Thanks for the numbers! I have to say, though, the response is part of what frustrates me about analytics. let me be clear that I don't mean this personally, more of just a general comment on analytics. Here 2 numbers were presented and the the one that better supported your thesis was chosen, in lieu of the (more accurate) one that goes against your priors. The bigger issue though is the idea that FIP is more meaningful than ERA when looking at last year. This is fundamentally misunderstanding statistics. FIP is a predicitve analytic. It MAY predict NEXT YEAR's ERA better than this current year's ERA, but FIP measures what may have happened if baseball was played in a vacuum while ERA measures what actually happened. FIP cannot predict last year's ERA better than ERA because we actually have last year's ERA - no predictions necessary. So the numbers presented absolutely point to the Twins replacing a better pitcher (Gray) with worse ones, on average (the bullpen). Maybe this doesn't conform to the analytics bible, but analytics is supposed to be about digging deeper to find the cold truth, and not being slavish to conventional wisdom.
That said, I am most interested in what happened specifically in the games that Gray pitched, and even more specifically with the pichetrs that replaced him. One thing I've noticed is the stats crowd always focuses on when to pull the starter, and almost never on who they plan to replace him with. Yes, third time through the order may result in a higher ERA for the starter, but it still might be better than the guy they are bringing in, and it is almost certainly better than the guy they are bringing it + the domino effect of wearing out your bullpen earlier. And this is magnified when you have a starter who is dealing. You KNOW how that guy is pitching, you don't need numbers to predict if he's pitching well. But when you yank him you are taking a risk that the next guy maybe just doesn't have it today, or the next guy, or the next guy. It just seems counterproductive to me. But, again, maybe I'm happy to be proven wrong with the data. I'll see if I can gather any numbers on this.