Scale and Analysis of Spatially Confounded Ecological Data

• Activity Number: 646
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Trevor Hefley* and Mevin Hooten and Ephraim Hanks and Daniel Walsh and Robin Russell
• Companies: Colorado State University and Colorado State University and Penn State and U.S. Geological Survey and U.S. Geological Survey
• Keywords: spatial confoundin ; regularization ; spatial risk factor ; wildlife disease

Abstract

Ecological data indexed by space and/or time are ubiquitous and statistical methods used to capture the spatial and temporal components of the processes are developing rapidly. Efforts to describe increasingly complex processes to gain deeper insights are commendable, but may also be motivated by constant reminders of the dangers of ignoring spatial and temporal correlation. Although the consequences of failing to properly account for latent autocorrelation are well described in the literature, the consequences of remedial measures are not. Similarly, spatial confounding has been recently recognized as an unfortunate byproduct of many spatially explicit statistical models, but there is no consensus on a general remedy yet. We explore the issue of confounding as it relates to the scale of the ecological process and observations. We then describe an approach to address spatial confounding based on regularization of both fixed and random effects and compare it to traditional methods and those explicitly designed to alleviate spatial confounding. To demonstrate these methods, we use the regularization approach to improve inference for spatial risk factors in wildlife disease models.