The limits of collaborative governance: The role of inter-group learning and trust in the case of the Estonian “ Forest War ”

Despite the reported success of introducing collaborative governance to the field of Estonian forest policy, conflicts over regulations escalated to an unprecedented extent in 2017. We analyze the institutional design and process of collaborative governance in this area in order to understand the reasons behind the failure of this governance arrangement. Our empirical analysis is based on a mixed methods approach combining network analysis with qualitative analysis of interview data. Our analysis reveals that the collaborative institutions were unable to generate shared understanding of the mission and the ground rules of decision-making, provided uneven facilitation, failed to build trust, and thus were unable to establish an arena conducive to learning. We further stress the role of network methods in capturing adequate information from an institutional setting involving multiple participants.


| INTRODUCTION
In Estonia, collaborative governance mechanisms have been mandatory in forest policy since putting in action Forest Policy strategy document in 1997. An analysis based on data from 2015 found that "the cooperation between forest owners and environmentalists in Estonia's forest policy making processes has been relatively good" (Teder & Kaimre, 2018, p. 59). By 2017, however, conflicts in the field had escalated to such levels that the media dubbed the situation a "Forest War" (Lõhmus, 2017;Väli, 2017;Vilbaste, 2018). This conflict manifested itself in the social movement "Estonian Forest Aid" (EMA), which has gathered close to 8000 followers on social media, led to several demonstrations in 2016-2017, and caused a heated media debate. An Internet search for "forestry" in Estonian newspapers received 96 returns for the year 2014 compared to 773 for 2017.
What happened?
We analyze this case as an example of a failure of collaborative governance on a national level to draw attention to the limits of this governance arrangement. It is hoped that collaborative governance strategies will help integrate knowledge from different stakeholders, diffuse best practices, and balance different interests, resulting in adaptive and flourishing socio-ecological systems (Bodin, 2017;Koppenjan et al., 2004;Moseley & Winkel, 2014;Sorensen & Torfing, 2005. Also, participation in the process of deliberation is seen as an integral part of democratic governance (Butler, 2017). On the other hand, co-governance does not automatically increase the legitimacy and compliance of non-state actors but should be assessed against wider ideals of equal citizenship and public reason (Birnbaum, 2016). Participatory ideals are often not implemented successfully (Nordberg & Salmi, 2019), and when they are, participatory governance is fragile and failure-prone, requiring careful and skillful This article analyses reasons for failure of collaborative environmental governance based on forest policy in Estonia using network and qualitative analysis. design and management (Rowe & Watermeyer, 2018;Sorensen & Torfing, 2007). We argue that collaborative governance can manage competing goals and provide win-win outcomes when the demand for resources is low but when demands increase and the resource becomes more contested collaboration is challenged and requires strong institutions for successful functioning.
Estonia offers a particularly interesting case for a "new institutionalist" analysis (Paavola, 2007;Scott & Thomas, 2017). This is because while it has been praised as a successful example of an ex-Soviet state that has adopted stable Western political institutions, it also practices participatory policy making, combining the power of the state with solutions based on voluntary cooperation (Bertelsmann Stiftung, n.d.). Although critical voices have been raised concerning the openness of policy processes to public participation in Estonia (e.g., Lepa et al., 2004), the situation has gradually improved (e.g., Uus et al., 2011). The ongoing concern, however, is that cooperation in policy formation is often "for show" and with little impact despite being legally required and exercised (Rikmann et al., 2019). We argue that the "Forest War" was a result of poorly designed and implemented collaborative governance mechanisms. Haphazard ground rules for decision making led to a situation that failed to generate shared understanding of the mission. This, in turn, resulted in a failure to build trust and hampered reciprocal learning. We demonstrate this by drawing on theories of collaborative, participatory and network governance and provide insights into the influence of institutional design and social-psychological factors on the outcomes of participation.
In particular, we wish to draw attention to policy-learning aspect of cooperative governance (Dunlop & Radaelli, 2013;Harvey et al., 2019) as the crucial element. The management of ecosystems challenges government arrangements because of the contested nature of the problem and the contradictory values, institutional regimes and objectives of the various participants involved (Defries & Nagendra, 2017). This is why it is commonly referred to as a wicked problem. We view participation a vessel for bringing new information to a policy making process that is otherwise inaccessible to the regulators. If this important aspect in collaboration is missing, the whole process is jeopardized. The journal Environmental Policy and Governance lately dedicated a special issue to knowledge coproduction (Rodela & Gerger Swartling, 2019). The editorial stressed the need to explore further the links between institutional design, learning outcomes, and their effects on environment. Although several examples exist of collaborative environmental management, fewer examples can be found of cooperative arrangements on the national level (Scott & Thomas, 2017;Yaffee & Wondolleck, 2000). As a small country, Estonia offers a well-bounded example of this kind of high-level collaboration effort. Ansell and Gash (2008) reviewed studies of collaborative governance and the factors that predict successful collaboration, and they proposed an approximate model for the institutional design of collaborative governance in a process that includes the history of cooperation or conflict, incentives for participation, power and resource imbalances, dialogue, trust, and shared understanding. We contribute to the model by adding a network analytical layer to their analysis and indicating its relevance to the policy outputs. This layer sheds light on the core aspect of collaborative governancenamely the relations between actors. With our analysis, we aim the answer the following questions: 1. How does collaborative forest governance function in Estonia? 2. What is the relationship between reciprocal trust and learning in the network of participating organizations? 2 | THEORETICAL BACKGROUND

| Collaborative governance in forest policy
Collaborative governance is an umbrella term for arrangements where public agencies involve non-state actors in a formal, deliberative decision-making process (Ansell & Gash, 2008). Collaborative governance tools are used in all policy domains, but they are particularly prevalent in natural resources management, with a number of applications in forest policy (see Beland Lindahl et al., 2017). Often, this has meant a focus on community-level management schemes in developing countries, but developed countries have also opened up policy-making to outside stakeholders (Maier et al., 2014;Nordberg & Salmi, 2019).
At its best, collaborative governance influences policy outcomes by making the decision-making process easier, more legitimate, and more democratic. This is achieved by balancing actors' interests (Bodin, 2017), thus making governance more accountable, opening policy gridlocks and avoiding litigation (Koontz & Thomas, 2006). In other words, collaborative procedures can be more socially legitimate (perceived as fair by stakeholders and society) and more normatively legitimate (satisfying the democratic norms of deliberation, political equality, and public reason) (Birnbaum, 2016). It is unsurprising, therefore, that stakeholders view participation positively, despite the fact that the participation process may not have an effect on the participants' ideologies and interests (Maier et al., 2014;Teder & Kaimre, 2018) .
An integral part of collaboration is integration of knowledge from different knowledge systems and generation of new knowledge through social learning (Bodin, 2017). Learning, in turn, increases the legitimacy of outcomes, as they are perceived as more desirable or more fitting to normative standards, such as justice (Birnbaum, 2016).
Nevertheless, policy learning is challenging, even when participatory processes are well designed. For example, a study of different locations in Germany (Sotirov et al., 2017) investigated whether participants' beliefs, values and cooperative behavior changed when they were forced to engage in "forward thinking" during the participation process, that is, considering the long-term development of forests.
They found that only some strategic learning and no substantial learning occurred. This lends support to the claims that collaborative governance does not work in highly contested and high-risk areas.
Therefore, at its worst, collaboration may lead to discrepancies between written regulation and policies, on the one hand, and actual governance practices on the othera sort of "window dressing" instead of accountable governance (Raitio & Harkki, 2014). Furthermore, in the context of Bulgaria and Germany, Winkel and Sotirov (2011) found that participation strategies had little effect on outcomes and actual practices because participatory governance either engaged environmentalists in lengthy and futile negotiation processes or simply created policy access for donors and powerful actors. Thus, cautionary examples of failures of collaboration and situations where failure is expected are plentiful (Ansell & Gash, 2008;Bodin, 2017). Consequently, collaboration should address the question of accountability alongside that of policy learning. The success or failure of learning in collaborative governance depends on how the initiating governing body establishes the processthe institutional designand aspects of the process: leadership and accountability.
We discuss these two issues in the following sections.

| Institutional design: Arenas, rules and leadership
The main component of collaborative design (Torfing & Triantafillou, 2016) that influences the success or failure of cross-sectorial collaboration is the creation and maintenance of arenas for interaction.
Arenas are venues within policy networks where negotiations are held and which provide the desired outcomelegitimacy, learning, social capital and trust (Ostrom et al., 1994). They include, for example, committees, conferences and round-tables (Koppenjan et al., 2004).
Rules influence the choice of participants in the arenas, their positions and actions, the use of information and control, as well as the outcomes of strategic behavior and the whole arrangement (Ostrom, 2005). If there were no rules, the result would be a Hobbesian state of nature where common good is extremely unlikely to emerge (Ostrom, 2005, p. 211). Moreover, it can be argued that rules should be in place to force participants to learn from each other. It is crucial that participants recognize mutual interdependence, as this facilitates joint production of meaning (Koppenjan et al., 2004). Ansell and Gash (2008) review characteristics of successful arenas and conclude that such arenas need to be inclusive, in the sense that they must include all the affected stakeholders; at the same time, however, they should be exclusive, in that they must be sole forum for discussing the issue at hand, with no alternative venues for bypassing them. Participants must both believe that a stalemate is undesirable and also refrain from utilizing other venues for achieving the results than the negotiations at hand (Leach & Sabatier, 2005). Otherwise, participants may block cooperation. The situation where a network of actors blocks either some ideas or some actors is called "closure" (Kickert et al., 1997;Schaap, 2007). In this case, participants may use alternative strategies to achieve their goals, such as expanding the network to include new actors or forming coalitions within the network.
A working arena requires good leadership (Ansell & Gash, 2008;Scott & Thomas, 2017). The governing body delegating its decisions to collaboration should act as a facilitator, organizer, encourager, and network-broker. Moreover, leaders must mediate and give a voice to participants in order to enable the synthesis of ideas from different organizations. Nonetheless, they must also introduce and follow rules that are recognized by other state institutions.

| Collaborative process: Learning and trust
Even in an arena that is effectively established, problems can occur in the collaborative process itself. The co-production of learning often requires meticulous design and attention to detail from the institutions that create it, and even then the transformative nature of new knowledge may be limited (Harvey et al., 2019). Nonetheless In settings with conflicting interests, institutional arrangements are expected to create trust between participants and, through that, enable learning. For a thorough overview of trust and its effects on cooperation see Klijn et al. (2010); however, we highlight some relevant points here. First, one of the most widespread definitions of trust is the "willingness of a party to be vulnerable to the actions of another based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party" (Mayer et al., 1995, p. 712). Trust has been extensively studied with game theory, and the results indicate that without a single central power, trust is a sine qua non for any Pareto efficient solutions (Axelrod, 1984), because trust allows participants in collective good management to refrain from egoistic behavior by reinforcing the belief that other participants will do the same (see Põder, 2010 for practical applications). Trust creates predictability regarding others' behavior, which is central to planning one's own strategy.
Trust and learning are intrinsically connected because trust influences how we gather and process information. Trust in people eventually influences trust in the information they convey. We tend to accept information more easily from trustworthy partners and question information from partners we do not trust. On the one hand, it influences our perception of the source of information. For example, Dunlop and Radaelli (2013, p. 602) argue that one of the central axes around which learning can be systematized is the "certification of actors." This concerns whether "teachers" have legitimacy and can be trusted. They observe that often it is the institutional structures (organizations, procedures) that provide these qualities. On the other hand, trust enhances participants' capacity to incorporate information. For example, Murphy et al. (2012Murphy et al. ( , p. 1700) demonstrate that trust increases the "absorptive capacity" of firms, that is, the "ability to recognize the value of new, external information, assimilate it, and apply it." In other words, trust is especially important in situations where we lack knowledge of the issue (Paglieri & Castelfranchi, 2014). In such cases, the only alternative is to rely on interpersonal trust. Paglieri and Castelfranchi (2014) base their argumentation on Tuomela and Tuomela's (2005, p. 71) idea that "[w]hen knowledge is lacking, trust is needed for cooperation." Ecosystem management has been described as exactly the kind of situation where precise knowledge is lacking and conflicting values discourage technical top-down solutions for knowledge creation (Bodin, 2017;Defries & Nagendra, 2017).
Therefore, new knowledge can be created only in conjunction with trust. In addition, Butler (2017) addressed this question from a philosophical perspective, arguing that deliberative democracy cannot achieve legitimacy without an epistemic learning process.
How, then, is an atmosphere of learning created in the policy process? First, participants must be able to discuss their "core beliefs" (Leach & Sabatier, 2005, p. 494). This means that sufficient space is created for discussing the values upon which the participants base their normative requirements. Second, there is a need for a neutral mediator in the negotiations. The task of the mediator is, among others, to force competing coalitions to justify their claims with scientific, quantitative data. If joint production of meaning is discouraged, an agreeable policy outcome is unlikely to occur (Harvey et al., 2019;Leach & Sabatier, 2005;Weible et al., 2009).
Few studies have directly measured trust or other relational qualities between governance network partners (Peters et al., 2017). Usually, what is measured is trust in general and how it influences the outcome of participatory (environmental) projects (e.g., Barroso-Méndez et al., 2016;Klijn et al., 2010;Leach & Sabatier, 2005). Measuring generalized trust and learning, however, fails to reveal who actually trusts and learns from whom in collaborative settings. This can lead to skewed understandings of collaborative processes and create false security by confirming that something, indeed, was learnt.

| DATA AND METHOD
Since we are interested in several inter-connected and context-specific aspects of collaborative governance, such as institutional design, trust and learning, we apply a mixed methods approach (see Saunders For gathering information about reciprocal trust and learning, we conceived a name-interpreter questionnaire (Carrington et al., 2005) with a list of all organizations that were mentioned in the preliminary interviews and document analysis. The questionnaire was answered by 27 respondents who were either heads of their organizations, or, in case of ministries, civil servants involved in forest policy field.
Respondents were able to add organizations to the list if necessary. This gave us a total of 45 organizations. For every organization, each respondent was asked to rate, on a Likert scale, one independent quality (perceived power) and four reciprocal qualities. These were intensity of interaction, information received, information given and trust. Trust was operationalized with the question "How much do you trust that the organization would take your point of view into consideration if forest policy were entirely their responsibility?" In turn, learning was operationalized with the question "How much useful information do you get from this organization?" The questionnaire data was analyzed using the ggraph (Pedersen, 2018) and igraph (Csardi & Nepusz, 2006) packages of software R. The network survey allowed us to identify 44 organizations with an interest in Estonian forest policy. Thirty of them were core members with multiple mentions and multiple ties. By contrast, 15 actors were mentioned only once and had a narrower interest in forest policy (such as the Roads Administration, Farmers Union, Academy of Sciences etc.). Consequently, these were excluded from the statistical network analysis for the sake of clarity.
We use cluster analysis to find groups of organizations that trust each other and learn from each other. We use an optimal clustering algorithm (Brandes et al., 2008) that maximizes the strength of links within each cluster and minimizes the strength between clusters. We then measured the average in-group and out-group "trust" and "learning" relationships. Exceptionally, we positioned the Parliament (RK) in a separate cluster in order to stress its role as the overarching political organization (e.g., the commissioner of the Forest Development The qualitative analysis used interview data, complemented with grey literature. Out of the 27 respondents of questionnaires, semistructured interviews (Brinkmann, 2014) were conducted with 19 people. These interviews covered the topics of recent changes in forest policy, the background of these changes, the functioning of significant arenas, and relationships between organizations. As our goal was data saturation, several people were interviewed multiple times, as new questions arose during the data analysis. The interviews lasted between 1 and 2 h and were recorded and transcribed. Grey literature included policy documents, reports and studies commissioned or published by organizations in the policy network, most importantly by the Ministry of Environment.
The semi-structured interviews were analyzed using thematic analysis (Guest et al., 2012), which combined deductive and inductive coding.

| BACKGROUND OF ESTONIAN FOREST POLICY
The development of Estonian forest policy after the restoration of independence in 1991 is an explicit example of an attempt to establish collaborative governance. Moreover, it is the first policy area in which the Estonian government applied this kind of governance after the collapse of the Soviet Union. Indeed, as the head of The FDP routinely triggers amendments to the Forest Act, and discussions in the FC may also lead to regulation change.
Since 1999 the Forest Act has been changed more than 20 times, This created the background for two important processes in 2016 that directly contributed to the "Forest War." First, RMK introduced a proposal to the FC to lower the felling age of spruce (Picea abies) on fertile soils from 80 years to 60 years. Second, planning of a 1-billion-euro timber refinery in Estonia was made public. Both of these processes became the central issues behind wide-scale protest on national and local levels. In that year, the social movement EMA was mobilized, with heated media debate and demonstrations following at the end of the year (for a more detailed discussion, see the results section below).

| RESULTS OF QUALITATIVE AND NETWORK ANALYSIS
We now analyze the institutional design and process of collaboration in order to understand why the collaborative partners were unable to respond to mounting tensions. Here, we focus on the post-2008 years that roughly correspond to the period of the last FDP (2010-2020).
Especially important are the years that follow the analysis by Teder and Kaimre (2018), who conducted their empirical work in 2015, because this is the period that led directly to the "Forest War." We build on Ansell and Gash's (Ansell & Gash, 2008, p. 550) model of collaborative governance to analyze both the institutional design and the collaboration process. Our interviews and document analysis reveal, however, that no formal agreement exists on the rules for reaching decisions in the FC.

| Institutional design
Nonetheless, in the interviews, the participants agreed that decisions were taken in a deliberative manner and different opinions were accounted for. As one of the key civil servants stated in an interview, the Ministry was accustomed to cooperation between organizations running smoothly, so they saw no need for the establishment of formal procedures.
The decisions we make there, we try to achieve consensus. This is of course not a place where financing is decided. We decide what kind of advice we should give to the Minister. And we look for common ground.
So there are no strict rules. The only rule is that we Nevertheless, the agenda setting rules of the arenas were opaque, and, in the interviews, the environmentalists frequently raised this question as problematic. Moreover, the "level of discussion"a proxy for the depth of the issues on the agendawas characterized as technical, with little debate on core policy beliefs (Leach & Sabatier, 2005).
Furthermore, the source of the agenda was not publicly discussed, although the topics were often highly significant, such as the maximum clear-cutting area, minimum felling age, and carbon sequestrating. Formally, the agenda was set by the Forestry Department of the Ministry of the Environment. However, the interviewees speculated that the topics for discussion at the Council arose from the management problems of the RMK. This organization runs a large-scale operation, since they manage close to 50% of Estonian forests.
It seems to be a legitimizing mechanism for the Minis- Hence, our analysis indicates that the rules of the arenas were neither clearly set nor followed. Moreover, although formally inclusive, the participation of important actors was often disregarded.

| The collaborative process
The institutional design set the stage for a collaboration process that deepened the inherent problems of cooperative governance. The rules of agenda setting are intrinsically connected to the use of expertise and scientific knowledge in the FC. Thus, instead of "common fact finding" (Ansell & Gash, 2008), our interviews and FC meeting transcripts indicate that academic analysis was often ordered to support a certain position rather than to study a problem in depth (e.g., the felling age of spruce, clear-cutting area, carbon sequestration). In most cases, fundamental research questions were not addressed, and thus scientific consensus was not achieved. The environmentalists argue that the ministry is not acting as a neutral facilitator who gives a voice to all parties. Some industrialists even argued that the ministry should not attempt any facilitation at all, as they were feeling certain their arguments will prevail. This led to tension and low levels of trust between the participants. This tension reached its zenith with the proposed change to lower the felling age for spruce in specific fertile soils. The argument advanced by RMK was that in such soils spruce begins to rot before it reaches its current felling age of 80 years. Thus, they proposed that the age be lowered to 60 years. This produced outrage among environmentalists and was criticized even by some industry figures for being too aggressive.
When we discussed the lowering of the cutting age for spruce, we had divided opinions. Environmentalists were quite opposed to this; for industry people, it was completely reasonable to cut them before they rot.
Well we could have done this anyway, but this change would have simplified the bureaucracy. And I remember there were those who said, "Let's not poke this. This is the wrong signal. We can cut these trees anyhow. This is the wrong signal if we start cutting young spruce. There could be trouble!" And those who said it were prophets! (Interview with a civil servant) In order to convince the environmentalists, the Ministry proposed an increase in protected forest areas in those fertile soils. From their perspective, it was a quid pro quo proposition. What made the offer controversial, however, was that the FDP 2010 had already included an increase in such areas, which had never been put to practice.
Hence, in the official transcripts, the issue was framed as "missing  (Schaap, 2007), as the criticism was blocked and diverted at the FC. It is noteworthy that no correlation was found between the organizations' perceived influence on policy outputs and their interaction centrality (r(17) = 0.06). This indicates that influential participants in policy making were not necessarily central to the interaction, lending support to the argument that the collaborative process lacked the ability to commit the actors to the collaboration process, a feature that is significant for the success of collaboration efforts (Ansell & Gash, 2008;Kickert et al., 1997;Schaap, 2007). It also confirms the qualitative finding that the arenas lacked exclusivity. Thus, powerful actors may have other channels to influence political decisions. Furthermore, our analysis revealed that although there is dense interaction between most organizations, the quality of that interaction in terms of trust and learning is highly differentiated, as shown in the following sections.

| Network of trust
It is evident from the qualitative analysis that the clusters of organizations are formed along the often-sighted division in environmental governance: forest as a raw material for industry or forest as part of a fragile ecosystem. The following network analysis shows that neither the institutional design nor the collaborative process was able to create trust among the participants of the quarreling groups. Moreover, although the majority of FC members enjoyed high reciprocal trust, this excluded two members: EKO (the umbrella for environmental organizations) and the University of Tartu, which is represented at the FC by an ecologist. Cluster analysis of the larger policy network indicates that, in terms of reciprocal trust, there exist two major clusters: one that includes industry-related and most state organizations, and another that includes environmental organizations (Figure 2). Only the Ministry of Culture, Animal Protection Society and an NGO EcoState Estonia that promotes sustainable development form a separate mini-cluster .

| Network of learning
Cluster analysis based on the "learning" relationships largely preserves the topography of the "trust" network, although it is more fragmented (Figure 3) as the organizations are more evenly divided between three clusters. The cluster with environmental organizations now includes also Ministry of Culture. This supports the idea that cultural heritage protection in the forests is among the challengers of current forest use. Second cluster is formed by organizations representing the forestry industry, state environmental agencies and RMK. This confirms shared understandings about forest use that was indicated with the "trust" network. The third cluster incorporates four ministries (Environment, Rural Affairs, Finance, Education and Research), Government Office, Forestry School and EcoState Estonia. Both network data on learning and interview data indicate that organizations are forming distinct clusters regardless that FC includes members of different clusters. Hence our conclusion is that FC as a platform was unsuccessful in creating learning relationships between its members.
It is notable that two universities inhabit two separate "learning"

| DISCUSSION
In this article we have presented the failure of collaborative governance in the face of increasing demands for forests from economical, F I G U R E 2 Clusters of trust in Estonian forest policy network. Forestry Council membership is highlighted. The numbers denote average trust between cluster members. Arrow direction indicates how much that group of organizations is trusted F I G U R E 3 Clusters of learning in the Estonian forest policy network. Forestry Council membership is highlighted. The numbers denote the average learning relationship between cluster members. Arrow direction indicates how much that group of organizations is learnt from ecological and cultural fields. Despite the earlier processes in the 1990s, which laid the foundation for the collaborative governance of forests in Estonia, the conditions no longer favored balanced power relations and bona fide collaboration in the aftermath of the 2008 economic crisis. Changes to the legislation largely favored industry, while increased demand for other forest services (such as conservation, recreation, or cultural heritage) made the field more contested. In the face of heightened tensions, the official arenas created to facilitate cross-sectorial discussion for forestry-related policy and planning exhibited problems in their design. These arenas lacked forum exclusiveness, clear rules and process transparency (Ansell & Gash, 2008).
This resulted in a collaborative process that failed to build reciprocal trust in the policy network. Cluster analysis indicates the existence of two major clusters with high in-group and low inter-group trust: one that includes industry-related and most state organizations, and another that includes environmental organizations. We consider the lacking trust between collaboration partners as the fundamental factor that limits opportunity for successful policy processes because clusters in the "learning" network largely overlap with ones in the "trust" network. Qualitative data reveals that although issues were debated and scientific evidence was presented, no real effort was made to discuss core policy beliefs (Leach & Sabatier, 2005) or introduce common fact-finding (Ansell & Gash, 2008) by collaboratively and systematically setting up research agendas to inform policy decisions.
The destructive influence of lacking trust is best exemplified by the situation in 2016 when the proposal to lower the felling age of spruce in exchange for expanded conservation areas failed to reconcile the warring factions and, instead, led directly to action outside the formal arenas (demonstrations and the creation of the social movement EMA), marking the beginning of the "Forest War" despite the participants of collaborative governance being generally satisfied with their involvement (Teder & Kaimre, 2018).
Our analysis of Estonian forest policy sheds light on the intense relationships between knowledge, trust and collaborative policy-making. We recognize the difficult position of policy makers who have to navigate the public demands for accountability from various directions with highly contradictory values and objectives (Defries & Nagendra, 2017). In the case of Estonian forestry, pressure came from climate change, the global economic crisis, increased need for cultural and recreational forest use. Under this pressure, institutions that were intended to encourage mutual learning failed in their task. We would like to draw two theoretical conclusions based in this data.
First, our material indicates that when the pressure for resource use increases, so does demand for design and leadership of collaborative governance. With this we directly build on Ansell and Gash (2008) who draw attention to previous history of conflict or collaboration as a factor that influences demand for facilitating leadership. In common language, collaboration can refer to very different practices from simply informing someone to discovering new goals through deliberation.
When the demand for a common resource is low, this haziness of the language and procedures does not pose a problem. Previous studies of inclusion of civil society into policy making in Estonia has shown how lack of real willingness to deliberate and to adapt to the demands of stakeholders does indeed create protest, but without significant consequences (e.g., Lepa et al., 2004). In forest policy so far, Estonia has been blessed with ample resources that have been able to provide various benefits to timber industry, nature tourism, cultural use and still been able to accommodate flourishing biodiversity.
However, when the pressure for resource use rises, collaboration as a tool for informing stakeholders of pre-meditated decisions is not enough (Vento & Sjöblom, 2018). Our analysis indicates that putting collaborative governance into practice under surmounting pressure requires carefully designed processes, but more importantly a different skillset from civil servants who mediate collaboration. We can only make an educated guess that "transformative mediation" (Ansell & Gash, 2008, p. 547) is by large not a part of university curricula and must be therefore acquired through practice. We argue that these skills become the main factor that limit the possibility of collaboration as we have referred to in the title of our paper.
Second, clustering of organizations in Estonian forest policy network allows us to refine and emphasize the relationship between trust and learning. Trust is often referred to as a key factor in relation to the success of cooperative governance (Klijn et al., 2010). We would like to emphasize its importance in enabling policy learning. Dunlop and Radaelli (2013) have argued for the importance of "certification of actors" for policy learning. While this can refer to formal position of the actors involved, our analysis suggests the informal mechanisms of certification, that is, increasing reciprocal trust in actors through collaboration. Network analysis indicates that the policy makers deemed certain actors and claims as legitimate and others as not.
Importantly, two major Estonian universities occupied places in different clusters. It is obvious that universities have high formal certification in the society and the question is in the way in which these sources of information were used. Collaboration process was not able to clear doubts that some scientific knowledge is unbalanced. Different data was used in the conflict to support one or another argument instead of using it for achieving higher order solutions. Thus, formal certification was challenged by lack of trust. Tracking trust levels between pairs of organizations and the clustering of trust should be a key predictor for success of collaboration.
Also, we encourage seeing the conflict that erupted in the policy field from a positive side as it opened the policy network up for new actors. The number of organizations involved in the next FDP now exceeds 30 and includes new participants, such as the Nature Tourism Union. Even the social movement EMA was, after a months-long debate, included in the ongoing process.
Finally, we draw attention to emerging literature of network methods for assessing decision-making networks (Cvitanovic et al., 2017). Our study adds to these methodological contributions by connecting them to network measurements that relate directly to theories of collaborative governance. Future studies should track the longitudinal developments of the network measurements and track the shifting topology over time.

| CONCLUSION
Our analysis suggests that although cumulative measurements or assessment of generalized trust in a collaborative process (Klijn et al., 2010;Leach & Sabatier, 2005) may be high, collaboration can nevertheless fail to achieve its goals. Growing demands for finite resources highlight the limits and possibilities of deliberative democracy. When inter-organizational trust is lacking, cooperation may be impossible because it renders trusting different knowledge sources difficult. Therefore, we suggest that relational trust should be closely monitored. It is possible that absolute consensus in forestry policy is indeed practically unachievable (Peterson et al., 2005), as scientific findings can be contradictory. But there is a fine line between haphazard collaboration and agonistic pluralism (Mouffe, 1999). Our analysis stresses the importance of clear and powerful institutions for common knowledge production that are able to sail the muddy waters of ecosystem governance. When collaborative governance is unable to adequately address tensions and conflicts in the network despite the well-crafted institutional design, mediated negotiations (Forester, 2006(Forester, , 2009 or other forms of intermediaries  may be suitable for articulating needs, aggregating knowledge and creating institutional support for collaborative governance. However, conflicts in the forest policy network should not necessarily be seen only from a negative point of view because this could drive policy network towards renewal and change.