This is the suggestion I just gave to a student:
1. check whether the model(s) you need are available either in ns-3 or
in one of the experimental repositories (the ones listed at
http://code.nsnam.org/);
2. inquire (here ;) ) about their level of completeness, especially if
they are experimental;
3. decide whether you are ready to tweak someting, sooner or later.
If (1)-(3) hold, then you should definitively use ns-3, it's a great
tool to work with. Otherwise, unless developing a model is (or can
become) the focus of your work, ns-2 is likely to be a more
conservative choice. Nevertheless, do not underestimate its learning
curve.