USGS Science⌕ Search
Home/Archive/USGS

USGS · 5224819

Analysis of multinomial models with unknown index using data augmentation

J. Andrew Royle·R.M. Dorazio·W.A. Link
View original source ↗Read source document ↗Export metadata ↓

Abstract

Multinomial models with unknown index ('sample size') arise in many practical settings. In practice, Bayesian analysis of such models has proved difficult because the dimension of the parameter space is not fixed, being in some cases a function of the unknown index. We describe a data augmentation approach to the analysis of this class of models that provides for a generic and efficient Bayesian implementation. Under this approach, the data are augmented with all-zero detection histories. The resulting augmented dataset is modeled as a zero-inflated version of the complete-data model where an estimable zero-inflation parameter takes the place of the unknown multinomial index. Interestingly, data augmentation can be justified as being equivalent to imposing a discrete uniform prior on the multinomial index. We provide three examples involving estimating the size of an animal population, estimating the number of diabetes cases in a population using the Rasch model, and the motivating example of estimating the number of species in an animal community with latent probabilities of species occurrence and detection.

Explore related subjects

Journal of Computational and Graphical StatisticsArticle

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

J. Andrew Royle, R.M. Dorazio, W.A. Link. 2007. Analysis of multinomial models with unknown index using data augmentation. https://pubs.usgs.gov/publication/5224819

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2020

A random forest approach for bounded outcome variables

Random forests have become an established tool for classication and regres- sion, in particular in high-dimensional settings and in the presence of non-additive predictor-response relationships. For bounded outcome variables restricted to the unit interval, however, classical modeling approaches based on mean squared error loss may severely suer as they do not account for heteroscedasticity in the data. To address this issue, we propose a random forest approach for relating a beta dis- tributed outcome to a set of explanatory variables. Our approach explicitly makes use of the likelihood function of the beta distribution for the selection of splits dur- ing the tree-building procedure. In each iteration of the tree-building algorithm it chooses one explanatory variable in combination with a split point that maximizes the log-likelihood function of the beta distribution with the parameter estimates de- rived from the nodes of the currently built tree. Results of several simulation studies and an application using data from the U.S.A. National Lakes Assessment Survey demonstrate the properties and usefulness of the method, in particular when com- pared to random forest approaches based on mean squared error loss and parametric regression models.

Journal of Computational and Graphical Statistics↗