A toolbox for fitting realistic spatial point process models
Speaker: Janine Illian (CREEM)
These days many, even rather complex statistical models can be fitted routinely as a result of widely accessible high-level statistical software, such as R and winbugs. For instance, even the non-specialist user can estimate parameters in generalised linear mixed models or run a Gibbs sampler to fit a model in a Bayesian setting both of which formerly required expert programming skills. In addition, assessing a model’s fit as well as the comparison of different models has become straight forward. Researchers from many different disciplines are now able to analyse their data with sufficiently complex methods rather than resorting to simpler yet non-appropriate methods.
For spatial point process models the situation is very different. Even though a rich theory has been developed fitting a point process model to data is complex and requires expert knowledge. There have been basically no attempts at providing methods for model comparison for Cox processes.
However, due to rapidly improving technology and a growing awareness of the importance and relevance of small scale spatial information, spatially explicit data sets have become increasingly available in many areas of science, including plant ecology, animal ecology, genetics, geology and medicine. Currently, these data sets are often analysed with methods that do not make full use of the available spatially explicit information. Hence there is a need for making the existing point process methodology available to applied scientists.
Here we suggest an approach to fitting complex log Gaussian Cox processes. We demonstrate that this allows us to fit many realistically complex data sets to spatial point pattern data and that it enables the routine fitting of complex point process models to spatially explicit data sets.