Modeling Categorized and Aggregated Dive Depth Data from Marine Mammals

• Activity Number: 161
• Type: Topic Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Megan Higgs*+ and Jay Ver Hoef
• Companies: Montana State University and NOAA
• Keywords: ordinal ; Bayesian ; hierarchical ; temporal

Abstract

Ordinal variables are pervasive in ecological and environmental data, often arising from practical constraints imposed at data collection. Variables hypothetically measured on a continuous scale are instead recorded as ordered categories. Ironically, this simplification complicates a statistical analysis. Models based on cumulative probabilities are developing sophistication to account for temporal, spatial, and/or cluster dependence, conceptually connecting the ordinal response to a latent continuous random variable through a vector of cut-points used to clip the continuous distribution. Dependence is induced through correlation among the continuous variables. We provide extensions for data acquired from Time-Depth Recorders affixed to Harbor seals by the NOAA National Marine Mammal Lab, where dive depth is recorded as a depth category, and data are aggregated over 6-hr intervals.