Page 30 - IDEA Study 8 2017 Direct subsidies and R&D output in firms

Firms that apply for international IP protection are fairly rare. It well might be, therefore, that the effects on this outcome are estimated imprecisely, because the data was insufficient in terms of the underlying variability and the sample size, thus statistical power. Arguably, if this is the case, our analysis could be prone to “false negative” errors in statistical hypothesis testing, erroneously concluding that there was no effect. Hence, as a rudimentary test of this possibility, we re-run the models on pooled data from all three IMPULS, TIP and ALFA programmes (see Appendix Table A7 for details). None of the coefficients come out statistically significant at theconventional levels regardless of the matching procedure used and whether the conventional or difference-in-differences estimation is used. In fact, the estimated effects appear to be even less statistically significant than before, which is reassuring. From this we conclude that indeed the subsidy programmes did not make any difference to international IP protection applications. Figure 1 provides a closer look at the magnitude of the estimated effects and respective 90% confidence intervals for Czech IP protection (see Tables 4 and 5 for details). Business enterprises that participated in the subsidy programmes are estimated to have been as much as 13 percentage points more likely to apply for Czech IP protection within three years of the start of funding than their comparable non-participating counterparts. The main effects differ noticeably both by programme and over time. As can be expected, there is not that much difference in the first year (t), because it is too early for activities that were stimulated by the subsidy to come to fruition; the effects start gaining steam in the second year (t+1) and the main thrust of the programmes is delivered in the third (t+2) year. Nevertheless, this holds primarily for TIP and ALFA, but to a substantially lesser extent for IMPULS, whose peak effect remains far below the other programmes. 28