Solving endogeneity problems in multilevel estimation: an example using education production functions

This paper explores endogeneity problems in multilevel estimation of education production functions. The focus is on level 2 endogeneity which arises from correlations between student characteristics and omitted school variables. Theses correlations are mainly the result of student stratification between schools. From an econometric point of view, the correlations between student and school characteristics imply that the omission of some variables may generate endogeneity bias. Therefore, an estimation approach based on the Mundlak [20 Mundlak, Y. 1978. On the pooling of time series and cross-sectional data. Econometrica, 46: 69–86. [Crossref], [Web of Science ®] , [Google Scholar]] technique is developed in order to tackle bias and to generate consistent estimates. Note that our analysis can be extended to any multilevel-structured data (students nested within schools, employees within firms, firms within regions, etc). The entire analysis is undertaken in a comparative context between three countries: Germany, Finland and the UK. Each one of them represents a particular system. For instance, Finland is known for its extreme comprehensiveness, Germany for early selection and the UK for its liberalism. These countries are used to illustrate the theory and to prove that the level of bias arising from omitted variables varies according to the characteristics of education systems.