Insect population experiments routinely produce short, replicated time-series of population counts under different treatments. Often these time-series are structured by species, stage, size, and/or space. Unfortunately, the most common method for analyzing population data, analysis of variance (ANOVA), does not treat the data as having been produced by biological processes such as growth, mortality, reproduction, and predation. This limits the utility of ANOVA for estimating and comparing biologically meaningful parameters, and therefore does not utilize all of the information in the data. An alternative approach would be to use population models for estimation and hypothesis testing. This approach rests on the same logical foundations as ANOVA, but conducts estimation and hypothesis testing in the framework of population models that consider processes such as growth, mortality, reproduction, and predation. However, a major difficulty in pursuing this approach is that relevant models are complicated and need to include both variability in biological processes as well as inaccuracy in observations, so the statistical relationships between models and data are difficult to calculate.
I will present computational methods for maximum-likelihood estimation and hypothesis testing with stage-structured models that incorporate biological variability and observation error. Key results are that under simulated data sets of typical biological scenarios, inference using population models can have higher power and estimate more biologically meaningful parameters than inference with ANOVA models. Experimental studies of biological control, species interactions, pesticide effects, and numerous other population-level questions could potentially benefit from analysis of population data with population models.
The ESA 2001 Annual Meeting - 2001: An Entomological Odyssey of ESA