Ecologists spend a great deal of time and effort learning modern statistical methods and tools for the analysis of their data. Traditional training in statistics, however, rarely goes beyond the basic tasks of description and estimation to consider how one evaluates the causal networks that regulate systems. As a result, there persists an intellectual separation between the enterprises of statistical description and causal modeling. We believe one factor contributing to this persistent separation is the use of separate toolboxes by statisticians and those using structural equation models to investigate causal hypotheses. In this talk we illustrate how to develop and evaluate structural equation models for the study of complex cause-effect hypotheses using traditional statistical modeling methods. Data from studies of wetland ecosystems and wildlife populations are used in these analyses.
Results/Conclusions
We illustrate structural equation modeling using two contrasting ecological examples. The first example illustrates the evaluation of different mechanisms whereby human disturbance can trigger dominance of invasive species in wetlands. In this example, it is shown how a change-point regression model can be subsumed within a larger structural equation model that traces cattail abundance back to human land-use intensity through human influences on water chemistry. The second example illustrates how Bayesian mark-recapture wildlife population models can be converted into causal hypotheses to investigate the relative influences of biotic and abiotic conditions on population dynamics in Atlantic puffins. In this second example, model results indicate that local climatic conditions have both indirect and lagged impacts on puffin survival through influences on local abundance of herring. Together, these examples show how the use of familiar statistical models can be developed into structural equation models representing causal hypotheses.