Indonesia, as an archipelago located on the equator, is prone to natural disasters. According to the National Disaster Management Agency (Badan Nasional Penanggulangan Bencana – BNPB), until mid-2018 a total of 1134 natural disasters had been recorded in Indonesia. These included tornadoes, landslides, floods, earthquakes and volcanic eruptions. Volcanic eruptions are frequent in Indonesia. This is not surprising because Indonesia has a record of 129 active volcanoes, 70 of which are considered very dangerous for the surrounding community (BNPB 2010).

Mount Merapi is a stratotype volcano with a height of 2980 m above sea level. Geographically, it is located at a latitude of 7°32.5′ South and a longitude of 110°26.5′ East. Administratively, it is located in the Sleman Regency in the Special Region of Yogyakarta and Magelang Regency, Boyolali Regency and Klaten Regency in Central Java Province. Mount Merapi has erupted numerous times. According to data from BNPB (2019), in the past few years, Mount Merapi has erupted approximately 40 times, with seven of these eruptions being significant. From 1768 to 1872, Mount Merapi has erupted more than 80 times, with major eruptions occurring in 1768, 1822, 1849 and 1872. Eruptions in this century have been more violent than were those in the 20th century, with hot clouds of ejecta reaching 20 km high. In the 20th century, there were at least 28 eruptions of Mount Merapi, with the most massive one occurring in 1931. There were fatalities in this incident; the damage increased to 13 villages, and 23 other communities were highly damaged. In 1933–1935 and 1961, eruptions caused lava flows. After that, in 1994, Mount Merapi ejected by breaking down the lava dome with a volume of 2.6 million m³. In that eruption, Mount Merapi spewed hot clouds of ejecta more than 6.5 km to the northwest and south, affecting 64 people badly and causing injuries. An eruption in 2006 destroyed the Kaliadem area, with two volunteers being killed by the hot cloud of the eruption. This eruption was marked by earthquakes and deformation of the landscape, followed by a rain of volcanic ash for 3 days in Magelang and Sleman, Central Java. The eruption of Mount Merapi in 2010 was the largest in the past 100 years. In mid-2018, Merapi began to exhibit increased activity, which continued until the end of January 2019. Merapi’s activity is still being monitored to minimise the danger it poses. Disasters require decentralised decision-making and intensive human interaction (Kapucu & Van Wart 2006; Kirschenbaum 2004). Managing disasters involves dynamic processes that are ideal yet demanding. Thus, collaboration between organisations and government agencies is essential for the development of an effective strategy and better performance during disasters.

Aside from the fact that collaboration often occurs between proximal and like agencies (Simo & Bies 2019), collaborative disaster management faces various challenges, which often leads to failure of the response operations. Poor communication, inadequate planning, misguided and poorly executed leadership and insufficient coordination with various stakeholders lead to collaborative failures (Streib & Waugh 2019; Wise 2019). Jatmiko and Tandiarrang (2014), in their study of the Indonesian Maritime Agency, found that the existing structure of the agency does not support good communication with agencies that are crucial to the agency’s performance. Meanwhile, Seng (2010) argued that the polycentric structure of Indonesian disaster management is ideal in responding to occurrences of tsunamis in the country. However, it does not fit with the norms of the Indonesian political community. Moreover, Nurmandi et al. (2015) studied different disasters in Indonesia and concluded that different structures are formed in the governance of each of the disasters they studied.

Emergency responses in Indonesia are often erratic, especially in terms of searching for and rescuing people and in coordinating both the collection and the distribution of aid to victims. Post-disaster recovery efforts are also not maximised. The inaccuracy of data in terms of the numbers of injured and the types of injuries suffered by victims makes it challenging to allocate the medical facilities needed for efforts to restore the health of victims. The complexity of the problem requires proper regulation, and careful planning of the response to disasters is needed to ensure that the implementation of disaster management will be more focused and integrated. Given the above problems, the Indonesian government’s implementation of regulations related to disaster management has not been optimal. This means that countermeasures that are taken do not follow systematic and planned steps. This study investigates the effect of cross-sector collaboration in disaster management by investigating the way in which collaborative governance is carried out by the government and non-governmental organisations in managing the Mount Merapi Eruption disaster in the Special Region of Yogyakarta.

For public sector organisations, the entire network approach is used widely as an analytical framework because it examines the connections that either exist or do not exist in a defined set of organisations, and this shows the extent to which organisations work with one another to achieve common goals (Provan & Kenis 2008).

Provan and Kenis (2008) examined multilateral relations that define the entire network and that are essential for achieving collective results. They suggested that the entire network must be analysed based on network governance, network leadership and management, and network performance. According to them, network governance refers to a network of coordination mechanisms and focuses on the network as a unit of analysis to guide the network in a steady state. There are three modes of network governance. The first mode is, shared or independent governance, which is characterised by easy form and high-level commitment but involves frequent meetings, lacks clear goals and experiences difficulties in reaching consensus. The second mode is leading the organisation, which refers to the efficiency of direction and precise network management but faces the potential for leader organisation dominance and low participation from members. The third mode is network administration organisations, which are entities that manage the networks but come with higher operating costs, more complex administrative processes, and the potential loss of control and decision-making authority for some network members. Meanwhile, Jung, Mazmanian and Tang (2009) suggested that leadership and management follow a framework that guides leaders and network managers between organisations in networks, irrespective of their chosen governance model.

Bryson, Crosby and Stone (2006) postulated that the structure of collaborative action changes and tends to be flexible because of membership ambiguity and complexity of the local environment. Such ambiguity arises from membership features, including perceptions about who is involved in the collaboration and what these members represent. Also, collaboration hierarchies, where individuals and organisations often become members of overlapping partnerships, further exacerbate the ambiguity of membership. Instead, governance between networks determines the survival and success of networks and collaboration. Bryson et al. (2006) viewed governance, characterised by initial agreements, leadership, planning, trust and conflict management, as a set of coordination and monitoring activities necessary for the network to survive. Governance is highly dependent on network structure, and as Bryson et al. (2006) emphasised, the choice of the type of governance structure tends to influence network effectiveness. The findings of Bryson, Crosby and Stone (2015) suggest that agreement is reached if public managers adopt an inclusive process that is made possible by flat structures.

Conversely, Dalton et al. (1980) explained that flat and tall structures pertain to the number of hierarchical levels of the organisation, where the span of control for the tall structure is narrower, whereas the span of control for the flat structure can be more comprehensive. They concluded that the organisation is structured in such a way that it fits its intended functions. And, therefore, the structure of the organisation may vary, but it may remain within a reasonable range in which there will be no difference in performance attributable to structure (Dalton et al. 1980).

This research uses a mixed method. The combination method combines two types of research methods: quantitative analysis and qualitative research. In combined quantitative and qualitative research, the two methods are not applied simultaneously, but they can be applied interchangeably. Hence, in this study, the quantitative analysis uses research instruments, and the qualitative analysis is a descriptive research done to obtain a detailed picture of post-disruption disaster management. The analysis performed is regression analysis using structural equation modelling through partial least squares (PLS–SEM); this is used to test the relationship between variables. The study involved 100 respondents from 28 institutions, practically representing the entire population identified for this study. Data validity depends not only on the number and characteristics of respondents but also onthe type of questions asked. When the questions for this study were tested, the results revealed that the number of statement items contained 50 of 7 variables (initial agreement, trust, previous performance, managing conflict, leadership, planning and existing relationship) and the value of Cronbach’s alpha for the analysis was 0.912. This value implies that the questionnaire used in gathering primary data is valid and reliable.

Analysis of the outer model ensures that the model with its variables and indicators is worth measuring, considering that this is valid and reliable. Thus, the external model analysis measures construct validity using convergent validity and discriminant validity as indicators.

As shown in Table 1, the highest heterotrait–monotrait ratio of correlations (HTMT) value among the variables is 0.772. Therefore, lower than the value of 0.85 or 0.90 is set. Also, the same findings apply to HTMT inference criteria that are defined by running a bootstrap routine. The bootstrap routine shows variable values below the confidence interval, and the confidence bias interval is corrected. All values differ significantly from 1. Therefore, discriminant validity is established for the outside models used in this study.

As revealed in Table 2, the structural model has a low predictive accuracy based on the values in Table 3. Furthermore, the R² value of the existing relationship initial agreement and managing conflict is considered unacceptable. And that indicates that the independent variable (previous performance) cannot explain the endogenous dependent variable. However, R² values are usually low in studies related to human behaviour and human relations with other humans because human behaviour is more difficult to predict (Sorrels et al. 2018). Therefore, even though the predictive capacity of structural models is low, important conclusions about the relationship between variables can still be derived from statistically significant predictors. As Paul (2017) said, a regression model with a low R² value can be very good, for several reasons. Therefore, (high or low) R² is not enough by itself.

Meanwhile, the effect size (f2) of the variables is also used to assess the structural model. The effect size (f2) values of 0.02, 0.15 and 0.35 for the significant independent variables represent weak, moderate and substantial effects, respectively (Chin 1998). The effect size of the hypothesised relationship between variables in this study, along with the p-value of the regression analysis.

The findings revealed that the previous performance of the network is crucially associated with the initial agreement, leadership and managing conflict, which validates the theory of Bryson et al. (2006). That sector failure facilitates cross-sector collaboration in terms of improving the initial agreement of the network as a way of making up for the shortcomings of particular sectors. The findings of Olu and Adesubomi (2014) are also supported as they stressed that a conflict management system based on the previous performance ensures a conducive environment in the process of collaboration. Lessons from sector failure serve as the basis for increasing informal and formal agreements in collaboration as well as enhancing the motivation mechanisms available (Bryson et al. 2015). Moreover, leadership in the network has a greater potential to evolve. Poor communication, misguided and poorly executed leadership by the federal and state government, and insufficient coordination with various stakeholders, as well as inadequate preparation among communities, lead to collaborative failures. And this study leads to the analysis of factors influencing the behaviour of public managers, such as the nature of the task performed and motivation mechanisms institutionalised in the network.

Moreover, the findings indicated that as leadership is significantly associated with the motivations in the initial agreement, leadership expands trust in the governance process. Generally, trust refers to a person’s confidence in the reliability of another person concerning certain outcomes, whilst the shared confidence held by the members of an organisation is called inter-organisational trust (Rashid & Edmondson 2011). The interdependencies amongst agencies and organisations through interactive processes such as face-to-face dialogues increase trust, build social capital and can develop into collaborative culture, which substantially increases the speed of decision-making and leads to fruitful collaborations (Ansell & Gash 2008; Emerson et al. 2011; Jung et al. 2009; Kapucu, Arslan & Demiroz 2010).

Significantly, the data show that trust in terms of dependability and competence is associated substantially with collaborative planning. The finding suggests that leadership enhances the reliability and expertise of the network, and trust enriches the planning process and output. These findings do not fully agree with the conclusions of Lester and Krejci (2019). Those findings postulated that the planning process should be participated by the leaders of the institutions involved in the disaster operations to ensure the successful result. However, this research supports the findings of Kapucu and Van Wart (2006) in the disaster response. Operations during the World Trade Centre attack, Hurricane Andrew and Katrina, when they postulated that the problems on weak or non-existent planning come along with incompetent managers. Thus, this research suggests that with good leadership, trust enhances the planning process of the network.

Given these findings, this study enriches the current understanding of cross-sector collaboration, which Bryson et al. (2006) referred to as an ideal but difficult and complicated approach towards the successful outcome. With its focus on the influences of the initial conditions to the aspects of governance process-leadership, initial agreement, trust, planning and managing conflict and its impact on the outcome of collaboration, this study reaffirms the previous studies conducted on cross-sector collaboration and disaster governance emphasising the relevance of the aspects of governance processes, particularly leadership in collaborative disaster management (Fung 2015; Jovita 2010; Kapucu et al. 2010; Lester & Krejci 2019). This study also joins the theoretical discussion on the relationship between the impact of the initial condition to the collaborative process that institutional design and sector failure, sets the basic ground under which collaboration takes place (Ansell & Gash 2008).

This study further reinforces the findings of Chang Seng (2010) that a structure may be ideal, but that does not necessarily imply that it is suitable for every society because factors such as social norms and political culture may prove to be deterrents. This finding also confirms the study of Kapucu and Van Wart (2006), which shows that decentralised decision-making with excessive dependence on centralised authority can cause more harm than good, especially if the authority is not fully committed to addressing needs and solving various problems. There will be challenges along the way. This study enriches the existing understanding of cross-sector collaboration, which Bryson et al. (2006) referred to as an ideal but difficult and complicated approach towards a successful outcome.

Practically, this research implies that at the national and regional levels, where many organisations are part of the network, centralised decision-making is needed and disaster operations must be decentralised (Kapucu 2005). However, shared governance must be fostered in local government units where the network has relatively few members and very dense relationships can be established (Gheblawi et al. 2020; Provan & Kenis 2008), more importantly, a disaster management ambassador, known in Indonesia as Pejuang Sigap Bencana, who is committed to serving the public. Such an ambassador should adapt to changing operating conditions and ensure that the heads of institutions attend collaborative activities (Rai & Khawas 2019). Also, tensions can be avoided in implementing sufficient skills.