Even the staunchest of SABR-followers would (or SHOULD) admit that the statistical models they've developed doesn't account for EVERYTHING that goes on. As an engineer all my analysis of steel structures are based on MODELS, not the real thing. While I recognize that my models aren't the real thing, they close enough to make meaningful conclusions from (the crane won't collapse at full load). Statistical analysis of baseball is the same thing. The best run estimators come very close to estimating the amount of runs a team produces in a year, but they will ALWAYS have an error associated with it. The model simply can't take into account every variable; Some variables are just too random (how many suicide sqeezes will result in a run? How many will even be attempted in a year?) or unquantifiable (fundamentally sound teams stealing a win here and there). That's not to say we can't extract meaningful conclusions from that run estimation model, however. If the run estimator can consistantly predict the amount of runs scored for a team over a season to within 5%, then we KNOW we've accurately modeled about 95% of what makes a team score runs. We can then begin weighing the things we CAN quantify to the things we can't. All mathematical models of complex events WILL have an error, ESPECIALLY when something like people are involved. Provided a person doesn't overstate the accuracy of their model, though, they can give valuable insite on whatever phenomina is being modeled. The most simple run estimating formula: runs = AB * OBP * SLG That can predict runs scored over a season to within 95% accuracy. If that can't convince people how powerful stats are in baseball, nothing will.