This paper reviews the principal econometric models used to measure the effects of public support for firm R&D investment. A taxonomy classifying papers according to the estimation method used (system of equations versus reduced-form), type of data (cross-sectional versus longitudinal), and type of policy variable (binary versus continuous) is provided. Through a historical reconstruction of the literature, the review starts by exploring the main features of early structural models and their recent refinements, by paying special attention to the issue of subsidy endogeneity. Subsequently, it discusses the second generation of the structural approach, the Selection-models, pointing out their ability in properly describing the two-player form of an R&D incentive scheme. It then goes on by treating and discussing more data-driven approaches, such as those based on Control-function and, more specifically, Matching. Finally, we address evaluation methods rooted in dynamic models of imperfect competition, considered as a novel promising stream of research within the last generation of 'harder' structural approaches. A discussion of features, advantages and drawbacks of each approach in a comparative perspective is offered to the reader, as well as suggestions for future improvements in this field of economic research. * This paper is part of the FIRB 2005-2008 strategic project on 'Models and tools for evaluating the short and medium term impact of firm R&D investments on the Italian productive system' financed by the Italian Ministry of University and Research. I wish to thank Bianca Potì and all the project's members for useful suggestions and the two anonymous referees for extensive comments that substantially improved the paper. All remaining errors are my own.JEL classifications: O32, C52, O38

In this article, I present , a command for estimating a dose–response function when i) treatment is continuous, ii) individuals may react het-erogeneously to observable confounders, and iii) the selection into treatment may be endogenous. I implement two estimation procedures: ordinary least squares under conditional mean independence and instrumental variables under selection endogeneity. Finally, I present a Monte Carlo experiment to test the reliability of the proposed command.

No abstract