Abstract
This paper evaluates the impact of Ecuadorian innovation support programs, which are intended to enhance firms’ technological and managerial capabilities, on firms’ innovative behavior and performance. In order to estimate the causal effects, we employ different Propensity Score Matching procedures. Results indicate that participating in a program increases firms’ internal R&D and innovation effort, the qualification of the workforce, the likelihood of introducing product, process and organizational innovations and the probability of establishing linkages with research partners. However, participants do not show greater external R&D intensity or a higher propensity to patent, nor are they more likely to cooperate with suppliers, customer or competitors.
Notes
2. ENAI classifies support programs in the following categories: personal training, innovation support, technology adoption and management, production certificates, entrepreneurship support and export promotion.
3. In ENAI, internal R&D is defined as creative work undertaken on a systematic basis in order to generate new knowledge (scientific or technical), to apply or leverage existing knowledge or to use knowledge developed by others, which includes basic research and experimental development; while external R&D refers to the same activities contracted out to external agents.
4. Section 3 describes all the variables.
5. By products we mean both goods and services.
6. The regional variables refer to the Ecuadorian 24 provinces.
7. Results of group differences by treatment status from both nearest neighbor matching estimates and common support analysis are available upon request from the authors. Note that both nearest-neighbor and five-nearest-neighbors also create a control group that is balanced in pre-treatment variables and requires one treated firm to be discarded from the innovative firms subsample.
8. Standard errors for matching estimators are, in practice, generated by bootstrap resampling methods. The use of bootstrapping is backed by Abadie and Imbens (Citation2008), who indicate that the standard bootstrap can be applied to assess the variability of kernel or local linear matching estimators. However, they show that standard bootstrap methods are not valid for assessing the variability of nearest neighbor estimators.