cladistics and outgroups

[Phillip Bigelow <n8010095@henson.cc.wwu.edu>]

Apologies in advance if this subject is getting too far off of the subject
of dinosaurs, but since there was a recent discussion of archosaurian cladistics
here, I have to ask this.
  When doing a cladistic analysis (sort of a statistically most probable 
family tree), the group of animals being analyzed is always compared to 
another group of animals, called the "outgroup".  "Outgroups" are, in a
rough definition, "other" closely related groups of animals.  Ideally,
a perfect outgroup would be the ancestors of the group being studied.
 The purpose of using outgroups is to determine the polarity of change of
character traits.  Therefore, in an ideal world, the perfect outgroup
should have ALL of its traits primitive, relative to the group being
studied.
  In the real world, it is impossible to be certain that an outgroup that
is picked for cladistic analysis is actually the ancestor, or whether it's
traits are all TRUELY primitive.

Questions:
Are there any "rules of thumb" for choosing an appropriate outgroup?
 
Also, lacking any known primitive fossil ancestors, is it possible to
get away with using primitive "sister groups" as outgroups?
(Actually, a better way to ask the question is, why are sister groups
often placed in the data matrix of some cladistic studies?).
 
And lastly, are there any caveats in the process of choosing an
outgroup (such as making a bad choice that nullifies your entire cladogram)?