The endogeneity problem in 'multilevel' models

5th November 2024, 09:00-11:00

Pr John Antonakis, UNIL

The endogeneity problem in multilevel models (2 hours)

Data are often hierarchical organized under an entity having measurements over time (longitudinal panel data) or sub-entities nested under it (pseudo-panal data). Such data are often incorrectly modeled by researchers in psychology and management. The key problem is that researchers often make an assumption that unobserved heterogeneity, the so-called "fixed-effect" in econometrics, is orthogonal to the regressors: If it is not orthogonal and if it also correlates with outcome under study, estimates will be biased and inconsistent (i.e., they will not converge to the true population value as the sample size increases). Popular go-to solutions and modeling choices often fail to examine this assumption. I will show how to test for this assumption and to ensure consistent estimation of multilevel data.

UNIL, Géopolis building, room 4799

20