Abstract
Observation error is a necessary evil when fitting models for population dynamics.
Animal abundances, or densities, are never known exactly but instead are estimated
in some fashion. This leads to observation/measurement error in fitting time series
models. This talk will provide a general non-technical survey of the work on observation
error in this context, both the impacts of ignoring it and methods that try to correct
for it. The latter include some methods where there is no additional information about
the observation error, but some simplifying assumptions are made, and methods that
try to incorporate information about the measurement error, in the form of standard
errors attached to estimated abundances or log-abundances. Attention is given to the
fact that the observation errors often have unequal variances, due to changes in the
population being sampled over time and/or changes in sampling effort.
The survey will touch on three general classes of dynamic models:
i) the random walk model often used in PVA (population viability analysis),
ii) linear autoregressive models, which are often used on the log scale to
model density dependence and delayed density dependence and
iii) general non-linear dynamic models.
John Buonaccorsi
Dept. of Mathematics and Statistics
Univ. of Massachusetts - Amherst