Introduction

The research presented here analyses how managers in central government agencies describe organizational interaction within the public sphere and across the public–private divide. Central government agencies are permanent, public organizations that are formally separated from their parent department, staffed by civil servants, financed mainly via the national government budget, and subject to legal procedures (Pollitt et al., 2004). Central government agencies contribute to the ‘pipe drains’ or ‘silos’ that characterize many contemporary national executives. The degree of specialization is high in this type of organizations, but their status can vary in terms of formal role and de facto autonomy (Yesilkagit & Van Thiel, 2008).

The aim of this research is to study the relationship between formal and organizational safeguards for agency autonomy on the one hand, and observed variation in how organizational interaction is described, on the other. In theory, organizational interaction can be more or less costly for the interacting parties, and cost is thus a determinant of interaction. For example, when performance management regimes make it necessary for an organization to concentrate on its own goals to be perceived as performing well, interacting with others will constitute a cost (Lægreid & Rykkja, 2015a; Molenveld et al., 2019, 2021). The theoretical frame of this research focuses on autonomy as a factor of potential importance for the perceived costs associated with organizational interaction. An assumption is that attitudes towards interaction are in part based on cost–benefit analyses. What these calculations look like, and what to expect in terms of attitudes or behavior, is a debated issue. The results from this research will be interpreted from within two competing perspectives. The first argues that there is a negative relationship between government agencies’ autonomy and their propensity to interact with others (cf. Wilson, 1989). This argument is supported by Bjurstrøm Hagen’s (2021) study of Norwegian government agencies, where a main conclusion is that the higher the policy autonomy of an agency, the less an agency coordinates with others. From within the second perspective, autonomous agencies can identify possible gains from interacting. This gives them a nuanced outlook on interacting with others, which in turn makes the outcome less evident (cf. Busuioc, 2016; Heims, 2019). In line with this latter perspective, the research presented here finds that strong formal and organizational safeguards of agency autonomy are likely to produce positive effects on organizational interaction.

The research design is comparative, including two Nordic countries, Norway and Sweden. Norway and Sweden both feature government agencies, but differ when it comes to vertical control and fragmentation of respective national executive. The Norwegian national executive is less fragmented than the Swedish one, the latter features a strong tradition of administrative dualism and prohibits ministerial rule. Importantly, the formal status of central government agencies differs between the two countries, with a stronger constitutional protection of the Swedish agencies’ autonomy (Balle-Hansen et al. 2012).

The data source used is the Coordinating for Cohesion in the Public Sector of the Future (COCOPS) Executive Survey on Public Sector Reform in Europe (Hammerschmid, 2015). The present study uses a sub-sample of this survey. This sub-sample represents only managers in central governmental agencies (and hence does not include respondents working in ministries), in Norway and Sweden. Compared to earlier studies of top managers’ views, and Nordic managers views on ‘coordination’ (cf. Christensen et al., 2015; Greve et al., 2016; Lægreid & Rykkja, 2015b) the design of this study is different in its empirical scope.

The next section is dedicated to the analytical framework of the study. Research design, methodology and data are then discussed, followed by a presentation of the main results. In the concluding section, these results and possible lessons are discussed.

Analytical Framework

Organizational Interaction

As discussed thoroughly by McNamara (2012), it is possible to distinguish between different types of organizational interaction, and create fine-grained categories. In essence, the differences between interaction types are a matter of both form and degree. For example, in McNamara’s terms, cooperation does not require organizations to work outside their own structures or to pool resources with others. Cooperation thus allows organizations to retain their autonomy versus other organizations, and does not require organizations to solve turf-related conflicts. In contrast to cooperation, both coordination and collaboration require organizations to give up some autonomy, albeit on different scales. The design of interactions, degree and formality of relationships, pooling or exchange of resources, and so forth are different in coordination and collaboration, respectively. While coordination is designed to create linkages between existing structures, collaborative interaction features collective efforts to set up new structures. Coordination makes the already existing, participating organizations dependent upon each other. In collaboration, power needs to be invested in the new structures that have been created jointly to solve a problem (McNamara, 2012). Thus, collaboration is the most demanding type of interaction, requiring participants to pool resources and give up their autonomy to a new structure created to manage the interaction.

Of the three types of interaction discussed in the research presented here, coordination is dominating the research field. Coordination takes place when public organizations actively align their actions across organizational borders (Christensen & Lægreid, 2007; Peters, 2018; Verhoest et al., 2007). Some categorizations of, again, coordination rather than organizational interaction, build upon the aims of the measures taken, distinguishing between negative (avoiding conflict or overlaps), positive (working together on mutually beneficial solutions), or strategic (designing policy programs) coordination (Peters, 2018). Moreover, research on coordination includes several institutional perspectives such as structural perspectives focusing on formal, top-down attempts to coordinate organizations and informal, negotiated orders in which organizations work together. Other perspectives focus on cultural-institutional aspects such as the organizational values and norms that support or inhibit interaction, or coordination as a myth among other public management myths (Christensen & Lægreid, 2007; Christensen et al., 2015; Lægreid & Rykkja, 2015a).

Interaction can take place between organizations operating within or across territorial borders. Christensen and Lægreid’s distinctions between vertical/horizontal and external/internal coordination, respectively, describe such patterns of interaction, providing a categorization of different types of organizations and their relationships (Christensen & Lægreid, 2008, 102).

Two Perspectives on Autonomy and Organizational Interaction

To say that the propensity to interact with others, and attitudes towards interaction, are based on cost–benefit analyses is only a first step in theorizing about possible outcomes. The results of this study will be interpreted using two competing perspectives. According to the first, the higher the level of autonomy a government agency has, the less it will engage in interaction with others (cf. Bjurstrøm Hagen, 2021; Wilson, 1989). From within this perspective, the relationship between autonomy and interaction is straightforward: when an agency is autonomous enough to decide for itself, it will not engage in interaction with others. In terms of different types of interaction, cooperative interaction is probably more common due to its relatively low costs, than both coordination and collaboration. Collaboration, which entails investing in new structures and giving up some autonomy, should according to this perspective be the least common strategy due to the high costs involved. Moreover, interaction internal to the public sphere is expected to be more common than interaction that involves actors that are external to the public sphere. The most common interaction is expected to be when public organizations at the same level (internal/horizontal) interact. Research shows that organizations with the task of accomplishing organizational interaction prefer not to be coordinated (or controlled) from above – that is, vertically (see e.g. Molenveld et al., 2019), but also that portfolio ministers can make agencies engage in cross-cutting activities (Molenveld et al., 2021).

According to the contrasting perspective possible gains from interacting with others makes the outcome for autonomous agencies less evident. For example, agencies can conclude that turf protection is facilitated by resources that are made available when organizations interact (Heims, 2019). One of the resources at stake in interaction is agency reputation, and interacting with others can be perceived as a possibility to build reputation that will benefit the agency (Busuioc, 2016). In line with this perspective, a positive relationship between autonomy and interaction is expected, especially when strong formal and organizational safeguards of agency autonomy lower the risks of losing turf or mandate to other organizations.

Methodology

The Comparative Design

The country comparison in this study includes Norway and Sweden, two countries that are similar in size, political traditions, and culture, and with national executives that feature a substantial number of government agencies (Greve et al., 2016; Kuhlmann & Wollmann, 2019; Balle-Hansen et al. 2012). However, Norway and Sweden differ in terms of national executive fragmentation and the formal autonomy of their respective government agencies. Norway applies the doctrine of ministerial responsibility, while the Swedish system is a dual executive system in which ministerial rule is prohibited (Ahlbäck Öberg & Wockelberg, 2016; Greve & Ejersbo, 2016). These differences makes a comparison between the two countries interesting, as it has been suggested that, besides organizational culture, the vertical control exercised by ministers explains public organizations’ propensity to engage in policy programs that demand coordinated action (Molenveld et al., 2021). In addition, the formal status of central government agencies differs between the two countries. While the independence of agencies is rarely formally codified or protected in Norway, it is in the Swedish case (Balle-Hansen et al. 2012; Ahlbäck Öberg & Wockelberg, 2015, 2016). Organizational and constitutional factors make Swedish government agencies’ autonomy unusually high, and importantly, higher than in the Norwegian case. The comparison between the two settings is motivated by these differences. In addition, and in comparison with other Nordic countries, Norway has been described as ‘leading in whole-of government efforts’ (Greve & Ejersbo, 2016, 50; see also Lægreid & Rykkja, 2015b).

The arguments for this comparative design are valid given that this study focuses exclusively on the government agency-level. Earlier studies of the Nordic group of countries find weak, if any, evidence of cross-country variation (Greve et al., 2016). In terms of managers’ role perceptions, Norwegian and Swedish national public managers give equivalent descriptions of having a collaborative role (Virtanen, 2016). However, these earlier studies have a different empirical scope, they are based on aggregated, country-level, survey data that possibly can hide differences that will appear as data is disaggregated.

Data and Analytical Methods

Descriptive statistics (compare means) and OLS regression analyses are applied to analyze self-reported data on organizational interaction. The COCOPS dataset contains self-reported perceptions from managers in ministries and governmental agencies in 21 European countries. This dataset has been used in studies of for example public management reforms, performance, different types of autonomy, politicization, and managers’ values and motivations (see e.g. Greve et al., 2016; Hammerschmid et al., 2016, 2019; Jeannot et al., 2018; Van de Walle, 2019). As self-reported data is used to measure all variables in the analyses, this section starts with a discussion of the risk of creating common source bias. This type of risk is high when self-reported perceptions and attitudes from a single survey are included as dependent and independent variables in the same model (Kelman, 2015). In the present study, the aim is to minimize the risk associated with including several self-reported attitudes in the same models by reserving this type of data to the measurement of dependent variables. Although the control variable, organizational position, is also self-reported, this is not a significant problem: organizational position is a factual statement that is unlikely to be associated with attitudes, beliefs, or perceptions (Table 1).

Table 1 Survey items used and their operational definitions
Full size table

The sub-sample used in this research contains data from Norway and Sweden (the response rates from the two countries were 27.9 and 40.4 percent, respectively; see Lægreid & Rykkja, 2016) and exclusively includes respondents from central government agencies. These are identified by question 1 in the survey, ‘What type of organization do you work for?’ and the alternative ‘Agency or subordinate government body at the central government level’ (Hammerschmid, 2015).

The survey question, ‘To resolve coordination problems when working with other organizations, we typically… (1 = Strongly disagree, 7 = Strongly agree)’, is used to study a range of organizational interaction types. This question asks for ways to solve coordination problems, and the sub-items represent measures of cooperation, coordination and collaboration. Sub-items 13_1 and 13_2 both describe a situation in which coordination problems are referred somewhere else, either upwards in the organizational hierarchy or to political decision-makers. Referrals can be perceived as associated with, in relative terms, limited costs, as the measure entails moving the issue to someone else’s agenda. Moreover, referring issues upwards but within an organization is less costly than referring issues to political actors and bodies, outside the agency. The latter type of referrals is, at least in theory, invitations to be controlled vertically from above, which is costly in terms of autonomy loss. Sub-items 13_3, 13_4, 13_5 and 13_6 all describe demanding and costly measures; in McNamara’s terms (see above), they require the organizations to pool resources and/or give up parts of their autonomy. The first three of this group entail collaboration, as new structures are created to enable interaction. The last (13_6) entails giving up autonomy to a lead organization, this does not necessarily count as the creation of new structures, and is thus here defined as a coordinative type of measure, but it is indeed a cost in terms of autonomy loss. Finally, sub-items 13_7 and 13_8 ask about consulting with private sector groups or relevant experts, respectively. A consultation entails an exchange of information, but does not require pooling resources, solving turf issues or investing in new structures. Consultations are thus cooperative measures, and as such relatively uncostly. Finally, these sub-items offer examples of interaction both with organizations that are internal to the public sphere and with external actors.

To study agency managers’ perceived quality of organizational interaction, survey item 11 is used. This item invites the respondents to evaluate collaborative activities that go on within their own policy field by passing judgements ranging from “Very poor” to “Very good” on a seven-grade scale. According to the theoretical definition used here, collaboration is a demanding, costly type of measure that entails creating new joint structures, giving up some organizational autonomy. The sub-items offer information about a range of possible collaboration. Sub-items 11_1 and 11_2 describe relationships that are internal to the national public sector. Sub-item 11_3 captures vertical interaction across public sphere organizations at the national and sub-national levels. Vertical interaction is also described in sub-item 11_4, which asks about national government bodies’ collaboration with international and supra-national organizations. Sub-item 11_5 describes interaction with organizations or actors that are external to the public sector, such as NGOs or scientists.

The statistical models will include individual sub-items for interactive measures and their quality. The alternative strategy, to constructs indices, proved to be unsuitable as attempts to do this did not report dimensions on high enough levels to guarantee reliability (Table 2).

Table 2 Descriptives
Full size table

Controls

The dataset is constructed to enable a comparison of central government agency managers in Norway and Sweden. This comparative design is, for technical reasons, somewhat limited in terms of the possibility to include control variables into the statistical models. For example, as the number of observations per policy area is too few, and slightly skewed, including this variable does not generate any robust results. Similar problems makes organization size – measured as number of employees – unsuitable as a control variable since the differences in national executives also is reflected in the fact that Sweden has larger government agencies than Norway (Difi, 2013; Statskontoret, 2021).

There is enough data available to control for the agency managers’ position. The significance of distinguishing between different levels of managers within organizations has been identified in earlier research (cf. Bach et al., 2020a, b; Egeberg & Stigen, 2018; Enticott et al., 2009; Jeannot et al., 2018; Molenveld et al., 2019). Organizational interaction measures, such as coordination, are ‘magic concepts’ (broad concepts with positive connotations, imply consensus rather than conflict and are fashionable in wide circles) (Pollitt & Hupe, 2011). Thus, it is likely that respondents to the survey in general express positive views. However, it is likely that managers at different levels of the agencies have different views. Heads of organizations have been found to express more positive views of their organization than members of other echelons do. One possible reason behind top executives’ particularly positive attitude is their role as the primary advocate responsible for the organization (Enticott et al., 2009). In the specific terms of this study of agency managers’ views on organizational interaction, it is likely that top managers express different, possibly more positive, views than lower level managers. Earlier research has suggested that cultural perspectives can help to identify causal mechanisms associated with the formal organizational position of agency managers (cf. Christensen et al., 2015; Lægreid & Rykkja, 2015a, b). Agency DGs arguably belong to a distinct, exclusive elite who have a mandate that is not only oriented towards agency leadership, but also entails close contact with the ministerial level of the national executive. In comparison, lower level managers are oriented more towards the organization and its operation. It may be that agency heads interact with others on a rather abstract policy level, and that they are thus not exposed to the practical problems and ‘nitty gritty’ of establishing and upholding inter-organizational processes.

To categorize agency managers, we use the question ‘What kind of position do you currently hold?’, where the possible alternatives allow us to distinguish between three hierarchical levels (Hammerschmid, 2015). The top level represents, for example, DGs in government agencies, while the second and third hierarchical levels represent managers in charge of lower levels of the organization, such as divisions and units. In the analyses, the variable of ‘managerial position’ is dichotomous, comparing agency DGs with a composite of layer 2 and layer 3 managers.

Results

This section starts with analyses of agency managers’ descriptions of common ways of solving ‘coordination problems’. Table 3 presents the results as calculated means for agency managers in Norway and Sweden, respectively. The most common strategy used is to refer issues to higher levels in the hierarchy. This activity is here interpreted as referring issues within the organization, a measure that is in relative terms not costly. On the latter note, it is interesting that referring issues to political actors and bodies is less common than sending referrals upwards in the internal organizational hierarchy. This result makes sense if the former type of referrals is perceived as exposing the organization to top down coordination. In that case, this finding is in line with earlier research showing that those responsible for organizational interaction prefer not to be coordinated from the top and down (Molenveld et al., 2019). This result also indicates that, even though political decision-makers are commonly in a position to apply holistic views on policy, coordination problems are not always brought to their attention.

Table 3 Country means for different types of coordination strategies
Full size table

It is in relative terms common to set up a work or project group that includes crossing organizational borders. Strategies that entail creating special-purpose bodies or appointing a lead organization are uncommon in both countries. This is expected, given the high costs of such strategies from the agencies’ point of view: these measures are associated with pooling resources and/or handing over autonomy to another organization. In both countries, permanent new bodies are less frequent measures than temporary new structures for interaction. Finally, we find it interesting that consulting with experts or private actors is reported to be relatively uncommon. Consulting entails asking for or exchanging information, and is an uncostly measure according to our theoretical understanding. A possible interpretation of this finding is that government agencies rarely consult with private actors. Here, then, we observe the relevance of an internal/external dimension in our understanding of interaction.

Comparing the results reported in Table 3 across countries, differences are small as managers in both countries give similar descriptions of the relative importance of the measures applied. To set up a temporary crosscutting project is more common in Norway than in Sweden, while crosscutting policy arrangements or programs are more common in Sweden than in Norway. To appoint an organization to lead some type of interaction is less common in Sweden than in Norway. Finally, the habit of consulting experts is less common in Sweden than in Norway.

Country differences surface when regression analyses are used to explain the observed patterns (see Table 4). The regression models include dichotomous variables for ‘country’, and for the control variable ‘managerial position’, respectively. The main result is that the ‘country’ variable produces significant results for most of the analyzed problem-solving strategies.

Table 4 Typical ways of solving coordination problem, OLS regression results
Full size table

As the dichotomous ‘country’ variable is coded with Norway = 1 and Sweden = 2, the coefficients reported for this variable (Table 4) represent the effect of ‘being Sweden’. A large positive and statistically significant effect (B = 1.182***) is reported for solving coordination problems by setting up a policy arrangement/program that crosses organizational borders; this result means that this strategy is clearly more common in Sweden than in Norway. Other statistically significant country differences can also be observed, albeit smaller ones: in comparison with agency managers in Norway, fewer Swedish agency managers report that coordination problems are solved by setting up temporary project groups (B = -0.493***), appointing a lead organization (B = -0.442***), or consulting experts (B = -0.546***).

In theoretical terms, it is interesting that many of the potentially costly interaction measures are more common in the Swedish context than in the Norwegian one. Managers in Swedish central government agencies are more likely to report interaction that entails permanent new structures as well as policy arrangement or programs. They are also reporting more referrals to political actors, a measure that could result in vertical coordination and thus autonomy loss. These results contradict the claim that the more agency autonomy, the less interaction. They also lend support for the competing perspective where agencies make nuanced analyses of cost and benefits, not necessarily shying away from advanced collaborative arrangements. The results suggests that it is the strong safeguards of the autonomous mandates of Swedish agencies that lower the risks involved in interacting with others in collaborative ways. The result that appointing lead organizations is more common in Norway than in Sweden is however in line with the idea of a negative relationship between autonomy and interaction, as are the results reported for external consultations.

Turning now to the issue of the quality of interaction, a first observation regarding the results (means) reported in Table 5 is that the respondents in general give highly positive evaluative marks. Regarding collaboration within the national public sector (i.e. internal interaction), managers are most positive in their evaluation of collaboration within their own policy area and less so when it to comes to collaboration between different policy areas. Vertical collaboration between national and sub-national bodies is evaluated in more positive terms by Swedish managers than Norwegian ones. In contrast, vertical collaboration between national and international bodies is given higher scores by managers in Norway than by their Swedish counterparts. Moreover, collaboration with private actors external to the public sector is described in positive terms. This is interesting, since consulting with private stakeholders, as a strategy to solve coordination problems, was described as being rare (as reported above). Thus, interaction with external actors appears to be rare yet perceived as well functioning. For most types of collaborative relationships, Swedish managers are more positive than Norwegian ones.

Table 5 How would you characterize collaboration in your own policy field between… (means)?
Full size table

The regression analyses presented in Table 6 show that the variable ‘country’ has strong and significant effects on the perceived quality of different types of collaboration. For all types of collaboration, except interaction between national and supra-national government bodies, Swedish respondents report a more positive perception than their Norwegian counterparts do. A positive judgement passed on collaboration is a cost–benefit analysis with a positive conclusion. From within this perspective, we conclude that Swedish managers find collaboration to be a more positive thing than their Norwegian colleagues do. On the one hand, this result further strengthens our interest in the suggestion that strong safeguards for agency autonomy lower the risks (costs) for organizational interaction. On the other hand, the respondents are evaluating interaction from within a more general perspective than one that focus on their own organization. It is also possible that the Swedish and Norwegian respondents respectively have different experiences from collaborating, at least when it comes to collaboration between national and supra-national bodies which possibly means something different in the EU member state Sweden than it does in Norway. Finally, the regression analyses presented in Table 6 confirm the importance of controlling for managerial level, as this control variable has independent strong, negative and significant effects on all dependent variables except the perceived quality of collaboration between national and local/regional government bodies. This result means that agency DGs find the quality of collaboration to be higher than managers operating at lower levels do; the same result is found when controlling for country. The largest effects of managerial level are reported for the quality of collaboration between national and supra-national government bodies, and for the quality of collaboration with private stakeholders, respectively.

Table 6 Quality of collaboration, OLS regression results
Full size table

Conclusions

The aim of this research was to study the relationship between formal and organizational safeguards for agency autonomy on the one hand, and observed variation in how organizational interaction is described, on the other. To this end, Norwegian and Swedish agency managers’ views on organizational interaction are compared.

A first set of empirical analyses concerned what measures that are adopted to meet needs for organizational interaction in central governments. The focus was on types of organizational interaction that vary in terms of how costly they are from the participants’ perspective. The descriptions are in this part relatively similar across both countries and managerial levels, which give them a certain weight. One conclusion from this part of the study is that organizational interaction that involves administrative actors horizontally within the public sphere is more common than interaction that entails transferring the initiative (power) vertically, i.e. to political actors. Setting up project groups, policy arrangements, or programs was common, while appointing another organization to lead an interactive arrangement – a high cost measure – was rare. Moreover, inviting coordination ‘from above’ was also rare. This means that political actors only to a limited extent are called upon to enable organizational interaction or solve coordination problems. This finding confirms earlier research showing that coordination measures that are controlled from the top down are not appreciated by those with the task of implementing them (Molenveld et al., 2019, 9f). Consulting with private stakeholders was reported as being uncommon, even though consultation is an uncostly measure. In sum, the respondents describe limited interaction across internal and external organizational borders. This is perhaps not surprising. The literature on integrated policy strategies describes failed policies rather than stories of success, and identifies a large number of constraining factors, such as poor design, sectoral priorities, lack of resources, and lack of political support (Candel, 2017).

A second set of analyses concerned the quality of interaction, in this part collaboration, between different sets of actors. The managers were in general positive in their evaluations. Collaboration with organizations operating within their own policy field was perceived as being of the highest quality, in comparison with all other types of collaboration. Interestingly, a high rating was also given to collaboration with private stakeholders. Hence, although this type of external interaction was described as rare, it was perceived as being of high quality. Collaboration across policy fields and vertically with public actors on the local and supra-national levels received relatively low ratings.

A third set of analyses had an explanatory focus. In theoretical terms, the most interesting result from this part of the study is that country had a significant effect on the type of measures that were applied, as well as on perceptions of quality. The two countries in this study represent different levels of agency autonomy, as Swedish central government agencies have stronger constitutional safeguards for their autonomy country than the Norwegian ones. The Swedish national executive is described as highly fragmentized due to its model of executive dualism and lack of ministerial rule. The regression analyses reveal that being a Swedish agency manager increases the propensity to report interaction, including crosscutting policy arrangements or programs and permanent engagements. A significant positive effect was also observed as Swedish respondents more frequently described referrals to political actors, a measure that could result in vertical coordination and thus autonomy loss. These results contradict the claim that the more agency autonomy, the less interaction (cf. Bjurstrøm Hagen, 2021; Wilson, 1989). There is however support for the competing perspective where autonomous agencies make nuanced analyses of cost and benefits, not necessarily shying away from advanced collaborative arrangements (Busuioc, 2016; Heims, 2019). It is possible that the unusually strong safeguards of the autonomous mandates of Swedish agencies affect the cost–benefit analyses of interacting with others in a positive way. In more general terms, the more stable the organizational boundaries are, the lower the risk of losing control over turf or mandate for those who engage in interaction with others. The indications of a positive relationship between strong safeguards of agency autonomy and organizational interaction rest on a structured country comparison, a different empirical strategy than the earlier contributions that are referenced here. The relevance of the results should thus be investigated further, preferably using more extensive data that allows more advanced statistical modelling.