Cross-sectional data were used to test the hypothesis and answer the research questions. The data were obtained through a web survey conducted in 2020 as part of the collaborative research project Determinants and effects of cooperation in homogeneous and heterogeneous research clusters (DEKiF). The main focus of the web survey was on collaboration problems in RCs and the solutions to such problems. The survey was sent to N = 15,595 spokespersons and principal investigators (PIs) from RCs of Coordinated Programmes¹ (German Research Foundation, 2017), Excellence Initiatives (German Research Foundation, 2019), and the Excellence Strategy (German Research Foundation, 2020) that were currently ongoing or had been completed after 2015 (Supplementary Figure A1 in Supplementary material). During the 7-week field phase, one invitation and four reminders were sent out to the target persons. A total of n = 3,875 participants completed the survey in full, while n = 1,451 participants completed the survey partially. The overall total of n = 5,326 participants amounted to a response rate (RR2) of 34.15%. No contact could be initiated with 4.27% of the target persons due to outdated email addresses. The sample consisted of 26% females and 74% males. As only 0.001% of the respondents belonged to diverse gender, these were excluded from the analyses due to the small group size. The mean age of the respondents was x¯ = 52.67 years with a standard deviation of SD = 9.52.

According to the German Research Foundation, the submission of a statement by an ethics committee is required if patients or persons with special protection needs are involved. The submission of an ethics approval is also required if a study involves physical risks to the participants, potential participants are not to be informed of a study, participation in a study involves deception, or if a study exposes participants (interviewees, those providing information, project staff, researchers, and research subjects) to exceptional risks (German Research Foundation, 2022). Because none of these cases applied to our study we did not seek approval from an ethics committee. The participants were asked to give informed consent as part of the questionnaire (i.e., written form). The text for the informed consent had been reviewed by the responsible data protection officer.

The RC of the funding lines, coordinated programmes (German Research Foundation, 2023b) and clusters of excellence (German Research Foundation, 2023c) comprise a wide range of differently constituted clusters: our sample includes inter- and intraorganisational, mono-, multi-, cross- inter- and transdisciplinary cooperating research clusters,² of different sizes in terms of personnel, heterogeneous in terms of disciplines, spatially distributed and of different durations (Supplementary Figures A2–A6 in Supplementary material). Beyond the different, overarching funding priorities of the targeted funding lines (such as, for example, the development of a scientific profile at the location of the applicant Higher Education Institutions or the establishment of internationally visible and competitive research institutions), what all RCs have in common is that they fund the work on ambitious, complex and long-term research projects. In this context, the German Research Foundation places central importance on close (often interdisciplinary) cooperative relationships between the PIs of an RC, which are directed toward the production of research results that clearly exceed the possible achievements of individual researchers or subprojects. In this respect, the PIs of an RC must cooperate closely on an ongoing basis in order to achieve their common research goals, fulfill their overarching function and thus ensure the continued existence of their RC (Defila et al., 2006).

The population we are addressing is interesting not only because it contains PIs and spokepersons from different disciplines who work closely together under different conditions in RCs of different sizes and with varying disciplinary heterogeneity, thus decisively shaping the realities of long-term research collaborations. The choice of the population was also methodologically motivated: Before the survey was conducted, a complete list of contact addresses of all target persons in the population was generated via the GEPRIS database (German Research Foundation, 2021), which made it possible to aim for a full survey and to obtain a sample that was robust in terms of inferential statistics (Toepoel, 2016). Due to the availability of a list of all target persons in the population, the difference between the inferential population, the target population and the sampling frame was eliminated (Weisberg, 2009). Biases in the representation of the sample could therefore only occur through unrealizable contacts (see above) and through unit non-response. A unit non-response analysis (Weisberg, 2009) of the sample showed that the non-response error is low with regard to the (1) professional affiliation and the (2) gender of the PIs and directors as well as with regard to their (3) affiliation to current and terminated collaborations (4) of the various funding lines: The relative frequencies of the characteristics of the variables mentioned deviate on average 1.9% in the sample from those in the population (see Supplementary Figures A7, A8 in Supplementary material). In this respect, it is assumed that the following analyses are based on a statistically reliable sample.

Finally, it should be noted that the research project Determinants and effects of cooperation in homogeneous and heterogeneous research clusters (DEKiF) was neither funded by the DFG nor supported by it in any way except by granting access to its collaborative research project database. The target persons were informed that the survey and metadata obtained would be used exclusively by the DEKiF project for scientific purposes, that individual persons and RC could not be identified from published results and that the results would not be exclusively communicated back to the DFG in any way.

To measure the perceived relevance of collaboration problems, we constructed two items for each of the seven collaboration problems. Respondents rated their agreement with the fourteen items displayed in Table 2 on a 5-point Likert scale from 1 = “Not at all” to 5 = “Completely.” The items were conceived based on the research literature on collaboration problems and on analyses of N = 18 expert interviews with spokespersons and PIs (Meißner et al., 2022). All numerical values of the fourteen Likert-scaled items were inverted before analysis, such that an increase in value of an item can be interpreted as an increase in the relevance of the respective problem dimension. To reduce the complexity of the item battery, we calculated a cumulative index of the two items of each of the seven problem dimensions (Spearman-Brown coefficient > 0.6; Table 2).

To analyze potential group differences in the intensity of collaboration problems, we specified eighteen subsamples based on the following seven group variables³: (1) the number of disciplines involved,⁴ (2) the duration of the RCs, (3) the number of their subprojects, (4) their collaboration mode, (5) their disciplinary composition, (6) the individual disciplinary affiliation, and (7) the employment position of the respondents.⁵ The values of the numerical variables (1), (2), and (3) were each assigned to one of three categories (see Figures 2, 3). The following analyses are based on the cumulative indices (from 2 to 10).

To answer RQ1–RQ3, we report the arithmetic mean values of the cumulative indices of the seven problem dimensions. To answer RQ2 and RQ3, a total of seven variables were used to reveal and test possible group differences in the perception of collaboration problems. These are (1) the number of disciplines involved, (2) the duration of the RCs, (3) the number of their subprojects, (4) collaboration mode, (5) their disciplinary composition, (6) the individual disciplinary affiliation, and (7) the employment position of the respondents.

The global relevance ratings of the collaboration problems (RQ1) by the PIs and spokespersons revealed that RCs generally only experience collaboration problems at the collaboration level to a small degree: The mean values for commitment, fairness, communication, difference, relationship, and certainty problems lay at around M = 4. These six problems were thus evenly scaled to have little relevance for both the PIs and spokespersons.⁷ Of even less relevance to the PIs and spokespersons were management problems, with a mean value of M = 3.23 (see Figure 1).

Regarding the differences in the perception of collaboration problems depending on different types of RCs (RQ2), the balloon plot (Jain and Warnes, 2006), visualizing the group differences in the mean values, showed only minor deviations from the results for RQ1 (Figure 2). In general, even after the group differentiation, the relevance usually lay at an arithmetic mean value around M~ 4. One exception was the management problem, whose relevance was assessed by almost all subgroups with a mean value of M~ 3, which was one scale point lower on average than the other problems. The strongest differences in the average relevance rating of the seven problems was apparent when differentiating the RCs according to different modes of collaboration: Those RCs that collaborate in an interdisciplinary or even transdisciplinary manner (see Supplementary Figure A9 in Supplementary material; Kocka, 1987; Jungert et al., 2013) showed, on average, somewhat lower relevance ratings of the seven problems compared to mono-disciplinary, multidisciplinary, or cross-disciplinary RCs.

Considering the visualization of the arithmetic means of the twenty sub-groups through the balloon plot (Figure 2), it can be seen that there are no significant mean differences between the groups for the variables number of disciplines involved, duration and number of sub-projects (Table 3). Deviating from this, slight group differences emerged regarding the variables collaboration mode and disciplinary composition: with the exception of the feature combinations cross-disciplinary vs. mono-disciplinary, cross-disciplinary vs. multidisciplinary, and mono-disciplinary vs. multidisciplinary RCs, all forms of collaboration differed mutually significantly with regard to assessments of the relevance of the problems (Table 4). Additionally, RCs with at least one PI from the humanities/social sciences differed significantly from RCs with at least one PI from the natural, life, or engineering sciences (Table 5). However, as the permutation test for homogeneous variances in the subgroups of the variables collaboration mode and disciplinary composition was significant in each case (i.e., homogeneous within-group dispersion could not be assumed), both test results should be treated with caution. On the basis of non-homogeneous within-group dispersion, it remains unclear whether the significant differences occurred because there was no homogeneity in the dispersion of within-group distances between the groups or because the groups had significantly different localized centroids in multivariate space (Anderson, 2001).

The findings regarding differences in the perception of collaboration problems depending on various individual characteristics of the PIs and spokespersons (RQ3) likewise showed only minor deviations from the results of RQ1 (Figure 3). Even after differentiating the relevance assessments of problems according to disciplinary affiliation and employment position, the arithmetic mean value mostly lay at around M~ 4. An exception is once again the management problem, whose relevance was assessed by almost all subgroups with a mean value of M~ 3. The strongest differences in the average relevance assessment of the seven problems can be seen in the differentiation of employment position: in particular, PIs and spokespersons who are not employed in the academic-scientific field assessed all seven collaboration problems as having somewhat higher relevance, with an arithmetic mean value of about M~ 5.

Accordingly, the PERMANOVA only revealed significant mean value differences between the groups of the variable current employment position (Table 6). As the permutation test for homogeneous dispersion for the variable current employment position was also not significant—i.e., homogeneous within-group dispersion could be assumed—this finding can be considered as stable. The post-hoc test showed that PIs and spokespersons who were not currently employed in the academic-scientific field rated the problems as significantly more relevant than did professors, academic councils, research assistants, and PIs or spokespersons who held another position in the academic-scientific field (Table 7).

In order to model information about the co-occurrence of collaboration problems (RQ4), we specified an MDS based on a distance matrix of Pearson correlations between the cumulative indices of the seven collaboration problems. Although the stress value of the resulting two-dimensional MDS configuration was 0.09, this was significantly smaller than the stress value expected from random data using a significant permutation test (p = 0.01). In this respect, the fit of the MDS solution with regard to the specified dimensions and the resulting stress value was therefore acceptable (Borg et al., 2018).

To ensure the replicability of the MDS solution and to avoid a premature stop of the iterative adaptation of the MDS at a local minimum of the stress value, different starting configurations of the MDS were specified (Mair et al., 2016). Furthermore, the stability of the MDS solution was tested using bootstrapping (Jacoby and Armstrong, 2014). The dashed ellipses around the indicators in the geometric space of the MDS configuration mark their 95% confidence regions (Figure 4). The more compact a confidence ellipse is, the more likely it is that the true position of the corresponding problem lies on or near the centroid of the respective ellipse. Figure 4 shows that all additive indices have a uniform, rather low uncertainty.

With a total stress value of 0.10, the specified MDS can be classified as acceptable. The stress per point (SPP) value can also be used to quantify the contribution of each input variable to the overall stress value of the MDS. The graph we specified (Figure 4) shows the SPP value of the seven problems based on the size of their localization point. Accordingly, Figure 4 shows that difference problems contribute most to the total stress value and commitment problems contribute least.

The center of the geometric MDS configuration (Figure 4) demonstrates that relationship problems are clearly associated with difference problems. The approximately central position of the relationship and difference problems in the geometric solution of the MDS further indicates that management problems, but especially certainty, commitment, fairness, and communication problems are associated with both relationship and difference problems. In this respect, relationship and difference problems occupy a central position in the network of relationships of the seven collaboration problems: Their occurrence is clearly associated with that of the other five problems. In addition, the lower right corner of the MDS chart shows that communication, fairness, and commitment problems are located in relative proximity to each other, indicating that these three problems are moderately associated with each other. Finally, the problems located at the outer edges of the MDS diagram are again largely uncorrelated with each other: Management and commitment problems as well as certainty and communication problems are thus essentially unrelated.

Overall, the MDS reveals that relationship and difference problems co-occur with management, certainty, commitment, fairness, and communication problems. Furthermore, fairness and communication problems co-occur relatively often. Management, certainty, and commitment problems, on the other hand, are not clearly exclusively associated with another problem dimension.

The objective of the present study was to gain a deeper understanding of the relative relevance of problems in collaborative research and the interconnections between them. Building on previous insights into the most current problems, we sought to find answers to four major issues, which we addressed through one hypothesis and three research questions: (1) researchers' perceived relative relevance collaboration problems, (2) differences in these perceptions based on the type of RC, (3) and on the characteristics of researchers, and (4) the co-occurrence of collaboration problems.

With respect to issue (1), the results of our representative survey of leading participants of RCs revealed that, broadly speaking, collaboration problems were experienced only to a small degree. Furthermore, we found almost no differences with regard to their perceived relevance in terms of their actual occurrence. The only exception is the management problem, which was experienced to an even smaller degree than the other six problems (i.e., commitment, fairness, communication, difference, relationship, and certainty). This finding is in stark contrast to the qualitative results we relied upon and which, based on a qualitative interview study, hypothesized that difference, commitment, and certainty problems were perceived as more important than the other problems (Meißner et al., 2022). Concerning issues (2) and (3), there were almost no differences concerning the perceived relevance of these problems depending on the type of RC or on the individual researchers. There were three exceptions: First, for all of the seven problems included in the study, non-academics perceived a higher relevance of collaboration problems. Second, collaboration problems occurred less frequently in RCs with a more complex collaboration mode (i.e., trans- and interdisciplinary) and third, in RCs with at least one PI from the humanities/social sciences. Lastly, addressing issue (4), we found close links between relationship and difference problems on the one hand and between fairness and communication problems on the other. The remaining three problems (i.e., commitment, management, and certainty) seem to occur rather independently. Furthermore, relationship and difference problems seem to play a central role in the overall picture of collaboration problems.

Taken together, the present findings do not reveal substantial variations regarding the perception of collaboration problems. Some groups generally perceive the seven problems to be more relevant: non-academics, participants of mono-, multi-, and cross-disciplinary RCs, and participants of RCs with members from the humanities/social sciences. While the result that trans- and interdisciplinary RCs perceive a lower relevance of collaboration problems is rather surprising, the other two results can well be explained in our view: First, non-academics might experience more struggles because most RCs might be more adapted to the viewpoints and routines of the academic sphere. Second, because of their scientific background researchers from the humanities/social sciences might be more sensitive to interpersonal problems. However, overall and within each group there is very little variation. Based on our findings, we cannot speak of problems that are of high or medium relevance, but only of less relevant problems. There are several potential reasons for this finding. For example, it might be the case that if research collaboration is perceived as problematic, this could affect all parts and aspects of the collaboration. In turn, this may result in an overall perception of the RC as problematic, in the sense of a halo effect. It might also be possible that researchers have difficulties to trace a problematic collaboration back to a specific problem, especially if confronted with very short items in a questionnaire without the opportunity to carve out a particular problem discursively like in a qualitative interview. Both of these explanations would imply that in practice, researchers cannot easily replicate the clear distinction between specific problems which we proposed in theoretical terms.

However, our results suggest that a problematic collaboration seldom occurs at all. The vast majority of researchers perceived a low relevance of problems, which implies that working in RCs is, at least for the leading personnel, generally experienced as smooth and unproblematic. Post-hoc analyses with respect to the overall problem level within the RCs in our sample support this insight: Only 9.7% revealed a high problem level and 1.0% a very high problem level, which in the words of Bozeman and Youtie (2017) can be called “routinely bad” or “nightmare” RCs. Instead, 56.0% may be called “routinely good” and 33.3% even “dream” RCs.⁸ Our study thus draws a more optimistic picture of the nature of collaborative research than one would likely expect. With regard to scientific practice, there seems to be little need for specific measures to address collaboration problems such as training for research leadership. Although this conclusion concurs with previous research (Bozeman and Youtie, 2017), it is rather abstract and should be treated with caution. As the aforementioned figures show, there are individual RCs that struggle with more or less intense problems that may even be likely to put the whole collaboration at risk. It would certainly be worthwhile to take a closer look at “routinely bad” or “nightmare” RCs in order to gain an even more nuanced understanding of collaboration problems and their implications—taking into account the principle that “each unhappy collaboration is unhappy in its own way” (Bozeman and Youtie, 2017). Furthermore, it would be interesting to take a closer look at researchers who are not employed in the academic field, because they are revealed to have a slightly greater problem perception than their colleagues from academia.

Concerning the co-occurrence of specific problems, our findings are also quite plausible. First, the co-occurrence of difference and relationship problems suggests that if differences are perceived as too high, it can be challenging to establish good relationships among the researchers. For example, different working styles may not only lead to conflicts in the workplace but also reduce the motivation to meet socially after work. Put differently, if researchers feel that they need to bridge various forms of differences, this can definitely strain personal relationships, probably even on a daily basis. The co-occurrence of fairness and communication problems is similarly plausible: Insufficient or one-sided communication can certainly result in researchers simply being unaware of their colleagues' work. Consequently, they might get the impression, whether justified or not, of an unfair distribution of individual inputs and outcomes. At the same time, greater communicative effort can be a valuable strategy to overcome any perceptions of unfairness in an RC, i.e., to resolve the fairness problem (Meißner et al., 2022).

Our findings can be considered as robust because they rely on a large-scale representative survey, which is rare in this area of research. Nevertheless, some limitations need to be acknowledged. First, we developed the items to measure the collaboration problems based on a qualitative study (Meißner et al., 2022) and did not test these items in advance. Thus, although the item development was theoretically and empirically informed and the items revealed to be valid in analyses apart from this study (Hückstädt, 2022), the construct validity of our measure (Cronbach and Meehl, 1955) needs to be tested in a further study. Second, our sample only included PIs and spokespersons of RCs, and did not include representatives of other roles assumed by researchers in collaborative research, such as doctoral students. It might well be the case that researchers in leadership positions have higher autonomy or power which might lead to different experiences than those of more junior scholars. Specifically, they might overlook or underestimate problems that younger participants in lower hierarchical positions experience. Thus, a more comprehensive sample might report more problems or assess the relevance of some problems differently. Furthermore, with their answers the speakers in our sample actually evaluated themselves, which may have led to certain biases due to a lack of critical self-reflection or honesty. Thus, further research is needed that takes the views of researchers on different career stages into account.

A third limitation concerns the specific RCs that we included in our target population: RCs funded by the German Research Foundation. While the German Research Foundation is the most important and prestigious funding agency in this country and the free accessibility of the data on these RCs provided a good opportunity for our research, our study is limited in geographical terms and does not include RCs funded by other bodies such as the European Union and specific programmes such as H2020. Furthermore, our sample does not include RCs who failed to raise funding or that were not dependent on external funding. Because dsyfunctional RCs might struggle already on the way to get funding and RCs without funding might be less interested in the success of their research, including them in the sample would probably lead to different results. Therefore, future research needs to explore whether our findings can be generalized beyond this specific sample. Lastly, our findings rely on self-reported data, bringing the usual issues: The respondents may not have been aware of collaboration problems or may have been unwilling to admit that their RC is not actually a “dream” one. Therefore, our findings need to be validated through more objective parameters such as bibliometric measures in order to produce findings that may be used to guide policy decisions. For example, a RC might experience itself as unproblematic but its productivity in terms of published papers might suggest otherwise.