A key assumption of regression analysis (or structural equation modeling) is that the modeled independent variables are not endogenous. Yet, the problems of endogeneity are not well known to researchers working in many social sciences disciplines (e.g., management, applied psychology, sociology, etc.). When the independent variable has not been exogenously manipulated, there is a strong possibility that its relationship to a dependent variable will not be correctly estimated, leading to spurious findings. This podcast gives a brief and vivid overview to endogeneity and why it is engendered. Prof. John Antonakis discusses the problems of endogeneity using non-technical language and intuitive explanations; he shows that the observed relationship that is estimated can be very misleading when the independent variable is endogenous.

A key assumption of regression analysis (or structural equation modeling) is that the modeled independent variables are not endogenous. Yet, the problems of endogeneity are not well known to researchers working in many social sciences disciplines (e.g., management, applied psychology, sociology, etc.). When the independent variable has not been exogenously manipulated, there is a strong possibility that its relationship to a dependent variable will not be correctly estimated, leading to spurious findings. This podcast gives a brief and vivid overview to endogeneity and why it is engendered. Prof. John Antonakis discusses the problems of endogeneity using non-technical language and intuitive explanations; he shows that when the independent variable is endogenous--which is also possible in experimental designs (when the mediator is endogenous)--the observed relationship that is estimated can be very misleading. Prof. Antonakis demonstrates how the problem of endogeneity can be solved using procedures borrowed from econometrics (i.e., two-stage least square regression estimator).

It is well known that endogeneity leads to inconsistent estimates. Unfortunately, many researchers working outside of economics are not aware of the problem of endogeneity and how to deal with it. Prof. John Antonakis shows how the two-stage least squares (2SLS) estimator recovers causal estimates in the presence of endogeneity (which includes the problem of common-method variance). He also shows that endogeneity can even be prevalent in experimental designs, when researchers estimate mediation models; that is, where the causal effect of an exogenous variable on a dependent variable is mediated by an endogenous variable (or a manipulation check).

Ulrich Hoffrage questionne John Antonakis sur la recherche qu'il a conduite avec Olaf Dalgas et qui donne lieu à une publication dans la revue Science de février 2009.

Ulrich Hoffrage questionne John Antonakis sur la recherche qu'il a conduite avec Olaf Dalgas et qui donne lieu à une publication dans la revue Science de février 2009.