Page 20 - IDEA Studie 07 2023 TACR

ARE SUBSIDIES TO BUSINESS R&D EFFECTIVE? REGRESSION
DISCONTINUITY EVIDENCE FROM THE TA CR ALFA PROGRAMME IDEA 2023
to a project. However, based on our conversations with TA CR representatives, in practice, they did so very rarely, and even in those cases the change typically did not affect which projects were ultimately funded.
We test the validity of the identifying assumptions in two ways. Firstly, in Figure A1 in the appendix, we show the results of running a McCrary (2008) test, which compares the density of the distribution of project scores below and above the cutoff. We see no significant discontinuity in density at the cutoff in calls 1, 3, and 4. In contrast, we observe a substantial and statistically significant discontinuity in the case of call 2. To avoid the risk that the scores were indeed manipulated around the cutoff in call 2 and that this would bias our results, we exclude call 2 from all subsequent analyses. In Figure A2, we subsequently show the results of the McCrary test that we obtain when we combine all projects in calls explored in the analysis, i.e., calls 1, 3, and 4. The figure shows no evidence of discontinuity in the density around the cut-off for these projects.
If the assignment of treatment conditional on the score received by a project around the cut-off is approximately random, we should not observe any differences between the treated and control observations around the cut-off. To see if this is the case, we conduct placebo tests in which we estimate a version of our estimating equation with outcomes given by various firm and project characteristics observed in the 4 years before the start of the project (years t-3 to t). As in the analysis of post-treatment effects below, we conduct the estimation separately for SMEs and large firms. We report the results in Appendix Table A3. In total, over 19 outcomes and 2 firm size classes, we estimate 38 placebo tests. The definition of significance levels means that, in the absence of any pre-treatment differences around the cut-off, on average, 3.8 tests should be significant at the 10% level and 1.9 tests at the 5% level out of pure luck. This is exactly what we see, with 4 of the 38 tests proving to be significant at the 10% level, 2 at the 5% level, and 0 at the 1% level. In other words, we see no evidence of differences in pre-treatment characteristics of firms below and above the cut-off.
In summary, after excluding call 2, we see no evidence of score manipulation based on the McCrary (2008) test, and no evidence of differences in pre-treatment characteris- tics around the cut-off. These two facts together make us reasonably confident that any differences in post-treatment firm outcomes, as presented in the next section, have a causal interpretation.
18