On Thu, 24 Dec 2015 09:04:52 -0600, dpb wrote:

On 12/24/2015 8:31 AM, Jack wrote:
...

This list in my mind is bogus, and if I were doing the testing, and
somehow a butt joint or miter joint came ahead of a domino, I'd keep it
to myself, and try to find out what I did wrong in the testing. Just my
opinion, but giant red flags here, making the whole thing suspect.

The overall ranking is of little consequence, granted, because there's
so much disparity between the joints as far as the specifics of them are
concerned (as I've noted several times previously ).

Also, as far a postulating, that's well and good, but the results from a
series of tests such as this are valid _only_ for the specific joints
down to the specific sizes of the various mating pieces; one canNOT
infer anything more than that regarding general conclusions.

To do the latter would require having a series of tests of each type in
which the single variable under study _only_ is changed (say width of
the tenon in the simple M&T for one) and then _only_ tenon length. The
problem when one attempts to undertake this kind of study then becomes
one that the number of tests required explodes geometrically and rapidly
turns in to the thousands or 10s of thousands. That's where one would
then need to turn to statistical design of experiments theory to develop
a test matrix that would allow for at least some of the variables to be
studied without confounding effects(*) with a reasonable number of tests.

But, for the particular set of joints incorporated, the simple butt did
_not_ beat the Domino and there's an explanation for why the simple
miter does as outlined above.

I don't think there's anything wrong with the tests themselves such as
they were; it's simply trying to draw too many conclusions from the
results that the amount of testing doesn't support.

(*) Confounding -- when an uncontrolled or unknown variable other than
the one under test has an effect on the result of the device under test,
the result of the test cannot be shown to actually have measured the
desired effect of the intended variable. The example easily seen here
is that between the M&T, F-M&T and the Domino the sizes of the tenons
aren't controlled; only the type used. Hence, one had _no_ controls in
place for the confounding variable and there's nothing that can be said
specifically about the effectiveness of the joining _system_ at all;
only that for the three specific cases with the specific dimensions that
the results were in the order observed.

After nearly 40 yr of consulting in the area for which I coined the term
"statistical engineering", being called in after the fact to try to make
something of results from such tests as the above was all too common
what a client was wanting. Unfortunately, in almost every instance, it
was too late in the process to salvage the work done to date other than
to try to complete an actual design for a series of experiments in which
the tests run could be used to fill in the necessary test matrix.

Brings to mind large laminated beam, what if one of the variables was
mid point in the beam?

One of the questions nagging me is, the amount of pressure used on a
glued joint. Is it really possible to squeeze out too much glue and
render the project worthless a few years down the road?