Steven Candy 5 Nov 2009

Thursday 5 November 2009, 11:30 am

Science In The Spotlight

Steven G Candy (SOE): Modelling nonlinear trends in longitudinal data using linear mixed models incorporating cubic smoothing splines: a demonstration using examples on Antarctic krill growth, wandering albatross growth, Antarctic fur seal dive profiles, longline sink profiles, fluorescence depth profiles from Broke-West, and within-season profiles of Adélie penguin foraging trip duration.

The AAD science program collects a diverse range of data many of which involve repeated observations of one or more response variables over time and/or space where these are typically denoted longitudinal, profile, or cross-sectional data in the statistical methods literature. This type of data can represent serious challenges in obtaining accurate inferences, predictions, and quantification of uncertainty. Often complex nonlinear trends combined with multiple levels of variation (e.g. between-animal, within-animal-over-time) and autocorrelation in time and/or space need to be modelled.

Linear mixed models incorporating cubic smoothing splines provide a relatively new and flexible method of modelling this type of data. The advantages and disadvantages of this approach compared to traditional linear and nonlinear modelling will be discussed using examples of some of the datasets collected by the SOE program.