Dear Katherine, I would like to add a futher point to this discussion: If you do model residues with poor or no electron density, then you should be sure of what you are doing and not just make more or less unguided guesses. This seems obvious but unfortunately it isn't, at least not to everybody. For example, I recently found a PDB entry of a glycoprotein that had an a-L-IdopNAc residue attached to an Asn side chain. Usually the first monosaccharide linked to an Asn is b-D-GlcpNAc -- there are only very few exceptions to that. Therefore, I had a look at the electron density and I saw that only part of the ring was resolved there; around the C5 atom of the sugar there was no electron density. b-D-GlcpNAc and a-L-IdopNAc only differ in the stereochemistry of the C5 atom, so it seems to me that the crystallographers arbitrarily placed the C6 and O6 somewhere and picked a wrong position, which led to wrong stereochemistry of the C5 atom. (This is in principle the same as adding an Ala CB atom somewhere near the CA atom and picking a position resulting in D-Ala instead of L-Ala.) In the example above it would have been better to leave out the unresolved atoms rather than creating an obviously wrong model, I think.
The problem of non-crystallographers non knowing about occupancy or B-factors might be solved (in part) by better pointing out "problematic" parts of the structures, e.g. by highlighting them in visualization tools by default and not only when the user sets color mode to "temperature" or something like that. Cheers, Thomas