Estimating Spatial and Temporal Components of Variation for Fisheries Count Data Using Negative Binomial Mixed Models. Issue 1 (12th December 2012)

• Record Type: Journal Article
• Authors: Irwin, Brian J.; Wagner, Tyler; Bence, James R.; Kepler, Megan V.; Liu, Weihai; Hayes, Daniel B.
• Is Part Of: Transactions of the American Fisheries Society. Volume 142:Issue 1(2013)
• Page Start: 171
• Page End: 183
• Publication Date: 2012-12-12
• Subjects: Fish culture -- Periodicals; Fishery management -- Periodicals; Pisciculture -- Périodiques; 639.2
• DOI: 10.1080/00028487.2012.728163
• Languages: English
• ISSNs: 0002-8487

Abstract

Abstract: Partitioning total variability into its component temporal and spatial sources is a powerful way to better understand time series and elucidate trends. The data available for such analyses of fish and other populations are usually nonnegative integer counts of the number of organisms, often dominated by many low values with few observations of relatively high abundance. These characteristics are not well approximated by the Gaussian distribution. We present a detailed description of a negative binomial mixed‐model framework that can be used to model count data and quantify temporal and spatial variability. We applied these models to data from four fishery‐independent surveys of Walleyes Sander vitreus across the Great Lakes basin. Specifically, we fitted models to gill‐net catches from Wisconsin waters of Lake Superior; Oneida Lake, New York; Saginaw Bay in Lake Huron, Michigan; and Ohio waters of Lake Erie. These long‐term monitoring surveys varied in overall sampling intensity, the total catch of Walleyes, and the proportion of zero catches. Parameter estimation included the negative binomial scaling parameter, and we quantified the random effects as the variations among gill‐net sampling sites, the variations among sampled years, and site × year interactions. This framework (i.e., the application of a mixed model appropriate for count data in a variance‐partitioning context) represents a flexible approach that has implications for monitoring programs (e.g., trend detection) and for examining the potential of individual variance components to serve as response metrics to large‐scale anthropogenic perturbations or ecological changes.