I get what I expect in the first and third histogram commands, but not
the second. The bars are stretched widthwise in the second histogram,
such that I don't think the density adds to 1. In some twoway commands,
changing the xscale to log when you have estimated a model on x and not
log(x) is an incorrect procedure. Is this another instance? If so, I
think it should be disallowed. Easy graph dialogs should be easy.