As a result of what I discovered in the course of my PhD research
(http://purl.org/tfinney/PhD), I have a few (loud-mouthed) opinions
about electronic critical editions.

I don't know about other fields, but in classical New Testament textual
criticism a critical text is constructed by weighing variants at every
point of variation. A number of principles are used to discern the
"best" reading. These principles are divided into internal (e.g. prefer
the shorter reading; prefer that which at first seems wrong but on
closer examination has native dignity) and external (e.g. prefer the
reading of a particular MS).

Now to my ideas. After looking at the results of a statistical analysis
of purely external evidence (i.e. multivariate analysis (MVA) of (1)
substantive and (2) orthographic variations among ~30 MSS in Greek of a
New Testament epistle called Hebrews), I am inclined to think that there
is grouping among the MSS. I think that I see three groups. I associate
these with three broad regions of the early (200-1000 CE) Greek-speaking
Christian world: Egypt, Palestine/Syria, Asia Minor. Your interpretation
of the results may differ. (For nearly all of the MSS, the grouping
found for substantive variations is the same as found for orthographic
variations. This is an unexpected result, and, I think, significant. It
indicates, to me at least, the existence of local texts.) 

So what does this have to do with an edition? In my opinion, the
external evidence can be used to extract a primitive text using the
following approach for substantive variations:

(1) isolate MS groups using MVA
(2) isolate within-group readings by taking the majority reading within
the group at each point of variation (if such exists)
(3) isolate between-group readings by taking the majority reading across
groups (if such exists).

E.g. 

group 1 majority text: The cat sat on the rat.
group 2 majority text: The cat spat on the mat.
group 3 majority text: The cat pounced on the mat.

Result: The cat (sat|spat|pounced) on the mat.

There is no between-group majority reading for the third word, so all
one can do is declare that the technique produces no winner. There is a
winner for the last word: mat beats rat. In general, readings can be
rated thus: (A) all groups agree, (B) a majority of groups agree (e.g.
two groups vs one), (C) there is a tie (e.g. two groups vs two groups),
(D) there is no majority between-group reading (e.g. two groups vs one
group vs another group vs yet another group.)

This can only work for MS traditions with many extant copies of the same
work. (Don't bother with it unless there are enough witnesses/MSS to
produce statistically significant results.) It is not always reliable
because the popular reading is not guaranteed to be the primitive
reading. Also, perhaps fatally, the number of groups is not always
clear. Nevertheless, this seems to me a useful approach when considering
purely external evidence--the kind of evidence that emerges from
computer-assisted capture and analysis of a text. (The approach works
best if you consider copyists to be isolated automatons rather than
omniscient editors. Try downloading, compiling and running my copying
simulation at http://purl.org/TC/downloads/simulation to see why.)

Is this a critical text? No--there is no critical thought involved. Is
it reliable? As a scientist, I am more inclined to believe an approach
based on evidence than one based on beliefs about what scribes are
inclined to do. In the end, it comes down to estimating the reliability
of each principle. Unfortunately, many of the internal principles are
untestable because one needs to know the primitive text (or texts)
before one can determine the probability that application of a principle
will yield the primitive reading (or readings).

Warning: My results and conclusions have been met by a deafening
silence. The reason may be that a purely external approach is considered
heretical, especially due to the influence of Westcott and Hort's
introduction to their 1880s critical text. They say "MSS should be
weighed, not counted." My response is, Westcott and Hort did not perform
MVA on collation results. 

Best

Tim Finney