The web reference for the item is here
The summary here is done from memory and omits some extra tests not
relevant to the point at issue which Chuck raised.
What I set out to do was first to look at 35 samples of 1000 words
from all three Synoptics. These were classified by genre: narrative
vs. mixed genre (i.e.parables, apophthegms) vs. sayings, as the greatest
stylistic differences tend to correlate with genre differences rather
than with author or source differences. The samples were also classified
by author of the text, and by the source to which 2ST attributes it.
What I wished to see was whether or not the stats _assigned_ the
samples to the source to which the theory _attributed_ them.
The stats involved counts of words or genitive endings of fairly
high frequency. The first test showed the three M samples higher
on the plot than the Q samples, though the greater difference was
that narratives went to the left and sayings to the right, with
mixed genre stuff in between.
I then used a different method which used smaller samples attributed
to M and Q, using only 5 of the most frequent words. (The latter was
needed, as for small samples you need very high frequency items; only
if you have have huge samples from novels can you have the luxury of
testing rare words, let alone whatever might get defined as a "hapax".)
GLM tested first for a genre effect and found it was significant, then
for a source effect allowing for genre and found that significant also.
I then used two further methods on the sayings samples attributed to
Q and M. Discriminant Analysis on samples on this size usually reaches
around 80% accuracy at best, and did _assign_ around 80% to the source
to which the theory attributes them, with "error rate" around 20%.
Cluster Analysis is "blind" to attribution and clustered most of the
Q samples together and most of the M samples together again with about
80% match to what the theory "expects".
So this was designed as a "crucial experiment". If the FGT is right we
would not expect there to be fairly consistent stylistic differences between
M and Q samples when genre is allowed for. If 2ST is right we would
expect such differences. My conclusion is that the results are sufficient
to infer a cautious preference for 2ST _on the evidence provided_.
An important proviso is that with all choices between complex over-arching
theories one has to plod through, and take account of, very large amounts
of evidence of very different kinds before reaching a final rather than
a cautious interim conclusion.
So in brief my answer to Chuck's query is that Q samples and M samples
do seem to differ from each other especially in the sayings. Some of
which were not assigned to the expected source were from passages often
debated as of uncertain attribution e.g. the woes against the scribes.
There didn't seem to be very many evident differences _within_ the M
samples, except, of course, for the very major differences between M
narratives and M sayings.
David Mealand, University of Edinburgh
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.