Assessing causal effect estimates in structural equation models using instrumental variables

Background/Question/Methods

Instrumental variable regression (RegIV) provides a means for detecting and correcting parameter bias in causal models. Widely used in economics, recently several papers have highlighted its potential utility for ecological applications. Little attention has thus far been paid to the IV methods that are implicitly built into structural equation models (SEMIV), which provide a contrasting analytical approach to the problem. In this talk, I present the motivations, requirements, and basic procedures for using SEMIV. In the talk I consider common sources of bias, which include omitted confounders, reciprocal causation, reverse causation, and measurement error, all of which can all be seen as a single problem – endogeneity. Using data from an ecological field experiment and simulation studies, I illustrate and compare the RegIV and SEMIV approaches.

Results/Conclusions

Empirical results are used to illustrate how IV methods can detect endogeneity and remove its influences. The evaluation of candidate IVs reveals valuable lessons regarding the theoretical requirements and empirical standards for IVs. One essential difference between RegIV and SEMIV approaches that is illustrated is the former’s focus on working around endogeneity and the latter’s focus on modeling endogeneity. In the immediate term, both approaches have individual strengths that can contribute to bias control. In the longer term, the techniques and diagnostics implemented in RegIV are beginning to be implemented within the broad framework of SEM. This opens up the possibility for a simultaneous pursuit of causal inference and explanatory modeling, a common pair of aspirations for ecologists.