Previous: All models are wrong (2021-08-18)
Although we know that "all models are wrong, but some are useful", that only handles the question of why a model can't do everything right. Right? A hurricane model might be able to get windspeed, precipitation, future track in 24 hours, future track in 2 weeks—but everything? No. We want the models to step right up, and give us everything, but each model is tuned to do some things well because each individual problem to solve in a model in the whole damned mess of problems is complicated enough on its own.
So "all models are wrong" is for the naysayers. It could just as easily be used for the True Believers—the ones who think there is such thing as a complete model of a problem if you just reach for it. A model isn't supposed to be right—"right" as in 100% correct about everything. But useful analysis often gets stopped in the quixotic search for a model that does it all.
Actually, maybe it's not a quixotic search, but a search for a holy grail. I always think of this exchange from Monty Python and the Holy Grail whenever someone gets wrapped up too tight in overtuning a model.
Camelot!
Camelot!
Camelot!
It's only a model.
Shhh!