Introduction

Dilemmas of Cooperation in University Networks

Activities in the field of higher education quality development (quality assurance procedures, academic development, digitalisation, joint degree programmes, etc.) are increasingly performed in the organisational form of university networks (Goedegebuure, 2012; Gunn & Mintrom, 2013; Harman & Harman, 2008; Rocha et al., 2018; Williams, 2017).Footnote 1

At first sight, this development can be seen as paradoxical since the predominant approach to university management is largely inspired by the new public management paradigm (Ferlie et al., 2008). This means that higher education institutions are required to act as competitors. With a view to universities’ relationships with other institutions, such competition-orientation rather implies that universities should conceal their internal strategies, weaknesses, and improvement needs. Effective networking, in contrast, assumes mutual openness of the participating partners, particularly concerning weaknesses and improvement needs and also with regard to previous experiences that the individual institutions have made and that might be of benefit to the other network partners. At second sight, however, networking can be expedient. Even in a competition-driven management environment, collaboration is vital since, due to the individual profiles that universities are urged to develop, not all necessary competencies are available at all universities. Some institutions are better than others in the field of digitalisation; others are lagging in this particular regard but have progressed faster than their peers in fostering the diversity of the university’s population. In this sense, university networks serve the purpose of increasing the mutual benefit of all participating partners: not everything needs to be developed everywhere from scratch, and in the best possible case, the network is more than the sum of its parts. At the same time, it is this mutuality that poses risks for the successful implementation of the network, at least when the motivation for the participation is conceptually limited to maximising the individual benefit of the partners, as is suggested by the respective theory of the rationally deciding actor (Becker, 1976).

The respective risks for the individual institution relate to the lack of control over the other participants’ contributions to the network or over their covert intentions to pursue an individual agenda at the expense of the network partners’ benefits. When examining the question of what the determinants for the successful establishment of university networks are, the question of what the mutual benefits are that can be gained from participation in the network arises. To answer such a question, a framework is needed that is capable of resolving the diverse dilemmas of cooperation. Such a framework can be derived from social capital theory, which helps explain the determinants of mutually beneficial cooperation in social networks by introducing concepts such as trust and norms of reciprocity to the explanation of social action (Putnam, 2000). In the case of university networks, establishing and maintaining voluntary associations between universities as such can be interpreted as a form of social cohesion. Hence, mutual trust and adherence to norms of reciprocity can be assumed to be determinants of the mechanisms that enable university networks to produce the desired benefits for all participants.

Social Capital as a Theoretical Framework for the Analysis of University Networks

Social capital as a theoretical framework has gained wide attention in the social sciences in the last decades (Ehlen et al., 2014). The number of studies that employ the concept in its different conceptualisations has increased over time and has produced a huge body of literature (Adler & Kwon, 2002; Baycan & Oener, 2023; Dika & Singh, 2002; Putnam, 2001). Several relevant aspects of this concept inform the empirical work of the present study. In generic terms, Putnam (2000) defines social capital as the interplay between ‘social networks and the norms of reciprocity and trustworthiness that arise from them’ (Putnam, 2000, p. 19). The elements of this definition (networks, social trust, and norms of reciprocity) can also be seen as the core of the concept that unites the existing range of different conceptualisations (Krenz, 2019; Putnam, 2001). Applied to the case of university networks, this means that the participants are forming a social network with a shared goal (i.e. successfully achieving the development aims of the network project). In consequence, their joint working environment is based on mutual trust and adherence to established norms of reciprocity. This refers to the mutual reliability among all participants regarding their willingness and ability to contribute to the network in return for the benefits they receive from their participation. As an outcome, social capital is generated, which can be understood as a network resource or as a medium to assess the trustworthiness of the participating actors. In this sense, social capital as an asset—despite not being tangible (Putnam, 2001)—is developed based on the network participants’ experiences with the other participants’ adherence to the norms of reciprocity.

Notwithstanding the relevance of these aspects in their entirety for the generation of social capital (Putnam, 2001), some authors argue that social trust and norms of reciprocity are not necessarily interchangeable concepts. On the contrary, they are distinct in that ‘social trust relies on strong value consensus and familiarity with behavioural patterns, [while] the commitment to norms of reciprocity is based on mutual acceptance of procedural norms in specific settings of social interaction and does not presuppose consensus between actors on more fundamental issues’ (Grundmann & Traunmüller, 2014, p. 598). For university networks, it can be assumed that social trust among the voluntarily connected network partners is a feature that is easy to build because the involved stakeholders follow the same cultural standards of academia, which exhibit specific ethical values and norms (e.g. in the sense of Merton [1973/1942] and his thought on the ethos of modern science, expressed in norms of communism, universalism, disinterestedness, and organised scepticism). Members of academia are bound to a strong value consensus, and they are familiar with behavioural patterns in the sense of the notion suggested above by Grundmann and Traunmüller (2014). Nonetheless, the concept of adherence to norms of reciprocity seems to mirror more precisely the features of social capital that are most appropriate to describe networks such as an alliance of universities. Reciprocity is particularly useful to analyse organisational structures at the meso level of social interactions (Claridge, 2017). University networks are assumed here to be constituted based on both the perceived benefits at the personal and institutional levels and the existing personal relationships between the participants. They may—at least in part—consist of players who may be connected by ties built on personal and social trust. However, when it comes to the establishment and operation of the network, they may also be connected by a more loosely woven arrangement in which the institutional interests towards a university’s participation in the network outweigh the relevance of mutual (personal and social) trust. Mutuality, both in the sense of accepting procedural norms, which drives the adherence to norms of reciprocity, and in the sense of negotiating and navigating issues of power (Mwangi, 2017), is therefore a determinant in the case of university networks. In this scenario, actors are bound to each other by the ‘act of law’ in the sense that they systematically build organisational patterns (decision structures, advisory boards, regulations, deadlines, partners’ responsibilities, regular meetings) in order to secure the smooth operation of the joint project. By this means they secure fairness, transparency, and the broad acceptance and bond of mutually taken decisions. Based on these considerations, we chose to narrow the concept of social capital in the present study by placing particular emphasis on the aspect of adherence to norms of reciprocity.

University networks usually originate out of a longer-lasting personal relationship between a group of actors. This also justifies the assumption that university networks bring together people who share the expectation that, based on previous experiences with each other, the cooperation will be a success. Nonetheless, the reason for the establishment of university networks can also be more merit-oriented, that is, achieving a specific goal that seems to be unreachable for a solitaire university (for an overview of rationales for the establishment of university alliances, see Fehrenbach & Huisman, 2022). The survey data which we employ for our analyses draw an undecided picture in this regard. About half of the respondents report that the networks under investigation have been founded based on already existing personal relationships, while the other half report the opposite, namely, decisions based on the merits favouring participation in the network. Moreover, the networks which we sampled are project-based, which is why they cover only a comparatively short period of time. This factor might impede the establishment of personal relationships, particularly in cases in which cooperation within a circle of acquaintances was not the point of departure. Wherever partners cannot ‘naturally’ expect to enjoy each other’s trust based on previous positive experiences with each other, it can be assumed that control mechanisms help fill the respective gap and secure a mutually beneficial manner of cooperation (see also Williams, 2017).

Research Questions

The control mechanisms that are employed to secure a steady surveillance of a network need to be conceptualised with more emphasis. It can be assumed that the existence of mutually binding formal rules can even promote the participants’ adherence to norms of reciprocity. If this holds true, it can be further assumed that knowledge of these formal rules’ traits (e.g. their functionality for building transparent structures and tools for the surveillance of the operational work, or their usefulness for reporting schemes) helps raise the understanding of the organisational designs of university networks and their capacity to balance between control-based management and trust-based, reciprocal, and voluntary action on the part of the involved stakeholders.

Along this line of argument, the research question of the present study relates to the relationship between the extent to which network participants adhere to norms of reciprocity on the one hand, and the control mechanisms that are employed to manage the network in the above sense on the other, both of which are presumed to be powerful predictors of the overall success of the partnership. Moreover, we assume that aspects that relate to reciprocal actions of the involved stakeholders and control mechanisms are not mutually exclusive. On the contrary, compliance with norms of reciprocity can even be promoted by mutually binding control mechanisms. In consequence, the overall success of university networks can be enhanced by control mechanisms if they are designed in a way that does not hamper compliance with the norms of reciprocity and mutually trustful relationships.

We further hypothesise that the immediate effect of reciprocal action on a network’s success is stronger in comparison to one that originates from control mechanisms, taking into account the nature of universities as institutions that value self-responsibility and professional autonomy (Pohlenz, 2022), and we give a higher priority to management approaches that emphasise independent self-management based on intrinsically motivated action.

Methods and Database

Operationalisation of Latent Constructs and Suitable Approaches to Data Analysis

According to the above hypotheses, reciprocity and control mechanisms are assumed predictors of a university network’s success. Moreover, the three involved variables are latent in that they are not immediately observable or measurable. ‘Success’, for instance, is context-sensitive and can mean different things depending on what output and outcome was intended and the subjective perception of the individual participants. In the present study, we operationalised ‘success’ with subjective indicators, such as the respondents’ overall satisfaction with the cooperation within the network (see Table 1). Additionally, we included more objective indicators, such as the extent to which respondents can report that the outputs of the network (i.e. products, such as specific intervention formats) have been adopted by the respective target audiences and to what extent they observe a gain of competencies as a result of the network’s impact and as an expression of the attainment of the network’s goals. However, given the diversity of the goals that are pursued by the networks that participated in the survey, it was not the aim of the study to specify the relations between intended and achieved outcomes. We focussed on the respondents’ perceptions of what determines a successful network project instead of measuring goal attainment objectively.

Table 1 Properties of the variables in the model (translation of the German language original questionnaire)
Full size table

Also, ‘control mechanisms’ can be considered a latent construct since control can be exercised in manifold ways. In the present study, we chose an operationalisation of ‘control’ that considers the organisational particularities of universities, namely a high degree of self-responsibility and autonomy. For the design of control mechanisms, this means that they need to be geared at facilitating transparency concerning the operational and managerial implementation of the network but not introduce an overly rigorous regime of sanctioning deviations from work plans. The composition of indicator variables assigned to the latent construct ‘control’ takes into account the voluntary nature of the universities’ association in networks, whether motivated by personal relationships or merit-oriented considerations. The voluntary nature of the networks justifies the representation of control mechanisms with indicators that emphasise the internal monitoring of the process rather than employing external accountability and regimes of sanctioning underachievement and deviations from set targets.

The operationalisation of ‘reciprocity’ draws on the notions of the concept of social capital (Putnam, 2001) and more precisely, the concept of norms of reciprocity. Social capital as such creates measurement problems (Collier, 2002) due to the variety of dimensions which need to be considered (e.g. structural, relational, cognitive; Claridge, 2017) and the various possible levels of analysis (e.g. from phenomena at the macro-level of the social structure to effects of the patterns of organisational structures and individual actions at the meso- and micro-levels; Claridge, 2017). With regard to university networks, we stated earlier that they may be built on both personal relationships and the institutions’ shared interests. This implies that the quality of the relationships that constitute social networks might vary and leave space for interpretation. For the implementation of empirical research in this field, the use of proxy indicators is an escape, regardless of the negative impact their utilisation may have on the validity and reliability of the analyses (Claridge, 2004; Stone, 2001; Stone & Hughes, 2002). The measurement of ‘reciprocity’ in the present study was performed using indicators that address the mutual support that stakeholders in the network grant each other and the multi-directionality of the flow of information that helped all participants gain from the contributions that all network members had made. Moreover, the aspect of timeliness of mutual support, in the sense that immediate action was taken to support one another when in need, served as an indicator of reciprocity (see Table 1). We assume that willingness to adhere to norms of reciprocity finds its expression in a working atmosphere that is characterised by solidarity and cooperativeness. In the same vein, a sense of belongingness to a group which stimulates reciprocal action can be measured by a high degree of self-identification with this group (Üzümceker & Akfirat, 2023). The indicator variables which we employed as measures for the outlined operationalisation of the three latent constructs are displayed in Table 1.

Structural equation modelling is a suitable approach to the analysis of relationships between latent variables. It is based on regression and factor analysis and aims at estimating the relationships between latent constructs based on their representation by measurable indicator variables (Hair et al., 2021; Zhang, 2022). Out of the range of approaches to structural equation modelling, we assessed partial least square (PLS) based models to be the most appropriate one for the present study: The analysis has an exploratory intention since the theoretical conceptualisation of the involved constructs has not yet been fully consolidated (Chua, 2023; Hair et al., 2021). Moreover, PLS estimations are suitable for smaller sample sizes and are less restrictive concerning violations of model assumptions of regression analysis, such as non-normally distributed data (Fauzi, 2022; Hair et al., 2014).

Sample

The data were drawn from a nationwide survey of the German higher education system. As the result of previous desk research (Merkt et al., 2024), we identified 174 network projects in the field of higher education development that were implemented between 2016 and 2022 and met the following criteria, which we had set out for university networks (Merkt et al., 2024). For the present study, we define university networks as institutionalised alliances between universities that pursue a shared goal of contributing to quality enhancement and increasing higher education effectiveness. The alliances have been initiated utilising (public) subsidies as temporary projects or lasting activities. The cooperation is based on agreements between partnering institutions concerning both the specific objectives of the cooperation and the processes geared at achieving the shared goals (cooperation agreements). These processes relate to both the culture of the cooperation (e.g. working in a collegial spirit) at the management and operational levels and the organisational structure (e.g. boards, responsibilities of the participating universities and their agents). Moreover, the network project is visible to external stakeholders, for example through a web page or other channels of communication.

Representatives of the identified networks were invited to participate in an online survey. Out of the 174 networks, respondents from 62 different networks took part in the survey, which equals a response rate of 35.6% at the level of the invited networks. According to the previous study (Merkt et al. 2024), the invited respondents preponderantly can be associated with the ‘third space’ of university management (Whitchurch, 2013). This notion describes a set of positions in between the classical line administration and academia that have been assigned a range of development tasks in higher education management and innovation (e.g. quality assurance and academic development; Whitchurch, 2013). However, third-space positions can also be located at different levels in the hierarchical strata of university management, which is why it is hard to trace what perspective the respondents were taking in terms of seniority or management responsibilities when administering the questionnaire.

At the level of individual participants, the response rate could not be estimated due to the snowball sampling procedure. As stated earlier, the networks in the sample exhibited considerable diversity with regard to the topics they addressed and the units that are/were involved in their implementation (e.g. international offices, faculty, quality assurance departments, etc.). It was thus not possible to trace who exactly was actively implementing the networks at the operational level, which is why we chose a sampling method that accepts self-selection to some extent. This can be considered a limitation of the analyses’ explanatory power; however, given the exploratory nature of the study, we believe this limitation to be acceptable. The final database that is used for the analyses of the present study comprises 160 cases. Table 1 gives an overview of the variables in the model, the mean values and standard deviations of their distributions, and the statistics of the Cramer-von Mises Test for normality of the data distribution (Becker et al., 2021). The latter test is provided by the smartPLS 4 software (Ringle et al., 2022), which we also employed for the estimation of the PLS structural equation model.

Data Analysis

Figure 1 displays the path diagram of the estimated model with the path coefficients for the relationships between the latent variables. In contrast to covariance-based structural equation modelling, no appropriate goodness-of-fit index for the overall assessment of the estimated model is available, or at least the use of respective estimators is not advised (Hair et al., 2021; Henseler & Sarstedt, 2013). The available criteria for the assessment of the model components’ (measurement and structural models) validity and reliability are discussed in the following paragraphs, not in the sense of an overall model evaluation, but rather with regard to their specific meaning for the estimated model parameters.

Fig. 1
figure 1

Path diagram of the estimated model (standardised coefficients)

Full size image

Measurement Model

The measurement model (‘outer model’) specifies the relations of the latent variables with their assigned measurable or manifest indicator variables, while the structural model (‘inner model’) specifies the relations between the different latent constructs (Henseler & Sarstedt, 2013). The measurement variables of the outer model are represented by the rectangles in the path diagram in Fig. 1. Table 2 presents the loadings of the indicator variables on their assigned latent constructs.

Table 2 Summary of the factor loadings and p-values of the indicator variables
Full size table

The model is reflective (i.e. the arrows point from the latent constructs to the indicator variables), which means that the coefficients represent the effect of the latent construct on the assigned indicator variables. The coefficients that display the single indicators’ ability to measure the respective latent construct take on values between 0.503 and 0.878 (Table 2). This means that in some parts, they do not meet the threshold value of 0.708 (which secures that at least half of the variance in the indicator is explained by the impact of the latent construct since the square root of 0.708 equals 0.5; Hair et al., 2022). Nonetheless, with a view to the other assessment criteria for the validity and reliability of the measurements, and with a view to the exploratory research intentions of the model design, we accept these limitations and consider them as an opportunity for further elaboration on appropriate measurements of the relevant constructs.

Table 3 summarises the parameters for the assessment of the measurement model. The values for Cronbach’s α and the composite reliability (ρc) are satisfactory, with values above 0.7 (Hair et al., 2022) while the measurements of the average variance extracted (AVE) confirm that the measurement model needs further elaboration and a better specification (values below the threshold of 0.5 in the cases of ‘control’ and ‘reciprocity’; Hair et al., 2022).

Table 3 Parameters for the evaluation of the measurement model
Full size table

However, in consideration of the discriminant validity as measured by the heterotrait-monotrait ratio of correlations (HTMT) and the respective empirical values < 0.90 (Henseler et al., 2015) that the data take on, the measurements of the latent constructs are tentatively accepted.

Structural Model

The structural model (‘inner model’)—represented by the elliptic shapes in the path diagram in Fig. 1—is usually assessed against four criteria, namely (a) the path coefficients, which estimate the relationships between the constructs within a structural model; (b) the coefficient of determination (R2), which indicates a model’s explanatory power with regard to the respective endogenous latent construct (i.e. the amount of variance in the endogenous latent variables that is explained by the effect of the assigned exogenous latent variables); (c) the effect size as a measure of the relative impact of the exogenous variables (i.e. predictors or independent latent variables in the structural model) on the endogenous variable; and (d) the model’s predictive power (Hair et al., 2022) as indicated by the prediction-oriented model comparison (Liengaard et al., 2021).

The path coefficients in the model correspond with the standardised β-weights in OLS regression (Wong, 2013). They can be interpreted as the respective exogenous variable’s relative contribution to the explanation of the endogenous variable’s variance. The path diagram of the estimated model indicates a strong direct effect of ‘reciprocity’ (0.673) and a moderate direct effect of ‘control’ (0.221) on the endogenous variable’s ‘success’. The indirect effect of ‘control’ over ‘reciprocity’ on ‘success’ (0.360) is stronger than the direct effect, and the total effect of ‘control’ on ‘success’ (0.581) is almost equal to the direct effect of ‘reciprocity’. The latter for their part are considerably correlated with control mechanisms as indicated by the respective path coefficient (0.535). All path coefficients are statistically significant (p = 0.000). The coefficients of determination are estimated for all endogenous variables in the model separately. In the present case, both ‘reciprocity’ and ‘success’ are endogenous since they are explained by the influence of ‘control’. The adjusted R2 value for ‘success’ (0.657) means that 65.7% of the variance in the latent construct ‘success’ is explained by the impact (direct, indirect, and total effects) of the two latent predictor variables, ‘control’ and ‘reciprocity’. The share of variance in ‘reciprocity’ that is explained by ‘control’ (R2 = 0.282) is comparatively small but remains considerable (Chin, 1998). The results suggest that well-functioning and well-designed control mechanisms promote a working atmosphere that encourages mutual support and a trustful working environment. This result supports the second of the above hypotheses. In addition to the coefficient of determination, the effect size (f 2) assesses whether the exogenous latent variables affect the endogenous variables. The respective constructs in the model take on values for the f 2 test statistics that are above the threshold of > 0.02 for a small effect and > 0.35 for a strong effect, respectively (Chin, 1998), as Table 4 shows.

Table 4 Effect sizes of the latent constructs’ correlations
Full size table

For the assessment of the overall model’s predictive power, the cross-validated predictive ability test (CVPAT) has been performed: ‘The test enables a pairwise comparison between theoretically derived competing models, and selecting the model with the highest predictive power based on a prespecified statistical significance level (…). Due to its reliance on cross-validation, CVPAT helps reduce generalisation error, thereby increasing the likelihood that the associated inferences will apply well to other datasets drawn from the same population’ (Liengaard et al., 2021, p. 364).

The respective procedure in PLS structural equation modelling is based on loss functions and quantifies the loss of predictive accuracy associated with the models being compared (Liengaard et al., 2021). With regard to the predictive power, the difference between actual and predicted models is expected to be significantly negative for models exhibiting predictive power (see Table 5).

Table 5 Test for the predictive power of the model with smartPLS CVPAT procedure
Full size table

Overall Model Evaluation

As stated earlier, PLS structural equation modelling differs from covariance-based models in that it does not provide a powerful overall goodness-of-fit estimator (Hair et al., 2021). The criteria that were applied to the measurement and structural models in the previous sections need to be discussed in their entirety. Most of the model parameters are satisfactory, except for the values for the AVE from the indicator variables assigned to the latent constructs (see Table 3). For a satisfactory model fit level, it is expected that more of the variance in the constructs is explained than remains unexplained. Thus, a value of > 0.5 for this particular measurement is the threshold, which is not fully met by the current model settings. In consequence, the composition of the indicators for the latent constructs needs to be reconsidered, particularly in the case of ‘reciprocity’. This is also indicated by the loadings of some of the undesirably low measurement variables. Further theoretical thought needs to be spent on the development of the indicators, representing the latent constructs. However, given that all other model parameters perform, or even outperform, according to the standards, and that the latent constructs correlate—some of them strongly—in the expected ways, these weaknesses of the measurement model are tentatively accepted, and the model is used as it is for further analysis.

Results

The data support the stated hypotheses. Firstly, we assumed that trust-based features (represented by the concept of reciprocity in the present study) of university networks and control mechanisms have distinctive predictive powers to explain the successful implementation of university networks. Secondly, we assumed that the trust-based and control-related traits of a network are not mutually exclusive. On the contrary, the development of trustful cooperation can be assumed to be encouraged by mutually binding control mechanisms that secure the operation of the network in a manner that is beneficial and reliable for all participating partners.

Concerning the first hypothesis, the data suggest that both exogenous latent variables, ‘reciprocity’ and ‘control’, exhibit predictive power for the explanation of the endogenous (dependent) latent variable ‘success’, which is indicated by the statistically significant positive path of coefficients in the structural model. As the path diagram in Fig. 1 also shows, the direct effect of the norms of reciprocity on success is stronger than the one that originates from control mechanisms. The findings suggest that university networks that can rely on the mutual compliance of their members with the written and unwritten rules of the respective cooperation turn out to be the most successful ones. Unwritten rules refer to the expectation of the network partners to receive support in return for the support which they have granted previously to others and to be part of a social network that takes its members’ needs into account.

Moreover, control mechanisms also contribute their share to the explanation of university networks’ successful implementation. Formalised regulations, for example regarding reporting on the network’s progress or the supervision of compliance with the stipulations, are necessary to secure a successful operation of the network. The respective direct effect is weaker compared to the impact of the level of reciprocity among the partners; however, the impact of binding rules on the network’s success needs to be acknowledged.

When looking at the results with regard to the second hypothesis, we also find evidence in favour of the assumptions. Control mechanisms and compliance with the norms of reciprocity are strongly correlated. The influence of ‘control’ as an exogenous latent variable explains more than one-quarter of the variance in ‘reciprocity’ as an endogenous variable (R2 = 0.282). This means that adherence to formalised rules is an important determinant of a working atmosphere that promotes compliance with unwritten rules. By doing so, compliance promotes the generation of social capital within the network, as suggested by the theoretical framework (Putnam, 2001). This argument is even more supported by the assessment of the indirect effect and the total effect of control mechanisms on the networks’ successful implementation. The indirect effect of ‘control’ over ‘reciprocity’ on ‘success’ has a value of 0.360, and the total effect of ‘control’ on ‘success’ is 0.581, almost equalling the direct effect of ‘reciprocity’ on ‘success’. This indicates that much of the variance in the variable representing the success of the network is explained by the influence of the formal structures and control mechanisms. The indirect effect over ‘reciprocity’ indicates that the establishment of binding formal structures helps develop a trustful and reciprocal working atmosphere, which is then conducive to the overall success of the network. This means that reciprocity, understood here as a feature of trust, is not only built based on personal social networks, but in contrast, its generation also considerably relies on the power of institutions and the force of law.

Discussion

Based on the analyses, the answer to the research question of the present study is that trust-based and control-oriented features of university networks do not mutually exclude or contradict each other. On the contrary, both aspects need to be considered in a network’s organisational design in order to ensure that it fully flourishes. Binding rules and the monitoring of the involved actors’ compliance with these rules can even be seen as preconditions for the generation of a trustful cooperation. This begs the question of what the most appropriate organisational structure that balances between the two would look like. University management in the era of the new public management paradigm in the last decades was characterised by an over-emphasis on indicator-based control regimes intended to hold universities accountable for their achievements (de Boer et al., 2007). The market-oriented approach of new public management has been subject to heated debate over the devaluation of academic professionalism (Dill, 2007; Westerhijden, 2007). As an alternative, value-based approaches (O’Flynn, 2007) and trust-based management approaches (Bringselius, 2017; 2023) have been advocated. By definition, ‘trust-based public management is governance and management control models focused on the needs of the service user, where each level of the policy process actively promotes delegation and coordination and attempts to secure its trustworthiness based on ability, integrity and benevolence’ (Bringselius, 2017, p. 3). Nonetheless, value-based or trust-based approaches to university management do not promote a return to state-bureaucratic, small-scale supervision of universities as a response to the deficits of new public management (O’Flynn, 2007).

In contrast, self-responsibility, autonomy, and professionalism are at the core of the concepts. These features can also inspire the organisational set-up of university networks. The relevance of reciprocal action for the respondents’ perception of what makes a network successful reflects their eagerness to create working conditions that are based on mutual trust and thus allow them to mutually grant self-responsibility and autonomy. Taking up the role of a network partner that is contributing to achieving the shared goals in self-responsibility and autonomy requires actors to be able to rely on sound professionalism, for example with regard to performing sound project management as a means of efficient controlling. In sum, the features of ‘post-new public management’ approaches to university management, such as value-based management, seem to be a valid reference point for the design of university networks in order to take into account the need to balance between trust and control. However, the organisational design of university networks cannot be uncoupled from the overall university management framework. In this sense, the discussion of the validity of ‘post-NPM’ approaches to managing university networks needs to be aligned with the overall management paradigm that is applied to universities today. At least, we can consider the case of university networks and the findings that reciprocal action and mutual trust are indeed determinants of successful cooperation as an argument to reconsider the larger scope of university management in its current, efficiency-driven form.

This study has to struggle with certain limitations. Firstly, the snowball sampling limits the representativeness of the data and in consequence, the statistical power of the analyses. This needs to be taken into account, even though the intention of the study is to develop an exploratory approach to the relationships between different features of university networks. Secondly, due to the relatively small sample size, we were unable to differentiate between networks that were newly established and those that have acquired a longer history. It can be assumed that the latter exhibit a level of maturity that almost naturally promotes a trust-based working atmosphere among the partners and their willingness to mutually support each other, regardless of any cost-benefit considerations. The former, in contrast, might be characterised to a higher extent by the need to establish the respective social structures at the beginning of the cooperation in order to build the base for a longer-term cooperation. With the available data, we were unable to control for the respective effects of the networks’ features, such as their longevity. The study can thus be seen as an attempt to heuristically approach the organisational nature of university networks. Further developments may focus on refining the involved constructs’ conceptualisation to provide more precise measurements of the respective relationships.