"Treating discrete variables as if they were continuous"

I confess that I see statisticians as the police of econometrics. If we start behaving badly, they will intervene and control the riot.

However, Andrew "Multilevel" Gelman surprised me. He wrote that there is no problem in principle with running regressions where discrete variables are treated as continuous. I thought that this was a crime, committed only by the most incautious members of the econ brotherhood.