The Problem Is That Your Test Statistics Are Inflated

If your model converges and your model is admissible id so it's proper in the way that we would think about. If those asemtotic properties of maxim likelihood are not coming on line, what typically happens is your test statistics are inflated. And they're all the problems we've had with that. Get those point estimates, get those standard airs, get those test statistics, and then adjust them for what we know to be the deleterious effects of small en yes, i really like the idea.