Perry de Valpine's research interests

My research interests lie in the intersection of population ecology, mathematical modeling, and statistics.  General topics are listed below.

Statistical methods for fitting population models to noisy time-series data

Population dynamics data typically involve inaccurate estimates of the state of a system undergoing stochastic dynamics.  Incorporating both measurement (or observation, or sampling) error and process noise (environmental and/or demographic stochasticity) as well as uncertainty in model structure has been a major challenge for statistical methodology.  I am interested in methods that fall under the nearly synonymous labels of state-space models, hierarchical models, mixed models, or hidden process models.

Applications of population model-fitting

I am interested in putting new model-fitting methods to work on real problems and in seeing those problems inspire new methods.  I have worked on fisheries and insect population dynamics.  Fisheries are particularly interesting because they represent the most common setting where population models and statistics play a perennial role in prediction and management.  Insect population dynamics are particularly interesting because they have important applications in agriculture and are often amenable to independent experimental testing.

Theoretical population dynamics

Drawing conclusions about population dynamics by fitting biologically sensible models to data is closely linked to understanding theoretical dynamics of the models themselves.  I am interested in age- and stage-structured models with realistic organismal development, species interactions, space, and stochasticity.

Computational Statistics

Methods for likelihood (and related) calculations with state-space or other hierarchical models require numerical implementation or approximation for all but the simplest cases.  I am interested in quadrature and Monte Carlo methods in general, including Markov chain Monte Carlo and sequential Monte Carlo.  I am also interested in the "normalizing constant" problem.  For both Bayesian and maximum likelihood estimation of hierarchical models, an important problem is to calculate model comparison statistics such as Bayes Factors and normalized likelihoods.  These are mathematically related problems of estimating normalizing constants, which are not easily available from MCMC output, for example.  Therefore additional computational steps are needed to estimate these quantities.

Life history evolution

Natural selection on the schedule of development, dispersal, reproduction and mortality is intertwined with the population dynamics generated by such demographic traits.  I am interested in theoretical models that connect evolution and population dynamics, for example to understand the interactions between  evolution of conspecific brood parasitism and population dynamics.  Here is an errata for typographical errors introduced in the publication process of de Valpine (2000, PRSL-B), in which I used adaptive dynamics theory to derive a general function of life history traits, incorporating stage-specific density-dependence, that should be maximized  by natural selection.

Machine learning, pattern recognition, and bioinformatics

I am interested in the problem of prospective power analysis of studies designed to discover diagnostic patterns from high-dimensional data, such as from genomics or proteomics studies, with low sample sizes.