While we know we're always working with a probabilistic model when dealing with projection, there's a lot more at work than a simple mathematical formula--and hopefully, if that were simply the case we'd see more sophistication and variation than such vanilla regression towards the mean. That a metric of pure data could deduce something like plate appearances based on such abstractions as injuries and minor league development and assumed playing time is damningly tenuous. There's biases (inevitably) at work both in terms of what data gets input-ed and what bases the mechanics of the formula are built. That we pretend that such prognosis are somehow mathematically innocent seems silly to me. The thing about good prognosis models is that they should seem organic in the macro.
I don't mind that projections are based on formula; but I want that formula to do better than some of the commonly held base assumptions (we already have that). Again, I'd like to see a projection formula that produced nuanced results player by player, and don't simply carry the assumptions of the aggregate to the particulars of the each specific player.
In my mind, there's difference between playing the odds and earnestly prognosticating specific player production. If current statistical modeling doesn't grant much confidence in specific player production (as with Mauer), then why make publicly available such prognosis? Again, what value do they have (as suggested with the population of 1(n=1))?
If the point is that such endeavors are better when the lens is pulled back, that the projections are more useful for projecting team wins than individual production, I think we can agree.
Edited by PseudoSABR, 28 January 2014 - 01:06 AM.