Evaluation of estimation quality of a general paradigm for indexing animal abundance when observations are counts

Authors

Benjamin Shulman, University of Colorado Denver Anschutz Medical Campus
Brandie D. Wagner, University of Colorado Denver Anschutz Medical Campus
Gary K. Grunwald, University of Colorado Denver Anschutz Medical Campus
Richard M. Engeman, USDA-APHIS-Wildlife ServicesFollow

Document Type

Article

Date of this Version

2016

Citation

Ecological Informatics 32 (2016) 194–201

Comments

U.S. Government Work

Abstract

Relative abundance indices are widely applied to monitor wildlife populations. A general indexing paradigm was developed for structuring data collection and validly conducting analyses. This approach is applicable for many observation metrics, with observations made at stations through the area of interest and repeated over several days. The variance formula for the general index was derived using a linear mixed model, with statistical tests and confidence intervals constructed assuming Gaussian-distributed observations. However, many observation methods, like intrusions to track plots or camera traps, involve counts with many zeroes, producing Poisson-like observations. To fill this inferential gap between Gaussian analytical assumptions and Poisson-distributed data we evaluated, via a broad Monte Carlo simulation study, variance estimation and confidence interval coverage when Gaussian statistical inference is applied to data generated from a Poisson distribution. The mixed effects linear model assuming Gaussian observations performed well in estimating variances and confidence intervals when simulated Poisson data were in the range found in field studies (88–96% confidence interval coverage). Estimation improved by increasing the number of observation days. Confidence interval coverage rates performed very well (even with few observation days) when day-to-day variability was small, while effective estimation resulted for a great range in station-to-station variability. These results provide a foundational basis for applying the general indexing paradigm to count data, strengthen the generality of the approach, provide valuable information for study design, and should reassure practitioners about the validity of their analytical inferences when using count data.