Current research

Research interests

  • Environmental and ecological statistics
  • Modelling high dimensional abundance data
  • Estimation and testing of allometric lines
  • Resampling-based hypothesis testing

Research plan 2007-2009

The overarching aim of David Warton's research is to further develop statistical methods for ecological research. Over the next three years, research will focus primarily on the community-level analysis of multivariate abundance data in ecology - abundances collected simultaneously for a large number of different types of organisms.

He has proposed a model-based approach to the analysis of multivariate abundance data. A model-based approach offers many approaches, some of which will be explored over the next few years. Model selection is an important problem: identifying which type of model is more appropriate for your data, and within a given class of regression model, identifying which variables are important in predicting community-level response. These problems can be approached quite naturally within the modelling framework he suggests, so he will develop automated methods for model selection, and apply them in contexts where specific predictions are of key importance, such as exploring the threat posed by climate change to natural communities.

Ordination is a tool commonly used in multivariate analysis, however the ordination methods commonly used are difficult to interpret as they are not based on an explicit model, and it is difficult to assess whether they are appropriate for a given dataset. I propose applying latent variable models, and exploring the application of canonical biplots to count data. These developments would provide the first model-based tools for ordination of data that specifically account for two key properties: overdispersion of abundance counts, and correlation between variables.

He will also review and apply modern methods of multiple testing, as a means of testing for taxon-specific effects in multivariate abundance data. This will involve applying recent innovations in multiple testing (such as false discovery rates and resampling-based multiple testing approaches) to generalised linear models fitted to multivariate abundance data.

Other planned projects include identifying a protocol for resampling-based hypothesis testing of generalised linear models, evaluating the use of generalised linear mixed models in modelling multivariate abundance data, and developing new, efficient regularisation methods for high-dimensional data analysis.

 

Key research quantity and quality indicators

Data from Scopus, last updated November 19, 2007:

  • Total number of publications: 17.
  • Total number of citations: 235.
  • Most highly cited article: [19], 51 citations.
  • h-index: 8 (i.e. 8 papers have been cited at least 8 times).