I just returned from the fourth International Statistical Ecology Conference that was held in Montpellier, France. As with the previous meeting in Norway, this one brought together a great mix of ecologists and statisticians, most of whom can easily move between both fields.
The plenary speakers were particularly good, with Simon Wood, Chris Wikle, and Mark Beaumont discussing new tools like ABC and synthetic likelihood for inference about complex ecological dynamics. Perry de Valpine and Ben Bolker spoke more generally about practical and philosophical issues associated with statistical modeling and the need (or lack thereof) for so many new methods, which continue to be developed at an amazing pace. Being someone who now advises graduate students, and having felt this way myself at one time, I know how this barrage of new methods can overwhelm and frustrate practitioners who don’t understand advanced topics like hierarchical modeling or who view much of this methodological development as unnecessary.
My opinion is that much of the methodological work may be unnecessary and useless from the perspective of an applied ecologist, but a large portion of it is dedicated to something that is extremely important. Specifically, I view ecology as a discipline with an enormous amount of rigorous theory, most of which has been developed by mathematical ecologists who haven’t worried much about application. We’re now at a point in which we need to apply ecological theory to describe, predict, and respond to the effects of rapid environmental change, and from my point of view, the best way to do this is to fit statistical models to data, data which unfortunately are usually fraught with unique sources of observation error.
Metapopulation theory is a classic example. The theory has been fine tuned for well over 40 years, and it now represents a cohesive set of spatially-explicit hypotheses about how metapopulations should behave. However, even though many applied ecologists are aware of metapopulation theory, until recently, it hasn’t been possible to apply the theory because you couldn’t fit spatially explicit metapopulation models to messy ecological data. But now we can, thanks in large part to computational advances that make modern statistical techniques feasible to implement. As a result, we don’t have to try to figure out how to cram metapopulation data into an ANOVA*. Instead, we can fit a metapopulation model to metapopulation data while accounting for issues like missing data and imperfect detection. This was the subject of my presentation:
One other example relates to species distribution modeling. Ecologists have been studying species distributions forever, and much theory has been developed to explain how density and/or occurrence probability should vary in time and space. But only recently were statistical methods developed to draw inference about these processes from field data. Many people at the conference discussed these techniques, which include spatio-temporal point process models, occupancy models, spatially explicit distance sampling, and spatial capture-recapture. Unfortunately, these methods aren’t always easy to grasp or implement and many researchers have chosen to use methods like MAXENT that, at best, allow you to study something proportional to density. And this is only true if the data were collected using random sampling and there was no spatial variation in detection probability. On top of that, methods like MAXENT don’t allow for inference about temporal dynamics or the underlying processes such as competition or spatial variation in survival. My point is not to bash MAXENT, but to emphasize that good statistical methods are necessary if we want to fit models based on ecological theory and use the models to inform conservation decision making. This was the theme I took away from the 2014 ISEC, and I’m looking forward to the next one in Vancouver.
* Don’t get me wrong, I teach a class on ANOVA and (generalized) linear models, which I view as the best methods available when you can design controlled experiments.