No prob. It's really a different way at looking at a GLM problem
On 26 January 2017 at 05:39, Lee, Laura <laura.lee@xxxxxxxxxx> wrote:
Thanks, Bob. I’ll definitely look into that. It’s an approach I hadn’t
*From:* qfish-bounce@xxxxxxxxxxxxx [mailto:qfish-bounce@xxxxxxxxxxxxx] *On
Behalf Of *Robert O'Boyle
*Sent:* Wednesday, January 25, 2017 9:07 PM
*Subject:* [qfish] Re: correcting for bias in gear characterization study
You can get guidance from Paul Conn's hierarchical analysis of noisy
survey data (Can J FAS 2010 67: 108 -120). While he uses a Bayesian
approach, a state-space approach is also possible. An unobserved abundance
is sampled by a suite of survey gears, which produces a suite of noisy time
series. In your problem, each set of your survey represents a progression
of the unoberved abundance over a a short time period, involving different
times of day. You could have coefficients for each time of day. Each set is
then sampled by gear from each side of the vessel. You could consider the
port and starboard sets as representing separate survey indices with
distinct catchability coefficents and CVs. A state space approach would
allow estimate of the time of day coefficients (and associated CV) and
port/starboard coefficients (and associated CV). Paul's paper gives some
guidance on how this problem can be formulated.
Hope this helps.
On 23 January 2017 at 08:48, Lee, Laura <laura.lee@xxxxxxxxxx> wrote:
This is not a stock assessment question, but I felt the audience of this
listserv would have the answer.
Some background…the study is a gear characterization project looking at
paired hauls from a trawling vessel to compare catches in the experimental
net to those of the control net. We are looking at three different
experimental gears. Ideally, the control and experimental nets are swapped
between the port and starboard sides so that each gear (control or
experimental) tows on each side the same number of times as the other gear.
This was not always the case. For example, in testing of one of the gears,
the control gear was on the port side 20 times and on the starboard side 4
times. If there is a side bias, this unequal towing could magnify
differences in results.
Another analyst I’m working with “solved” this problem by randomly
identifying and eliminating 16 of the tows in which the control gear was on
the port side so the analysis was restricted to tows for which each gear
was on each side 4 times (for the example above). This does not seem
correct to me.
I’ve run randomization tests to test for side bias (and day/night bias) in
the catches of select species. There is a side bias (and sometimes a time
of day bias) in some cases. What is the best way to adjust for this bias so
I can determine whether the difference between the control and experimental
gear catches is due to the gear modification alone.
I hope this makes sense. Thanks so much for your time!
*Laura M. Lee*
*Senior Stock Assessment Scientist*
Division of Marine Fisheries
Department of Environmental Quality
252 808 8094 <(252)%20808-8094> office
252 726 6062 <(252)%20726-6062> fax
3441 Arendell Street
P.O. Box 769
Morehead City, NC 28557-0769
*Email correspondence to and from this address is subject to the*
*North Carolina Public Records Law and may be disclosed to third parties.*