Beyond ANOVA: Design and analysis of multi-stressor experiments

Chair: Alistair Hobday

Peter W Dillingham (1)*, Philip W. Boyd (2), Catriona L. Hurd (2), Christopher E. Cornwall (3), Christina M. McGraw (1)

1 School of Science and Technology, University of New England, Armidale, New South Wales 2351, Australia
2 Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, Tasmania 7005, Australia
3 School of Earth and Environment & ARC Centre of Excellence for Coral Reef Studies, University of Western Australia, Perth, Western Australia 6009, Australia

Most laboratory-based ocean acidification experiments manipulate only a small subset of environmental properties predicted to change by 2100. In order to quantify cumulative effects of multiple stressors, experimental designs that move beyond two- or three-factor studies are required. Here, we describe a collapsed factorial experimental design for the manipulation of multiple stressors which is both tractable and efficient. Building on single-factor studies that determine a likely dominant physiological control, the collapsed factorial design groups the remaining variables into a single, collapsed factor. This design balances the desire to quantify the effect of individual factors with the need to make accurate predictions of physiological responses in a complex and changing ocean. This design was recently implemented to study growth rate and other physiological responses of the Southern Ocean diatom Pseudonitzschia multiseries (Boyd et al. 2015, Nature Climate Change, doi:10.1038/NCLIMATE2811). Manipulating five oceanic properties (temperature/CO2/nutrients/iron/light), this design was more efficient than either full or fractional factorial designs that it replaces. Combining the new experimental design with analysis techniques more sophisticated than classical ANOVAs (e.g. model averaging), we gained further insight and efficiency when making predictions based on laboratory studies.