Recently the independent multinomial selections model (IMS) with the
multinomial logit link has been suggested as an analysis tool for radio-telemetry
habitat selection data. This model assumes independence between animals,
independence between sightings within an animal, and identical multinomial habitat
selection probabilities for all animals.
We propose two generalizations to the IMS model. The first generalization is to
allow a Markov chain dependence between consecutive sightings of the same animal.
This generalization allows for both positive correlation (individuals persisting in the
same habitat class in which they were previously sighted) and negative correlation
(individual vacating the habitat class in which they were previously sighted). The
second generalization is to allow for heterogeneity. Here, a hierarchical Dirichlet-multinomial
distribution is used to allow for variability in selection probabilities
between animals. This generalization accounts for over-dispersion of selection
probabilities and allows for inference to the population of animals, assuming that the
animals studied constitute a random sample from that population..
Both generalizations are one parameter extensions to the multinomial logit
model and allow for testing the assumptions of identical multinomial selection
probabilities and independence. These tests are performed using the score, Wald, and
asymptotic likelihood ratio statistics. Estimates of model parameters are obtained
using maximum likelihood techniques, and habitat characteristics are tested using
drop-in-deviance statistics.
Using example data, we show that persistence and heterogeneity exist in habitat
selection data and illustrate the difference in analysis results between the IMS model
and the persistence and heterogeneity models. Through simulation, we show that
analyzing persistence data assuming independence between sightings within an animal
gives liberal tests of significance for habitat characteristics when the data are
generated with positive correlation and conservative tests of significance when the
data are generated with negative correlation. Similarly, we show that analyzing
heterogeneous data, assuming identical multinomial selection probabilities, gives
liberal tests of significance for habitat characteristics. / Graduation date: 2001
| Identifer | oai:union.ndltd.org:ORGSU/oai:ir.library.oregonstate.edu:1957/32513 |
| Date | 16 May 2000 |
| Creators | Usner, Dale Wesley |
| Contributors | Ramsey, Fred L. |
| Source Sets | Oregon State University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
Page generated in 0.0022 seconds