one fellow at 6'1". We don't discount that he is actually 6'3", but we also don't know yet if the average height of men in St. Paul is 6'1". We will keep measuring other men to deal with the fact that there is variation. The more measurements we gather, the more accurate our estimate can become. None of our men have the "wrong" height, but none of their individual heights is the answer we are looking for. I don't know if that explanation makes things more clear or less.