Age and complementarity in scientific collaboration

Abstract

I model research quality as the outcome of a CES production technology that uses human capital measured by publication records as inputs. Investigating a sample of scientific publications with two co-authors, I show that the CES-complementarity parameter is a function of the age difference of the authors. Complementarity is maximized if the age difference between the authors is about 10 years. Two theories are presented which may explain this finding. According to these models, older and younger researchers differ not only in their skill levels but also in the types of their skills and their interpersonal relationships.

Appendices

Appendix 1: The survey

The survey used in Sect. 6 was conducted via mail in June 2011. I sent the questionnaire to 579 economists and business researchers. Among these scientists, 434 were affiliated with German, 72 with Austrian, and 73 with Swiss institutions. I picked a random sample of pairs from the original sample. Each individual researcher was asked about one paper only. Initially, it was intended to ask researchers outside Germany, Austria, and Switzerland, too. However, this idea had to be abandoned for organizational reasons, which is why not always both co-authors received the questionnaires.

The survey participants were not informed that the aim of the study was to relate their answers to the age structure of the collaborating pair. People may have suspected that the survey would be used to investigate plagiarism, e.g., professors letting their students do all the work and then publishing under their own name, which might have reduced the response rate. In order to avoid this, the survey informed all scholars that their co-authors were asked the same questions.

Each letter contained one sheet of paper with a cover letter on the front and the questionnaire on the back and a stamped and self-addressed envelope for the reply. The survey was administered in Germany via the University of Konstanz, in Switzerland via the Thurgauer Wirtschaftsinstitut (TWI), which is located in Kreuzlingen, Switzerland but part of the University of Konstanz and in Austria via the University of Vienna. The survey was sent out in early June 2011; responses were received until mid-August 2011.

The cover letter read as follows:

The questionnaire contained eight questions which could be answered by checking the corresponding boxes. To avoid going too much into the intimate details of their personal relations, people were only asked to distinguish between their relationship being purely professional or friendship.

Table 9 shows additional descriptive statistics beyond the ones already displayed in Sect. 6. It compares the data for respondents with those for the overall sample. 317 of the surveyed scholars responded, which corresponds to a response rate of 54.75 %. No significant differences between the two groups can be observed. The sample of survey respondents is representative of the overall sample, non-response bias does not appear to matter.

Table 9 Survey data: additional descriptive statistics

Table 10 shows descriptive statistics for the 83 articles, for which responses by both authors were available. Although there is substantial variation, on average the shares of the three tasks that the two authors claimed for themselves, respectively, add up to about 100 percent. This provides further support for the assumption that the respondents answered the survey questions honestly.

Table 10 Survey data: papers of which both authors responded

Appendix 2: Survey evidence: Supplementary outputs

Table 11 shows the complete regression output for specification (4) in Table 3. As also discussed in Sect. 4.2, there is a small positive correlation between quality of the journal and article appears in and both authors being women, and a strong negative correlation between journal quality and both authors identifying as business researchers.

Table 11 Non-structural estimates

Table 12 repeats regressions from Tables 7 and 8, but also reports the coefficients for the additional controls. Only very few of these coefficients are significant. Only one distance measure ($<$100 km) is correlated with a scholar’s share of the idea that lead to the article. The better a scholar’s prior publication record, the less of the technical tasks he performed. The better the co-author’s publication record, the smaller a scholar’s share in writing down the results. If, on the other hand, the co-author is female, scholars tend to write more. Scholars affiliated with Swiss universities are more likely to have been friends when they started collaborating. If the authors knew each other for longer than 3 years, it is also more likely that they were friends before they started to work together. Mentor–protégé relationships tend to reduce the probability of having been friends before. That two co-authors have ever, i.e., before or after their collaboration, applied for the same jobs, is less likely the higher-ranked the journal in which the article they have written has appeared in. Competition on the job market is also less common among scholars affiliated with Austrian institutions.

Table 12 Coefficients on control variables

Table 13 shows robustness checks for the regressions from Table 7 in the main body of the paper. Two restrictions apply to the sample now: (i) I only use papers of which both authors responded; (ii) the sum of the contributions that I use as dependent variable must not lie outside [80, 120], which roughly corresponds to a range one standard deviation around the mean (see Table 10). For this restricted sample, it may seem more likely that the authors responded correctly. The statistical significance of some coefficients in Table 7 may, thus, be due to misreporting of individual contributions, which may be correlated with the respondents’ age and the age difference with their co-author. On the other hand, this restricted sample also includes a high number of the modal respondents, who reported 50 percent. Some of them may have check “50 percent” simply because they did not remember the correct individual contributions, which may downward-bias the estimates in Table 13. Also note that the sample size decreased by more than half. With all the unreported control variables, not many degrees of freedom were left, anymore.

Table 13 Concept, technique, writing, and age for subsample with sum $\in [80,120]$