Endogeneity

• In Event: Methodological Pluralism

Abstract

The problem of endogeneity is ubiquitous in the empirical analysis of social phenomena. Social, political and economic processes are complex and this implies that everything depends on everything else. Yet, our theories and empirical models are notoriously under-determined and often do not take into account causal heterogeneity across units and over time. Endogeneity comes in different shades: unobservables, selection, simultaneity, co-determination, reversed causality and many more. The empirical examples are manifold from institutions and economic growth to wage and performance.

The underlying problem is the unobservability of counterfactuals. The identification revolution in the social sciences has understood this very problem and the simple solution is random assignment to treatment, or the simulation of this. However, not all research questions that are of interest lend themselves to this simple solution and if they do, questions of external validity and generalizability arise.

This paper discusses these issues, the theoretical and empirical solutions and the problems these solutions create. I will theoretically discuss and empirically investigate the performance of endogeneity tests and show that they are rather part of the problem than the solution. I will discuss the merits of instrumental variable approaches as well as exploiting variance over time with fixed effects and first-difference models as proposed solutions to endogeneity.

Author

• Vera Eva Troeger, University of Warwick