Dana L. Thomas

Study Designs and Tests for Comparing Resource Use and Availability II

Abstract We review 87 articles published in the Journal of Wildlife Management from 2000 to 2004 to assess the current state of practice in the design and analysis of resource selection studies. Articles were classified into 4 study designs. In design 1, data are collected at the population level because individual animals are not identified. Individual animal selection may be assessed in designs 2 and 3. In design 2, use by each animal is recorded, but availability (or nonuse) is measured only at the population level. Use and availability (or unused) are measured for each animal in design 3. In design 4, resource use is measured multiple times for each animal, and availability (or nonuse) is measured for each use location. Thus, use and availability measures are paired for each use in design 4. The 4 study designs were used about equally in the articles reviewed. The most commonly used statistical analyses were logistic regression (40%) and compositional analysis (25%). We illustrate 4 problem areas in resource selection analyses: pooling of relocation data across animals with differing numbers of relocations, analyzing paired data as though they were independent, tests that do not control experiment wise error rates, and modeling observations as if they were independent when temporal or spatial correlations occurs in the data. Statistical models that allow for variation in individual animal selection rather than pooling are recommended to improve error estimation in population-level selection. Some researchers did not select appropriate statistical analyses for paired data, or their analyses were not well described. Researchers using one-resource-at-a-time procedures often did not control the experiment wise error rate, so simultaneous inference procedures and multivariate assessments of selection are suggested. The time interval between animal relocations was often relatively short, but existing analyses for temporally or spatially correlated data were not used. For studies that used logistic regression, we identified the data type employed: single sample, case control (used–unused), use–availability, or paired use–availability. It was not always clear whether studies intended to compare use to nonuse or use to availability. Despite the popularity of compositional analysis, we do not recommend it for multiple relocation data when use of one or more resources is low. We illustrate that resource selection models are part of a broader collection of statistical models called weighted distributions and recommend some promising areas for future development.

Feeding ecology of phocid seals and some walrus in the Alaskan and Canadian Arctic as determined by stomach contents and stable isotope analysis

Feeding habits of ringed (Phoca hispida), bearded (Erignathus barbatus), spotted (Phoca largha) and ribbon (Phoca fasciata) seals and walrus (Odobenus rosmarus) were studied using stomach contents and stable carbon and nitrogen isotopes. Bearded seals fed benthically, primarily crustaceans and mollusks. Both zooplankton and fish were significant prey for ringed seals, while fish was principal spotted seal prey. Few gastric contents were available from ribbon seals. δ15N was positively correlated with age in ribbon seals and δ13C was positively correlated with age in ringed and ribbon seals. δ15N was highest in spotted seals, in agreement with their fish-dominated diet. δ15N was not different between Alaskan-harvested ringed and bearded seals, while δ15N was lowest in ribbon seals and walrus. Carbon-13 was most enriched in bearded seals and walrus reflecting benthic ecosystem use. Canadian ringed seals were depleted in 13C compared to Alaskan pinnipeds, likely because of Beaufort Sea versus Chukchi and Bering seas influence.

Estimating prefledging survival: allowing for brood mixing and dependence among brood mates

Estimates of juvenile survival from hatch to fledging provide important information on waterfowl productivity. We develop a model for estimating survival of young waterfowl from hatch to fledging. Our model enables interchange of individuals among broods and relaxes the assumption that individuals within broods have independent survival probabilities. The model requires repeated observations of individually identifiable adults and their offspring that are not individually identifiable. A modified Kaplan-Meier procedure (Pollock et al. 1989a,b) and a modified Mayfield procedure (Mayfield 1961, 1975; Johnson 1979) can be used under this general modeling framework, and survival rates and corresponding variances of the point estimators can be determined.

Survey and Comparison of Methods for Study of Resource Selection

Wildlife management studies often compare relative use and availability of resources (e.g., habitats). When resources. are used disproportionately to availability, use is said to be selective. Designs and analyses for resource selection studies are reviewed and compared with respect to the type of data collected, underlying assumptions, weighting of observations, distributional requirements and usefulness in comparing selection among subgroups or time periods. Common misuses of analyses are noted. Practical problems in studying resource selection, such as which resources to consider, the choice of study area, and spatial and temporal dependencies, are discussed and suggestions for future development are given.

A Bayesian Random Effects Discrete-Choice Model for Resource Selection: Population-Level Selection Inference

Abstract Modeling the probability of use of land units characterized by discrete and continuous measures, we present a Bayesian random-effects model to assess resource selection. This model provides simultaneous estimation of both individual- and population-level selection. Deviance information criterion (DIC), a Bayesian alternative to AIC that is sample-size specific, is used for model selection. Aerial radiolocation data from 76 adult female caribou (Rangifer tarandus) and calf pairs during 1 year on an Arctic coastal plain calving ground were used to illustrate models and assess population-level selection of landscape attributes, as well as individual heterogeneity of selection. Landscape attributes included elevation, NDVI (a measure of forage greenness), and land cover-type classification. Results from the first of a 2-stage model-selection procedure indicated that there is substantial heterogeneity among cow–calf pairs with respect to selection of the landscape attributes. In the second stage, selection of models with heterogeneity included indicated that at the population-level, NDVI and land cover class were significant attributes for selection of different landscapes by pairs on the calving ground. Population-level selection coefficients indicate that the pairs generally select landscapes with higher levels of NDVI, but the relationship is quadratic. The highest rate of selection occurs at values of NDVI less than the maximum observed. Results for land cover-class selections coefficients indicate that wet sedge, moist sedge, herbaceous tussock tundra, and shrub tussock tundra are selected at approximately the same rate, while alpine and sparsely vegetated landscapes are selected at a lower rate. Furthermore, the variability in selection by individual caribou for moist sedge and sparsely vegetated landscapes is large relative to the variability in selection of other land cover types. The example analysis illustrates that, while sometimes computationally intense, a Bayesian hierarchical discrete-choice model for resource selection can provide managers with 2 components of population-level inference: average population selection and variability of selection. Both components are necessary to make sound management decisions based on animal selection.

Confidence Bands for Percentiles in the Linear Regression Model

Abstract Simultaneous confidence intervals for percentiles of the normal regression model similar to those given by Steinhorst and Bowden (1971) are considered. Kanofskys (1968) confidence band for a single normal distribution is modified and extended to the regression model. The confidence bands that are simultaneous in all percentiles provide corresponding confidence bands for the cumulative conditional normal distribution functions. The various procedures are compared with respect to bandwidth.

A Bayesian multinomial model for analyzing categorical habitat selection data

Modeling the number of uses of discrete habitat types by animals with a multinomial distribution, we illustrate the use of Bayesian methods to estimate selection. An advantage of this approach in assessing selection is the construction of credibility intervals that do not rely on large sample normal theory. In addition, credibility intervals for ranked selection of habitats are easily obtained. Bayes factors and Bayesian p values (posterior predictive values) are used to test the hypothesis of selection for each animal, test selection across all animals and for multiple comparisons among habitats. We compare our method to alternative methods for a real dataset. Freely available WinBUGS software is used to fit the model and test hypotheses.

Introduction to resource selection studies

In this chapter we provide motivation for the study of resource selection, define terms, discuss study designs and sampling, and give an historical perspective on the statistical evaluation of resource selection.

A Unified Theory for the Study of Resource Selection (Availability and Use) by Wildlife Populations

A unified theory of analysis is developed for comparing resource use and availability for wildlife populations based on the concept of resource selection functions. Each resource unit is characterized by values for p variables x = (X 1 ,X 2 ,…,X P ), and a resource selection probability function is defined to be the probability of a unit with X = x being used by time t. We propose that the function that gives the probability of a resource unit with xnot being used by time tshould be approximated by the proportional hazards model, which is widely used for survival distributions in other contexts. On this basis, we discuss how the resource selection probability function can be estimated from censuses or samples of available, used, or unused resource units taken after either one or several units of selection time.

Resource Selection by Animals: Statistical Design and Analysis for Field Studies

Introduction to resource selection studies. Examples of the use of resource selectory studies. Examples of the use of resource selection functions. Statistical modelling procedures. Studies with resources defined by several categories. Estimating a resource selection probability function from a census of resource units using logistic regression. Estimating a resource selection probability function from a census of resource units at several points in time using the proportional hazards model. Estimating a resource selection function from samples of resource units using proportional hazards and log-linear models. Estimating a resource selection function from two samples of resource units using logistic regression and discriminant function methods. General log-linear modelling. Analysis of the amount of use. The comparison of selection for different types of resource unit. References. Index.