With regards to estimating Ricker model parameters via the
loglinear formulation ln(R/S) = ln(a) - bS: Steve Fleischman was
entirely correct in pointing out the fact that the null
hypothesis to be tested is that R is linearly proportional to S.
However, there is much more to be aware of when estimating
Ricker parameters in this way.
Such estimates are unbiased and accurate *if* the assumption
of independent identically-distributed normal error residuals
is vaild. Indeed, if this is so, the method also gives the
correct error SSQ and confidence intervals for the parameter
estimates. However, the regression SSQ from the loglinear
model in *incorrect* (when compared to the non-linear regression
equivalent) and hence any R^2 goodness-of-fit measure will
be erroneous, as will F-test results. Such diagnostics should
therefore not be used to compare models: rather, the error
SSQs themselves shoud be invoked.
Ideally, the loglinear approach should perhaps be used in
conjunction with the full non-linear estimation, with the
former being used to generate starting estimates for the
latter, which in turn will yield the correct diagnostics that
are required for model comparison.
Some good references on this topic are:-
TC Iles. 1990. A review of stock-recruitment relationships
with reference to flatfish populations. Neth. J. Sea. Res.,
32 (3/4), 399-420. (esp. Appendix 2)
R Hilborn. 1985. Simplified calculation of optimum
spawning stock size from Ricker's stock-recruitment
curve. Can. J. Fish. Aqu. Sci., 42, 1833-1834.
By the way, is the new book from Quinn & Deriso
publicly available yet?
Best wishes,
Coby Needle.
--
Mr Coby L. Needle Email: [log in to unmask]
Population Studies Tel: +44 (0) 1224 295 456
FRS Marine Laboratory Fax: +44 (0) 1224 295 511
PO Box 101 Web: http://www.marlab.ac.uk
Victoria Road
Aberdeen AB11 9DB "We're gonna need a bigger boat!"
Scotland
***************************************************************
<To leave send SIGNOFF FISH-SCI to [log in to unmask]>
<For information send INFO FISH-SCI to [log in to unmask]>
***************************************************************
|
