Not only is this wrong about physics/astronomy, I regularly use the arxiv version because the typography is better (e.g. in the published paper an equation is split with part of the equation being at the bottom of one column, and the top of the next, whereas the equation is on one line in the arxiv version).
