Assessing the Effect of R&D Contributions to Luxembourgish Firms: A Non Parametric Approach through the Application of Spline and Kernel Estimators
Michela Bia1, Alessandra Mattei2
1Population et Emploi, CEPS/INSTEAD, Differdange, Luxembourg; 2Department of Statistics “G. Parenti”, University of Florence, Florence, Italy

Firms' innovation policies are an important feature to support local enterprises. Technological investments are considered an efficient strategy to guarantee competitiveness both at the firm-level and for the economy as a whole. Research & Development (R&D) investments fall in the class of interventions expected to set up technological progress and facilitate growth in the long-run. In this paper we focus on the impact evaluation of R&D financial aids, provided to the Luxembourgish enterprises in 2004 and 2005, using the Community Innovation Survey.

Our contribution to the existing literature is twofold. First, we advance the evidence on the evaluation of financial measures to firms in Luxembourg. Second, a distinct feature of the present paper is that we are interested in assessing the impact of the intensity of R&D subsidies. As the amount of financial aid is related to the local labour market conditions or to firms' performances, the expectation is that firms receiving different amounts of contribution will differ in their labour market outcomes. For this reason, we argue that it is important to go beyond estimation of public policies causal effects employing a binary discrete intervention, and instead estimate dose-response functions of receiving different levels of R&D financial aid. A key identifying assumption is that selection into levels of the treatment is random conditional on a set of observable pre-treatment variables (unconfoundedness). Under unconfoundedness, Generalized Propensity Score methods can be used to estimate dose-response functions. In this paper, we use the GPS to estimate average treatment effects (on the treated) of different levels of exposure to R&D subsidies on firm's innovation sales, by employing both parametric and semi-parametric estimators of the dose-response function. As far as the semi-parametric approach is concerned, we apply the nonparametric partial mean estimator and the nonparametric inverse-weighting estimator based on the kernel method and propose a new nonparametric estimator based on spline technique. More specifically, we aim at i) comparing all different approaches by simulation and ii) applying them to the dataset CIS 2004-2006.

References:

Rosenbaum and Rubin (1983) The central role of the propensity score in observational studies for causal effects. Biometrika, 70. Imbens Hirano and Imbens (2004). The Propensity Score with Continuous Treatments, Applied Bayesian Modelling and Causal Inference from Missing Data Perspectives.

Mattei and Bia (2008). A Stata package for the estimation of the dose-response function through adjustment for the generalized score, The Stata Journal, StataCorp LP.

Flores-Lagunes A., Flores C., Gonzalez A., Neumann T. (2010). Estimating the Effects of Length of Exposure to a Training Program: The Case of Job Corps, The Review of Economics an Statistics.

Cerulli (2010). Modelling and Measuring the Effect of Public Subsidies on Business R&D: A Critical Review of the Econometric Literature, The Economic Record.

Keywords: Continuous Treatment; Dose-response function; Generalized propensity score; Non-parametric methods; R&D investment

Biography: During my Ph.D in applied statistics, I have acquired particular interest in causal inferece issues. In my dissertation, I have developed extensions of such techniques, with particular attention to the Propensity Score based methods.

My last job at LABOR-Centre for Employment Study (Collegio Carlo Alberto) gave me the opportunity to advance my knowledge in such technique, also improving my programming skills.

I am currently working as researcher at the Centre for Population, Poverty and Public Policy Studies in Luxembourg. The National Funds of Research provided me with a grant, for a project focused on 'policy evaluation methods'. My main goal is to develop 'semiparametric and nonparametric techniques in the estimation of direct indirect effects'.

With respect to my mid-long term standpoint, it is very important for me to improve my research expertise and proficiency on applying causal inference methodologies, as my plan is to remain active in a scientific and academic career.