I think that non-modelers have an idea that modelers know ahead of time what will happen when they put X, Y, and Z into a model. "Everything that comes out is just stuff you put in." This is so wrong. Even super simple social models surprise me all the time.

Because I build simple models, upon observing results I can understand what features caused them. And this means I can write a paper that clearly outlines these causal processes. But they are rarely just what I expected them to be.
2

For example, our student Nathan Fulton just built a model looking at a network where individuals share evidence. We asked: what if these individuals exhibit confirmation bias? I assumed the whole group would learn worse. Wrong. They learn better.
3

The bias slows down the learning of the group and creates transient diversity a la the work of @KevinZollman. And the point is this sort of thing happens all the time. You have to do the analysis to actually get the results.

4