Efficient progress in understanding complex ecological systems can be achieved by formulating mechanistic models and and fitting them to time-series data. Frequently, the models that embody our hypotheses are naturally formulated in continuous time but the data are taken at discrete intervals. For this problem, approaches of varying degrees of ad-hockery have been proposed, but accessible methods of general utility have remained elusive.
Results/Conclusions
In this talk, I describe a recently-developed likelihood method for statistical inference with several nice properties. It can accommodate models in continuous or discrete time with nonlinearity, process noise, measurement error, non-stationarity, missing data, and hidden variables. Most importantly, the method has the “plug and play” property: one can use it to perform statistical inference for any model one is capable of simulating. Using measles dynamics as a case study, I show how likelihood based on mechanistic models is a versatile and valuable tool for scientific investigation.
