Fit model to only unfished things to demonstrate performance drop-off (I hope)

[DanOvando]

The idea here is simple. Redo, but only fitting to unfished critters. You'd think there'd still be a signal (e.g. habitat clustering), but it should be a hell of a lot worse than the fished species