The impact of local and external university knowledge on the creation of knowledge-intensive firms: evidence from the Italian case

For the statistical definition of KIFs, see below. According to the literature, typical examples of KIFs are R&D laboratories, high-tech firms, law and accounting firms, and management, engineering, and computer consultancy companies (Alvesson 1995).

One may contend that academic spin-offs (Rothaermel et al. 2007; Colombo and Piva 2012) and university–industry collaborations (Perkmann et al. 2013) are important mechanisms of knowledge transfer from universities to the productive system that should be taken into account when considering the effects of universities on the creation of new KIFs. However, in this paper, we are explicitly interested in how far in space university knowledge extends its effects, depending on its codified versus tacit nature. As academic spin-offs and university–industry collaborations encompass both the production of tacit (e.g., know-how concerning a production process) and codified knowledge (e.g., a patent), their introduction into the analysis might have confounding effects and lies beyond the scope of this paper. In excluding them from the analysis, we are consistent with mainstream research on the impact of universities on new firm creation at the local level. Indeed, the impact of academic spin-offs and university–industry collaborations has been studied mainly with reference to innovative regional and local activities, while research on knowledge spillovers and new firm creation has largely focused on university knowledge embedded in academic patents, scientific publications, and graduates.

http://​www.​infocamere.​it/​movimprese.​htm; see Sect. 3 for a detailed description.

http://​www.​scimagoir.​com/​. This split corresponds roughly to making a distinction between the top and bottom 50 % of the distribution.

Several studies have found that R&D expenditures by universities have a positive effect on new firm creation at the local level (e.g., Harhoff 1999; Woodward et al. 2006; Kirchhoff et al. 2007).

The authors considered only the effect of the closest university, ignoring the effects engendered by other universities.

According to Baltzopoulos and Broström (2013), this is particularly true for students who choose to relocate to attend a specific university. They find themselves in new environments and have the chance to build entirely new social networks.

We use an exponential specification because we assume that the dependent variable follows a negative binomial distribution. See Sect. 3.2 for further details.

As explained in Sect. 4, we do not consider the raw count of publications but rather an alternative measure obtained by weighting publications depending on the research areas to which they belong.

One might expect agglomeration externalities to extend far beyond the border of a province. In Sect. 5.3, we describe how we control for this effect.

The variable Border _i equals zero if the province shares borders with one of the two enclaves within the Italian territory, i.e. Republic of San Marino and Vatican City State.

Because in some provinces (11) there are both high- and low-quality universities, we cannot specify interaction terms in the model, discriminating between provinces with high- and low-quality universities.

However, as a robustness check, we also run instrumental variable regressions to estimate Eqs. (1) and (3). See Sect. 5.3 for a detailed description.

The NUTS classification is a hierarchical system for dividing up the economic territory of the EU. It subdivides each member state into NUTS level 1, level 2, and level 3 territorial units (for further information, see http://​epp.​eurostat.​ec.​europa.​eu/​portal/​page/​portal/​nuts_​nomenclature/​introduction). In Italy, NUTS level 3 units correspond to intermediate administrative divisions (provincie). During the period 2005–2009, seven new provinces were created (Olbia–Tempio, Ogliastra, Medio Campidano, Carbonia–Iglesias, Monza–Brianza, Fermo, and Barletta–Andria–Trani). Therefore, the current number of Italian provinces is 110. However, data on new KIFs and on territorial characteristics are not available for these new provinces.

See Appendix 2 for the list of knowledge-intensive industries included in the sample.

It is likely that some of the KIFs created during 2010 are academic spin-offs. Unfortunately, we do not have data on the exact localization of these academic spin-offs, and so we cannot exclude them from the sample to check whether their presence affects our results. This is undoubtedly a limitation of the present analysis. However, given the low number of spin-offs with respect to the total number of new KIFs, it is very unlikely that their presence would bias our results. According to the NETVAL report (NETVAL 2012), in Italy in 2010, 117 academic spin-offs were founded across all industries, representing 2.5 % (117/4,716) of the total number of new KIFs in 2010.

Criteria for inclusion were the following: the existence of institutionally recognized research units, the existence of an official research mandate, the presence of regular PhD programs, the consideration of research in strategic objectives and plans, and the regular funding of research projects by public agencies or private companies. See Bonaccorsi et al. (2012) for a more detailed description and full-scale analysis of these data.

We thank one of the anonymous reviewers for raising this important point.

Scimago is generally considered a reliable source of data for comparative analysis because it does not measure only publications in top journals or by highly cited scientists but rather covers a wider range of publications. Nevertheless, being based on international publications, it clearly underestimates the quality of research in the humanities and social sciences, in which a larger share of output is published in books and national-language journals. We consider this limitation acceptable because the research production most relevant to new KIF creation, as shown in Bonaccorsi et al. (2014), is from scientific and technical fields, which are well covered by the raw Scimago data. Another limitation is that the Scimago rankings of institutions are based on four indicators, three of which are independent of size (percentage of international collaborations, normalized impact score, and percentage of publications in high-quality journals) and one of which (number of publications) is not (see http://​www.​scimagoir.​com/​methodology.​php?​page=​indicators for details). This may explain why some small high-quality Italian universities do not appear in the top 40 list. For these reasons, use of the label “high-quality” or “low-quality” does not imply at all an overall evaluation but is rather a convenient shorthand for comparing universities with respect to research production in fields that are well covered by the Scimago data and are most relevant to new KIF creation.

The LR test reported at the bottom of Table 4 confirms that including Patents _i ^external in the regression significantly improves the log-likelihood with respect to a restricted model in which Patents _i ^external is set to zero. Conversely, the LR tests concerning Publications _i ^external and Graduates _i ^external do not reject the null hypothesis that the values of these latter variables are equal to zero.

The LR test results reported at the bottom of Table 5 confirm that including xLQ _i ^local and xLQ _i ^external in the regression does not significantly improve the log-likelihood with respect to a restricted model in which these variables are set to zero (for all types of university knowledge). Hence, we cannot reject the null hypothesis that the impact of knowledge (both local and external) produced by low-quality universities is zero.

For illustrative purposes, let us consider a dedicated training program taught by professors of a high-quality but distant university or a large incubator located on the premises of a local university. In the former case, entrepreneurship in the focal area is likely to be positively influenced by the knowledge produced by the distant university, while in the latter case, the knowledge produced by the local university is likely to remain highly localized.

Results concerning Eq. (1) are similar to those reported here. They are not shown for the sake of synthesis, but are available from the authors upon request.