Page 17 - IDEA Studie 07 2023 TACR

ARE SUBSIDIES TO BUSINESS R&D EFFECTIVE? REGRESSION
DISCONTINUITY EVIDENCE FROM THE TA CR ALFA PROGRAMME IDEA 2023
In total, there is a panel of 7,293 observations, of which nearly three-fourths are SMEs. A closer look at the data confirms that SMEs are significantly different from their large counterparts not only in size, but also in many other structural characteristics, including the likelihood of foreign ownership and operating in the manufacturing sector. However, there do not seem to be significant differences between SMEs and large firms in terms of their time since incorporation, labour productivity, and, interestingly, in their access to public funding for R&D. The average number of proposed project participants is three, and almost all proposals include a research organization. For detailed descriptive statistics of the complete dataset, see Appendix Table A2.
Because the characteristics of proposals are derived from administrative data, they are available for all observations and indicate the maximum size of the sample at our disposal. The drop-out rate due to missing data for other variables ranges between 14 to 18% in SMEs and 7 to 9% in large firms, which is arguably very small for this type of study. It is also noteworthy that the drop-out does not differ significantly above and below the cutoff. The size of effective samples for estimation is lower, because data for all variables included in the estimate need to be available at once, but in none of the estimates presented below does the drop-out rate reach a level which could be expected to entail a significant bias.
Methodology
Estimation strategy
To identify the causal effects of the subsidies, we exploit the fact that the project proposals submitted are evaluated and ranked by independent referees, but the final decision on which will be funded, i.e., where the line is drawn, depends on available funds, which are not known at the time the evaluation and ranking takes place. Intuitively, we can expect firms just above or just below the threshold evaluation points to be very similar, except for the fact that the former group received a grant and the latter did not. If that is the case, subsequent differences in performance between the two groups can be interpreted as effects of the grants.
To formalize this intuition, we adopt the RD approach first proposed by Thistlethwaite & Campbell (1960). It assumes that assignment of treatment conditional on the running variable – in our case, the score assigned to a project – around the threshold is approximately random. We estimate the following stacked RD regression:
15