Hypothesis generation prior to empirical data collection allows researchers to evaluate whether variance generating mechanisms under study can be reliably identified from observed response patterns. Connections between generating mechanisms and observed response patterns can manifest either as systems where the signal does not differentiate the mechanism or where mechanisms are noisy. In the former case, collected data cannot be used to test hypotheses about the target mechanism, whereas in the latter case, parameter estimation and predictive power are hindered. We suggest that agent-based modelling (ABM) can provide value as a hypothesis generation tool to evaluate the nature of connections between generating processes and response patterns. Inference from ABM has historically been challenging because the calculation of a likelihood was computationally intractable. Although modern methods enable the calculation of a likelihood, we suggest that use of these models for hypothesis generation efforts should be given more consideration within ecology.
Results/Conclusions
Using ABM for hypothesis generation offers researchers the ability to explore and quantify potential relationships between predictor and response variables to assess the efficacy of study designs for estimating relationships between generating mechanisms and response patterns. This application is particularly advantageous when researchers are unable to specify testable hypotheses as closed-form equations. ABM allows ecologists studying complex, stochastic, and/or multi-scaled systems to generate and quantify hypotheses from first principles. We present a generalized framework for conducting and interpreting hypothesis generation using ABM prior to data collection, and a case study to illustrate how ABM can be used to test the rigor of conclusions from conventional data analysis of spatial movement data to infer resource selection behavior for a population of flammulated owls (Psiloscops flammeolus).