The Distinction between Dictatorial and Incentive Policy Interventions and its implication for IV Estimation

IZA Logo

IZA Seminar

Place: Schaumburg-Lippe-Str. 9, 53113 Bonn

Date: 01.12.2009, 12:15 - 13:30


Presentation by 

Christian Belzil (Ecole Polytechnique, Paris)


We investigate if the economic nature of a policy intervention affects the capacity of IV methods to estimate relevant treatment effect parameters, in a context where the error term of an outcome equation is generated by a multi-state intertemporal (possibly dynamic) model. We consider two distinct classes of experiments; those that are dictatorial in nature (which require either a minimum or maximum consumption level), and those that are based on some incentive provision. We show that incentive-based policy interventions generate more post-intervention randomization (a lesser degree of selection on individual endowments among the sub-population affected) than dictatorial interventions. For this reason, and for a relatively wide class of data generating processes, IV methods using incentive-based policy reforms should always outperform IV estimates that make use of dictatorial policy interventions. We illustrate these concepts within a calibrated dynamic life cycle model of human capital accumulation, and focus on the estimation of the returns to schooling using two popular types of interventions; namely a mandatory schooling reform, and a set of education subsidies. In the human capital accumulation framework, the class of model, for which incentive-based instruments are superior to dictatorial ones, incorporates (i) the models in which post-schooling accumulation decisions are independent of schooling, after conditioning on heterogeneity (the static model), and (ii) those in which schooling reduces the cost of accumulating further skills, after conditioning on heterogeneity (the dynamic skill-complementarity model). For instance, IV estimates using compulsory schooling regulations, and implemented on both the static and the skill-complementarity models, lie well outside the support of the distribution of the returns to schooling (in the negative orthant), and are significantly far from the population treatment effect. At the same time, those IV estimates using various education subsidies, and implemented on the same model structures, lie within a very narrow distance to the relevant population treatment effect.

For more information, please contact