Hi:

I looked at the Hayes study.  Sincere effort, but the skunk predation and
queen loss preclude drawing conclusions.  The bottom entrance hives with
queen excluders were either killed or impacted by skunks. Only two
survived, and I suspect that their low production (30 and 41 lbs) was as
likely to be a result of the skunks in the yard.  If one colony was killed,
two heavily preyed upon, I doubt that the skunks left the remaining two
alone.  Regardless, none of the other yards had this severe problem.

If you drop the queenless colony from comparison of the top
entrance/excluder hives versus the controls, the average yield is only a
couple of pounds different, certainly not statistically significant.

Finally, we too have noticed plugging up of the brood chamber when queen
excluders were put on bottom entrance hives - with one important
difference.  During the day when a heavy flow is on, the field bees dump
nectar into open cells in the brood nest.  Over night, they move this
nectar upstairs.  We have lots of digitized data of brood/honey/pollen
areas in commercial colonies -- the honey bound brood nest observation
depends on time of day that the colony is opened and inspected.

Finally, although I don't think one can draw any real conclusions from this
study - too few hives and lots of skunk problems at the site with bottom
entrances/excluders, I know from our data that excluders cause lots of
changes in a colony, at times causing the queen to suspend laying, and in
some cases, balling of the queen and supercedure.  I'm not a fan of
excluders and we could use some good studies.  This one made an honest
effort, but too many confounding variables (skunks, queenless, etc.).

Expect 25-50% coefficient of variations (or relative standard deviation) on
honey production among hives at a given site - based on years of our studies.

To get the C.V. or RSD, divide the mean or average by the standard
deviation, multiply by 100.  I recommend no fewer than 12 colonies per site
for a study like this - and would toss in a couple of spares.  Elevate the
hives on sawhorse stands to eliminate the skunk problem.  Toss out colonies
that go queenless (AND MARK YOUR QUEENS AT THE START OF THE TEST).
Hopefully, you will have 10-12 viable hives for comparisons.  Oh, also, be
sure to pull all of the hives together before the trial, rank them by
population strength and brood nest size, then randomize and assign hives to
each site.  The objective is to have the same mix of weak, good, and strong
hives at each location.  Anything less means that you can't make any
comparisons.  I don't advocate trying to equalize hives, it never really
works.

Other than that, this test is one that any beekeeper with sufficient hives
can do.  Oh yes, you really do need at least two sites for controls, two
for bottom excluders, two for top excluders.  Without replication of sites,
you have no idea as to whether the effect is due to the excluders or to
some other variable such as blossom availability (or as happened in the
cited study, skunks).

I offer these as suggestions for improvement, not as a criticism of what
was tried.  One of those 20/20 eyesight, when looking back.  Unfortunately
for the authors, Murhphy's Law prevailed - but the results are suggestive
and they clearly show where the study can be changed to improve the design.

Cheers

Jerry