Endogeneity is one of the main problems faced by any economist wishing to study the relationship between several variables. Mathematically, this problem corresponds to the fact that the (central) assumption of no correlation between the explanatory variables (x) and the error term (e) will be violated. This results in a bias in the estimation of the interest coefficient and therefore in misleading conclusions. There are three main sources of endogeneity in econometrics:
– Simultaneity: if the economist wants to show that x causes y, variable x must not be influenced by y. See more details on this previous insight here. In this context, one solution is to instrument variable x with another variable w that is not influenced by y.
– Unobserved heterogeneity: it may be that the relationship between x and y is simply linked to the effect of a third factor z. See more details on this previous insight here. In this context, if possible, a variable controlling for this factor z should be added to the model.
– Measurement error: some variables may not be measured accurately, which will affect the estimate, resulting in a bias in the estimation of the coefficient of interest and therefore misleading conclusions[1].
Julien P
Notes
[1] Note that we could add one last source of endogeneity that is often cited as important: the case of time series data where a lagged independent variable is included in the regression (i.e., the « y » from the previous period explains the « y » from the current period) while the residuals are autocorrelated.
