The Treatment of Endogenous Selection Bias in Household Surveys by Heckman Model

The objective of this working paper is to describe the conditions under which selection bias related to non-response in household surveys can be corrected. Generally,the correction methods implemented assume an ignorable non-response (i.e. missing at random) mechanism. However, when there is an endogenous non-response problem, then the non-response mechanism is no longer ignorable, and the estimators derived from conventional correction methods are biased.

To correct this bias, we propose a weighting based on a Heckman model. This model consists of simultaneously modelling the participation and the variable of interest that we are trying to estimate. However,the identification of the model is conditional on a certain number of hypotheses, such as the existence of an instrument that explains participation but not the variable of interest. In order to have such an instrument, an adapted protocol with independent sub-samples can be setup. This paper details the conditions under which this type of protocol allows an estimate corrected for endogenous selection.

The link to the associated GitHub respository is Ouvrir dans un nouvel onglet the following .