When creating the correlation matrix of organizations and directly comparing it with the matrix of correlations at the meso level, several differences stand out.
Generally speaking, the correlations of organizations are more scattered and spread out. For instance, the matrix is made of a higher number of clusters of correlation that vary in the size of its organizations. The cluster that contains the Technical University of Denmark for example, is quite diverse: with presence of universities from China, the US, Singapore, Portugal and Sweden. This particular cluster is only made of universities and contains no businesses or associations. Other clusters follow this pattern, the fact of being made of only a particular type of institution, which means that universities and the industry still have a “gap” between them.
When focusing on organizations another pattern emerges, these are rarely highly related to other organizations, and instead only strongly connected to a smaller number of institutions.
Finally, one can say that universities tend to aggregate themselves in clusters of innovation, while organizations tend to be more dynamic and targeted in their capabilities. This because the correlation matrix seems to separate universities and businesses. Thinking of the university-industry relationships, these relationships seem to be in the early stages of development: organizations are still “shy” in what regards the targeting of capabilities, while universities are more open and related to themselves.
The organizational profiles of correlation are very good proxies for the innovation strategy that organizations follow. On one side, universities have a fuller spectrum with a higher average correlation, which indicates a more scattered field of capability. On the other side, organizations have a lower average which indicated that they focus on particular areas of interest in a more “closed” fashion.
The Technical University of Denmark is a good example of the above. The top 5 organizations that are more related to it, 3 are Chinese universities, with correlations upwards of 65%. The Chinese university of Tsinghua follows the same pattern. On the other hand, American universities are characterized by higher collaboration with associations (ARS, CSIR) and other American universities.
Finally the comparison tables between DTU and Tsinghua Universities are a good example of how a high correlation expresses itself. The top 10 tables of term pairs have 7 term pairs in common. This indicates that the research strategies of these two universities are extremely aligned, at least in terms of area. This could also be an expression of research partnerships, which are known to exist between these institutions (source), and would be a good quantitative measure of the success of these partnerships.
The characterization of the nature of collaborations has several aspects worth discussing.
On the first hand, most universities (>90%) appear to have at least some sort of collaboration with other universities, which is an expression of the above hypotheses. Danish universities are particularly well classified in the spectrum with all having at least 60% of collaborations made with other universities. This same pattern occurs in Chinese and American universities. However, the bottom 4 universities that collaborates the least with other universities are the 2 Brazilian and the 2 Canadian institutions. This might be related to several factors: the fact that these countries are vast in size and their universities are “isolated”, or the fact the organizations in these countries do not see value in academic collaboration. A report was found (link) where 37% of Canadian organizations did not find collaboration with universities to be relevant for example.
On the other hand, while observing non-universities, the landscape is quite the opposite. Most organizations simply do not collaborate with universities of any sort. This is more evidence of the fact that organizations are still closed to outside ventures. Of the organizations that collaborate with universities, most are associations such as ARS, ACAD, or USDA ARS. Which could mean that associations serve as a bridge between universities and the industry. Meanwhile, Danish organizations such as Novozymes or Dong Energy tend to go against this trend by collaborating more intensely with universities, this might be due to cultural factors (more openness to industry-university collaborations in Denmark), or strategic factors.
In summary, perfect university-industry collaboration seems to be still in its very early stages. Universities form clusters of academic partnerships and collaborations, but industry is still cautious, and reserved in opening up. Denmark is one of the countries “swimming against the current” and investing more seriously in these types of collaborations which is clearly shown by the data.
The first limitation that was the most challenging in this subchapter of the analysis was related to the database itself. The AMICa pathfinder project successfully organized a database where every asset was connected to an owner. However, this characterization is not perfect, particularly in the names of organizations across the database, some names are not particularly well formatted, and other organizations are repeated under different names. This is a consequence of the fact that the AMICa project mixed a number of different databases. Although this does not seem to affect the “big picture” perspective, it could have some repercussions such as misinterpretation etc.
The collaboration part of this analysis was based on the simple detection of the string ‘Univ’ in an organization’s name. Although this serves as a good proof of concept, not all universities and academic institutions use ‘Univ’ in their name, which could distort some of the collaboration numbers.
Finally, throughout the micro analysis, a distinction between Universities and Non-universities was made. However, the world is made of a wide range of different types of organizations: universities, businesses, associations, research centers, and more (Etzkowitz, 2001). Therefore, this simplification is limitative on the type of conclusions that could be drawn.