178the overall impact and effects of these R&D grants. For example, some support programs seem to yield positive results on innovation and growth, while the results from other programs are less clear, and some even demonstrating negative effects.

Dvouletý et al. (2021) provide a review of empirical studies that have investigated the effects of targeted R&D grants on ﬁrm performance in 28 European Union member countries. The authors only include studies that employ techniques to estimate the counterfactual outcome of the grants, such as propensity score matching (PSM) and regression discontinuity design (RDD). The review covers several different outcome variables, including employment growth but not demand for skilled labor. The results show that 20 studies have investigated the effects of targeted R&D grants on employment growth, and that 18 of these studies report positive employment effects. This indicates that R&D grants targeted toward SMEs can be successful in promoting employment growth. However, the results also reveal signiﬁcant differences depending on the length of the post-support period, ﬁrm size, region, industry, and size of the grant. Zúñiga-Vicente et al. (2014) offer another extensive overview of the impact of R&D grants, based on a compilation of 77 studies from different countries. Their main conclusion is that the results are rather mixed in terms of overall impact, but that there are four clear tendencies. First, the crowding-out effect of a support Do Targeted R&D Grants toward SMEs Increase Employment and Demand for High. . .

179scheme, i.e., the decline in private investments following a government grant, seems to be affected by the ﬁnancial restrictions (e.g., bank contacts, ability to attract venture capital) faced by the individual ﬁrm. Second, the effect of support programs differs between basic research and development projects. Third, the impact of the grant is most likely larger for smaller R&D projects or when the grant is relatively large compared to ﬁrm sales. Finally, there is a time lag before any positive effects of a grant are realized. Koski and Pajarinen (2013) argue that the time lag between a grant and its impact tends to be somewhere between one and three years. One problem with time lags is that the more time that passes between grant receipt and outcome, the greater the risk that the causal impact of the grant is contaminated by unobserved factors that take place during the post-treatment period (Mian & Suﬁ, 2012). However, when investigating the effects of R&D grants on ﬁrms’ demand for labor, Koski and Pajarinen (2013) found that the R&D grant had a positive impact on employment during the time of the support program but diminished after the support program ended. Previous studies also indicate that the effects of targeted R&D grants seem to be larger for small ﬁrms as compared to large ﬁrms (see e.g., Bronzini & Iachini, 2014).

Söderblom et al. (2015) try to address the selection problem when analyzing the effects of a targeted R&D grant program among Swedish innovative startups by comparing data on ﬁrms that received support with those that applied for funding but were rejected at the last stage of the decision-making process. Their identiﬁcation strategy is thus to compare the development of ﬁrms that received subsidies (treatment group) with ﬁrms that applied but were rejected in the last instance (control group). The logic is that those ﬁrms who were the last out offer the closest comparison to the ﬁrms that received grants. The ﬁnal treatment group consists of 130 ﬁrms that received funding during 2002–2008, compared with 154 ﬁrms in the control group that were rejected at the last stage. The results indicate that the targeted R&D grants had a positive and statistically signiﬁcant effect on employee and sales growth, implying that small startups grants can be an efﬁcient way of promoting the growth of new innovative companies. A similar comparison strategy was used by Autio and Rannikko (2016) when investigating the effects of a Finnish R&D program also focusing on growthoriented new ventures. Although not focusing on employment effects per se, they found that ﬁrms participating in the R&D program increased sales by 120% compared to the control group of non-targeted ﬁrms. Howell (2017) analyzed data on ranked applicants to the US Department of Energy’s SBIR grant program, ﬁnding a large positive effect of the R&D grant on revenues and patenting. However, this study does not focus on the effects of targeted R&D grants on employment nor on demand for skilled labor. An implicit assumption behind the identiﬁcation strategy described above is that there are great similarities between those ﬁrms that were supported, and those that almost received support from the program. This kind of identiﬁcation strategy is thus only valid if the ﬁrms that received support were randomly chosen at the last stage of the selection process (Angrist & Pischke, 2008). However, government agencies tend to select those ﬁrms that receive support based on metrics and data from S.-O. Daunfeldt et al.

180personal interviews and expert group evaluations. It is thus likely that there is a systematic difference between the treatment and control group based on the subjective perceptions of these interviews and evaluations, and that those ﬁrms that received support would have performed better than the treatment group even in the absence of support. Note also that the treatment and control groups might be different even if the ﬁrms are endowed with similar observable characteristics. The selection of the ﬁrms that received support might depend on factors that are unobservable to the researchers but are correlated with the future growth of the companies. If we believe that the decision-makers select and recommend ﬁrms that have a higher probability of success, then we would expect that these ﬁrms perform better over time regardless of whether they receive subsidies.

3 Data and Programs Analyzed

To estimate the average treatment effect of a targeted grants program (ATT), information is required about the targeted ﬁrms, the amount they received, and when they received it. We obtain this information from the Micro Database of Government Supports to Private Business (MISS), which is a comprehensive dataset on government support programs compiled by Myndigheten för tillväxtpolitiska analyser och utvärderingar (the Swedish Government Agency for Growth Policy Analysis). The dataset includes a unique ﬁrm identiﬁcation number, which makes it possible for us to merge MISS data with a matched register-based employer– employee dataset from Statistics Sweden that covers all limited liability ﬁrms in Sweden. This dataset includes information on number of employees, investments, sales, value added, industry afﬁliation, and educational attainment of workers, among other variables. We investigate the effects of two R&D grant programs included in MISS, Vinn