Using FIP as an example: it's not an experimental result. It's just a formula that incorporates events that have already transpired (e.g., strikeouts). It can't have 'error' in a statistical sense. However, the degree to which FIP (or a similar statistic) is useful depends on the properties of the underlying events. One inning pitched tells you basically nothing regardless of how you want to express the results (ERA, FIP, xFIP, etc.). After 5000 innings, you can safely use ERA without worrying about regression issues. Metrics that regress more heavily are better for small samples, because of the tendency for regression to the mean. In other words, xFIP is preferable to FIP for a 20 inning sample, because the former assumes pitchers have little control over HR/FB rates. But in reality, pitchers do have some control over that, and after a couple seasons it's better to look at FIP. Stats writers have gone to very, very great lengths breaking everything down. Just have to look for it.