Page 13 - IDEA Study 02 2020 Recruitment of Researchers

WHERE DO UNIVERSITIES RECRUIT RESEARCHERS? 2020 Robustness test To test, how significant is the limitation mentioned above, we directly found out for a sub-sample of researchers, from which university they graduated and compared conclusions with regards to inbreeding based on this information with the benchmark results based on the affiliations in their initial articles. The aim was to gauge, how far off the mark we are using our main method. The test covered three disciplines: i) Biochemistry, genetics and molecular biology; ii) Physics and astronomy; and iii) Social sciences. In each discipline we randomly selected 30 researchers, for which we manually found out from publicly available sources (such as ORCID, personal web pages, etc.), whether they are currently based on their alma mater. As the alma mater we considered university, in which the respective researcher obtained her highest academic degree, in the vast majority of cases a Ph.D. diploma. For 8 researchers we could not find this information, hence replaced them by another random draw. The classification of researchers, whether they are currently based on their alma mater, was the same using both approaches by 77% in biochemistry, genetics and molecular biology, 90% in physics and astronomy and 87% in social sciences. The initial affiliations thus turned out to provide a relatively reliable indication of where the researcher graduated. The differences between disciplines are small, which is surprising given the large differences in the propensity of doctoral students to publish, but perhaps this confirms that even if dissertation research is published only after finishing the studies, the researchers still tend to acknowledge their alma mater in their publications. The classification differed for 14 researchers. Of those there was the same number of 7 false positives and false negatives, which tend to offset each other, thus the impact on the presented ratio to total researchers was even more limited. As far as disciplines are concerned, 7 were from biochemistry, genetics and molecular biology, 3 from physics and astronomy and 4 from social sciences, and regarding the affiliations, 4 were from American and British universities, 6 from Western European continental universities and 4 from the Visegrad universities. Hence, it seems that the bias tends to be relatively evenly spread across the board. There were three reasons for the different classification. First, the researcher published her initial articles with affiliation to her employer, not the university from which she graduated (5 cases). Second, the researcher had a relationship to her current university, hence published with affiliation to it, before or during doctoral studies elsewhere (3 cases of false positives). For example, the researcher published already during master studies, then obtained a doctorate at other university and came back. Third, there are measurement errors, such as a joint Scopus ID for two authors with a similar name or inaccuracies in matching the initial and current affiliations (4 cases of false negatives). Overall, the test indicates that the presented results are in fact fairly robust to this kind of limitations. The differences in classification can hardly account for the significant differences between universities in our main results, especially if one keeps in mind that there are both false positives and false negatives that cancel out each other. Even though the robustness test is encouraging, however, we proceed with caution when interpreting the results, as the match of affiliations in current and initial articles is only an indirect indicator of inbreeding. 11