Causal mediation analysis: a new methodology for the identification of causal mechanisms through an RD approach

This Ph.D thesis is comprised of three self-contained, but related essays, corresponding to the three different chapters, on the causal mediation analysis. Causal mediation analysis is a statistical framework used to study causal mechanisms. In such a context, a mechanism is defined as a process in which a causal variable of interest, known in literature as treatment, affects an outcome through one or more intermediate variables, called mediators, that lie in the causal pathway between treatment and outcome. This methodology has been developed above all in sociology, psychology and epidemiology. Surprisingly, few studies used this approach in economics, despite the great importance in knowing causal mechanisms of policy interventions and of economic phenomena. In fact, the main limit of the traditional policy evaluation approaches is that the causal effects can be estimated but without knowing nothing about the causes of the effects. In other words, we can estimate how large is an impact and if it is positive or negative, but we cannot know what is due that impact, leaving the causal effect as a "black box". This thesis tries to make new developments in this direction. First of all, we try to use this approach in the economic field: causal mediation analysis is an important tool with a great potential, that permits to go deeper with the analysis and know more about economic phenomena. It is no more sufficient to know if a policy intervention worked or not, but it's becoming more and more important to know "why", in order to design more efficient policies. Secondly, we propose a new estimator trying to go beyond the limits imposed by structural models: until a few years ago, researchers used predominantly structural equation models to study causal mechanisms. Only recently, some researchers moved towards new approaches, like counterfactual methods and quasi-experimental designs. Following these studies, we propose a new estimator that takes advantage of Regression Discontinuity Design (RDD) to solve the limits of the traditional mediation framework. Thirdly, we propose an interesting application to validate this model. In particular, we estimate the EU 2007-2013 Regional Policy on the 2006-2015 GDP per capita growth rate at NUTS 3 level, investigating if part of this effect is driven by Research and Development (R&D). The results suggest that the EU Regional Policy has a positive and significant impact on the per capita GDP growth rate, estimating a total treatment effect of 9.4%. A little part of this effect, 1.5%, is driven by R&D investments, confirming to be a mechanism of transmission of EU Regional Policy, even if not statistically significant. To conclude, the idea of this thesis is to give a contribution to causal analysis in order to have better interpretations of the results and, then, to better know the phenomena that surround us. After a critical survey of the literature reported in the Chapter 1, the aforementioned issues are directly faced in chapter 2 and 3.