Who teaches and who learns? Policy learning through the C40 cities climate network
This study examines the network structure of policy learning in the C40 Cities Climate Leadership Group, which is a network of the world’s largest cities committed to tackling climate change issues. Among forty members and nineteen affiliate members, we ask the question with whom do cities learn and why? How are policy-learning relationships associated with cities’ multi-stakeholder governing body, policy performance, and cultural similarities? While studies on learning have analyzed conditions facilitating learning, quantitative studies of local government learning in global networks are rare. To facilitate the investigation into learning, we conceptualize learning as a process comprising information seeking, adoption and policy change, and focus on information seeking as the foundation step in the learning process. This social network analysis using the exponential random graph model reveals the cities that seek information and those that are information sources are different subgroups. Furthermore, analysis of nodal attributes suggests that transmunicipal learning in the C40 network is facilitated by the presence of a multi-stakeholder governing body; homophily of culture (language and regional proximity); and higher level of climate change policy performance. Creating a multi-stakeholder governing body could ensure participatory representativeness from citizens and relevant stakeholders to enhance climate change policy engagement and decision making as well as policy learning.
KeywordsPolicy learning Climate change Transnational network Multi-stakeholder governing body Social network analysis Exponential random graph model
Akaike information criterion
C40 cities climate leadership group
Carbon disclosure project
Central intelligence agency
Cities for climate protection
Exponential random graph
International council for local environmental initiative
Human-induced climate change is a critical challenge, posing substantial risks at the local level. City governments are important local level actors, providing many services to the community, for example waste management, drainage, and transport services, while closely communicating and interacting with the community, and thus are an important conduit for information flows and policy leadership on climate change.1 Additionally, urban populations are forecast to increase in the future (UNPF 2009); consequently, cities are critical sites for addressing climate change.
More importantly, cities have become international actors, collaborating with other cities around the world to combat climate change. Climate change policy is divided into two fields: mitigation (reducing greenhouse gas (GHG) emissions) and adaptation (adjusting to climate change impacts). While both mitigation and adaptation policies are important, they can be in synergy or conflict, which creates complex challenges for urban studies scholars and practitioners (Pizarro 2009), in both developed and developing countries (see for example, Hamin and Gurran 2009; Laukkonen et al. 2009; Revi 2008). These challenges are exacerbated by the different temporal and spatial scale foci of the climate change policy approaches; mitigation is long term and large scale, while adaptation is typically considered over shorter temporal and smaller spatial scales (Biesbroek et al. 2009). At the macro international scale, we have witnessed policy failure due to tugs-of-war between developed and developing countries over the distribution of responsibility to mitigate greenhouse gas emissions. Yet, cities are taking initiatives to tackle climate change mitigation and adaptation by forming a variety of international climate change networks such as the C40 Climate Leadership Group and the International Council for Local Environmental Initiative’s (ICLEI) Cities for Climate Protection (CCP) campaign. These transnational municipal networks2 for climate change primarily aim to facilitate learning best practices and collaboration among member cities. Thus, learning and collaboration for local level climate policies takes place across state borders.
An important approach to dealing with increased uncertainty, such as that generated by climate change, is being able to respond to changing circumstances. Learning provides a central mechanism for improving adaptive capacity (Scholz and Stiftel 2005) and facilitating policy adaptation (Weible et al. 2010). Furthermore, exchange among actors through learning in the field of climate change has the potential to stimulate improvements in other policy domains (deLeon and Varda 2009). Learning focused on specific policy problems is termed “policy learning.” As both theoretical studies (e.g. Glasbergen 1996; May 1992; Sabatier 1988) and empirical studies (e.g. Betsill and Bulkeley 2004; Fiorino 2001; Flynn and Kroger 2003) have illustrated, policy learning resulting in policy change is a complex process, involving many actors within the learning network in varied and dynamic processes. The key research questions arising from the “black box” of learning interactions in networks include from whom do actors in a network learn and why? In short, how do we use actor characteristics to explain learning relationships?3
Theoretically, this paper aims to contribute to the learning literature by investigating the nature of policy learning, seeking to identify the drivers for information seeking and the characteristics of both information seeking cities and those cities considered to be information sources. In particular, we analyze the impact of multi-stakeholder governing bodies, similarity in language, geographical region and geographical structure, and performance level on the formation of learning relationships in a network.
This study investigates the global C40 network, a network of developed and less developed cities focused on learning to adapt and respond to climate change as an example of a transmunicipal learning network (C40 Cities 2011). In forming these transmunicipal learning networks, cities (who are assumed to be competitive) decide to cooperate and collaborate to overcome the collective action problem of climate change. Most learning research has analyzed learning efforts among individuals and organizations, companies or international organizations within the boundary of one region or state or single international organizations (for example, see the following studies: Beem 2006; Edmondson 1997; Fiorino 2001; Gabler 2010; Haas 2000; Hall 1993; Mattes and Bratton 2007; Rashman et al. 2005; Siebenhüner 2008). Little is known about how cities learn in international arenas, and the C40 network with the primary aim of promoting learning and information sharing is therefore an appropriate case (C40 Cities 2011). The C40 network is different from other climate change transnational municipal networks as it has been initiated by cities (not a separate organization), and therefore, the C40 network does not have a large centralized governance structure (C40 Cities 2011). Examining learning activities in the C40 network enhances our understanding of cities’ interactions across state boundaries. We anticipate these findings could contribute to the identified lack of knowledge of internal network operation (Kern and Bulkeley 2009), knowledge generation, and information sharing (Betsill and Bulkeley 2004) in the fields of transnational municipal networks and climate change governance.
In terms of methodology, by employing quantitative social network analysis (the exponential random graph model), this study aims to test the relevance of cities’ attributes as driving overall network structure. Potential relationships between information seeking as a learning activity and cities’ attributes such as language, region, geographical structure, knowledge of other cities’ performance, city governance, and the perception of risk are developed. We hypothesize that learning relations in the C40 network are positively influenced by the following: the government level where responsibility for climate change policy is held; the presence of a multi-stakeholder advisory committee; similarity in geographical region, language, and geographical feature (e.g., delta, mountain); and the performance of cities that serve as information sources.
This paper continues by discussing the concept of policy learning within networks and developing our three-stage learning process, followed by the theorization of the city attributes (variables) to be tested. The social network analysis methodology used and the data collected are described, followed by the presentation and discussion of the results. The final section points to the potential implications of the study and opportunities for future research.
Conceptualizing learning: relation and process
A learning approach to investigating policy processes has become popular over recent decades as a means of explaining policy change (see for example Glasbergen 1996; Hall 1993; May 1992; Sabatier 1988). Although there are many different definitions of policy learning, we adopt a broad view, defining it as the use of information and knowledge to make predictions of the future, which are then used to make decisions (Bennett and Howlett 1992: 278). In the following discussion, we explore policy learning and the concept of policy networks, including transmunicipal networks, before elaborating on our conceptualization of the policy-learning process.
Policy learning: relations
Policy-learning theory provides numerous conceptualizations of learning with different terminology and definitions. For example, in environmental policy, Glasbergen (1996) identifies technical, conceptual, cognitive, and social learning, while in political science, May (1992) discusses political, instrumental, and social learning; and Hall (1993) categorizes first-order-, second-order-, and third-order social learning.4 (See also Sabatier (1988) and Siebenhüner (2008) for different definitions and contexts.) Considered as a whole, these different conceptualizations move from learning about policy tools or problems to reviewing the policy problem more conceptually, to considering the normative basis for the policy objectives (Bennett and Howlett 1992) and focusing on the interaction and communication to create continual learning opportunities (Glasbergen 1996). As the learning becomes more abstract and related to core beliefs and values, the challenge to undertake the different types of learning is more difficult (Sabatier 1988). Additionally, identifying the causal link between actors who learn and how they learn becomes increasingly challenging given the large number of actors (e.g., industry, politicians, community, and other stakeholders) and external events (e.g., national and/or international events) influencing the policy-learning process (Beem 2006; Weible et al. 2009).
Policy learning involves many different actors, both individuals and organizations, which come together to form policy networks (Sabatier, 1988). Policy networks could comprise elected officials (a restricted view), or elected and relevant agency officials and leaders of stakeholder organizations (a broader definition) (May 1992). An even more inclusive view includes public and private organizations actively concerned with a policy problem, and also potential (latent) actors, such as those with inadequate information (Sabatier 1988).
In defining the policy network in more detail, Sabatier (1988) argues that actors with similar beliefs are likely to form a smaller policy network, which then seeks influence over other smaller policy networks to implement their desired policy program. Actor characteristics (e.g., similarities and differences in beliefs) and their relationships (network structure) can influence the nature and extent of policy change (Howlett 2002). Network structure and operation is further complicated by roles different actors play, for example, the influence of policy brokers who seek to build transitive (triadic) social networks (Henry et al. 2011). Investigating networks, including factors influencing network relationship development, can provide insights into the policy process (Carlsson 2000; deLeon and Varda 2009). Given that learning takes place among numerous stakeholders and information and knowledge flow in different directions, learning, especially in networks, reflects diverse relations among network participants.
A policy network may exist at different scales, within individual cities or between cities, such as a transmunicipal network. Transmunicipal networks have been observed for hundreds of years (e.g., Hanseatic League in Europe during the 13th century) (Ewen and Hebbert 2007). More recently, transmunicipal networks have been advocated as a means of addressing complex cross-boundary environmental problems, such as climate change (Bulkeley 2005). These transnational municipal networks focusing on climate change provide opportunities for learning, information sharing, networking, generating legitimacy, exchanging values, and collaboration (Betsill and Bulkeley 2003, 2004). Therefore, we contend that transmunicipal networks are potential sites where policy learning could be initiated or actually occur.
Learning among the member cities in transmunicipal networks is influenced by policy entrepreneurs, who are critical for the successful city-based implementation of initiatives which originate in the transmunicipal network (Kern and Bulkeley 2009). A focus on implementation relates to instrumental learning, which is learning about the viability of policy instruments or their implementation designs (May 1992). Instrumental learning excludes examination of the policy goals (May 1992) and therefore is similar to Glasbergen’s (1996) technical learning, which involves policy adjustments, rather than substantial changes. Learning about policy goals and scope is undertaken by smaller groups of actors (sometimes called policy communities) in policy networks (Bennett and Howlett 1992).
In summary, we argue that policy learning is inherently relational; it is dependent on the interactions among different actors. Transmunicipal networks provide a site for potential policy learning, which can be transferred to individual cities. The importance of transmunicipal networks has increased over recent years, particularly in the field of environmental governance, and therefore, investigating learning in a transmunicipal network can provide insights into policy-learning literature, environmental governance, and understanding network operation. We next focus on the process of policy learning, which forms the basis for our empirical study.
Policy learning: process
Information seeking is where actors use their social relations to identify and access information from other actors. This could involve accessing information to generate different types of knowledge, such as knowing what (knowledge about facts), knowing how (ability to do something), knowing why (knowledge about principles), and knowing who (knowledge about who knows what) (Johnson and Lundvall 2001). Both explicit (formal) and tacit (implicit) knowledge are important for learning (Lam 2000). Explicit or formal knowledge can be separated from the person who holds the knowledge (e.g., written down) and can be transferred across time and space, independently of that person, while knowledge depends on close interaction and development of trust and a shared understanding (Lam 2000). Both types of knowledge can be transferred through networks (Marsden et al. 2011; Cross et al. 2001). External information sources are important for international organizational learning and policy change regarding climate change (Haas 2000). Assimilation of new information is critical for learning (Hall 1993; Henry 2009), and therefore, information seeking is a core activity of policy learning (Sabatier 1988; Siebenhüner 2008). Information is particularly important when developing causal models of the factors influencing a policy problem, which then form the basis of policy learning (Sabatier 1988; Glasbergen 1996).
Definitions of learning typically separate the acquisition of information from its processing (e.g., Huber 1991, and also the cognitive part of learning); in our three-stage policy-learning process, our second stage is adoption, which primarily depends on the internal learners’ characteristics, needs, and belief systems. During the adoption stage, learners may discard or modify information and knowledge gained from others based on past experience or discussions reflecting problems and aims they pursue. Internal information processing involves individuals attaching meaning to information based on prior knowledge and connecting the new information to the existing processed information (Alvesson and Kärreman 2001). It can involve aggregating and codifying information (Swan and Scarbrough 2001). In conceptual policy learning, adoption may require a reexamination of individuals’ core beliefs, and possibly an acknowledgment of incorrect assumptions or contradictions of their goals (Sabatier 1988). Lesson drawing literature discusses the adoption stage in terms of evaluating and judging information, developing a conceptual model of the policy, and finally developing a new program (Rose 1991). In this way, the adoption phase focuses on evaluating and making sense of information in relation to the specific policy program. While the internalization of information may be influenced by discussions with others (as per the discursive learning approach, see for example, Lipschutz 1997), the adoption of information is primarily an internal process (James and Jorgensen 2009) and therefore is not considered relational.
The third stage of the policy-learning process (Fig. 1), policy change, focuses on the outcome of the information seeking and adoption stages. In response to the new insights obtained from the previous stages, policy change in a learning process involves adjusting policy goals and techniques (Hall 1993) or developing a more integrated solution to the problem (Fiorino 2001). In contrast to adoption, a largely internal process, policy change in a learning process involves multiple stakeholders. However, the extant learning literature is not often explicit regarding the link between information seeking, adoption, and policy change (see for example May 1992). Significant events in the policy sphere (such as a change in government or senior policy leadership) can also bring about review and reflection of policy goals and thus influence policy change (Fiorino 2001). External actors such as the media or other actors outside of the policy-making sphere can introduce new ideas which challenge the existing perspective and therefore can also exert pressure on policy makers (Hall 1993). Factors such as institutional capacity, policy needs, power relations among stakeholders, ideology, and the roles different stakeholders play influence policy change (Nowlin 2011), resulting in a highly contested policy change process (Betsill and Bulkeley 2004). However, the influence of external factors on policy change cannot be understood without concurrently examining the internal relations of the stakeholders involved in the policy network (Marsh and Smith 2000; Weible et al. 2009). Improving our understanding of policy learning can therefore contribute to our understanding of policy change. Information seeking and adoption of the information does not automatically change policy objectives, content, or measures directly; rather it can be used indirectly as a source of inspiration to act (Kern and Bulkeley 2009).
The policy-learning process can be observed in the example of the bicycle policy of Changwon (South Korea). Changwon, a C40 member city, identified Paris as a city that has implemented a successful bike policy to reduce the GHG emissions from motor vehicles. Changwon, located in the southern part of the Korean peninsula, is the first planned city in South Korea. Using its planned urban structure and flat topography, Changwon City aimed to design and implement a bicycle-based transportation policy. To develop this policy, Changwon city government sought information by benchmarking high-performing cities and their policies. Research trips by city officials to Paris enabled Changwon to learn about Paris’ “Velib” program—a bicycle-based public transportation system (information seeking process). Changwon then adopted its own bike-based transportation system, named “Nubiza” (adoption process), incorporating advanced information technology such as global positioning system (GPS) and mobile phone payment systems. A new management team was created and other relevant policies such as bike insurance and a light system for bikes was established (policy change process). As seen in the process of policy learning, information seeking, and adoption of knowledge may lead to policy or institutional changes.
The policy-learning process is particularly complex in networks (Betsill and Bulkeley 2004) and our three part conceptualization is anticipated to facilitate policy-learning investigation in networks by articulating different process stages. In this study, we analyze information seeking as the foundation part of the policy-learning process. Information seeking is more likely to be relational than adoption, where an individual or organization is likely to undertake this process internally, and identifying policy change as an outcome of learning is difficult. Therefore, we contend that a greater insight and understanding of information seeking within a network can contribute to an improved understanding of the relational aspects of policy learning.5 In order to assess the relational aspects of information seeking, three elements should be identified: the presence or absence of learning relations among cities, cities who learn and are learned from, and directions of relations.
Theorizing learning in a transnational municipal network
This study focuses on the learning observed within a transnational municipal network (the dependent variable) and examines how city attributes (the independent variables) explain the learning patterns observed in the network. We now turn our discussion to theoretical perspectives of the independent variables.
Theories of policy learning generally identify actors who learn but do not often explicitly describe the actor attributes which facilitate or constrain the learning process, beyond Sabatier’s (1988) identification of actors with similar beliefs forming advocacy coalitions. To identify actor characteristics relevant for investigating information seeking in a transnational municipal climate change network, we reviewed a range of policy learning, information seeking, and organizational learning literature, which are related to information seeking and learning in a policy network. As there is insufficient learning literature in environmental studies, we extended the review to the general literature in these fields. The literature review revealed four attributes which were related to both cities’ learning in a transmunicipal climate change network: level of climate change responsibility, homophily of language, region, and geographical structure, knowing of other cities’ performance, and perception of risk of climate change impacts. These concepts and how they are related to the C40 transnational municipal network are discussed below.
First, learning within and between organizations is considered part of a political process, and therefore local government leaders (i.e., mayors) are important actors (Rashman et al. 2009; Rose 1991). Leaders can influence others by bringing people together and creating an environment conducive to learning (Rashman et al. 2009). In international climate change cooperation, unilateral leadership (where nations lead by setting a good example by formulating goals and implementing policies) was identified as an influential leadership style (Saul and Seidel 2011). Similarly, political leaders strongly influence an organization’s learning policies and structures (Dekker and Hansen 2004) and initiate information seeking (Marsden et al. 2011). A city’s leadership level (i.e., at political or senior executive levels) can influence a city’s approach to learning or how readily the city looks beyond its borders for information (Marsden et al. 2011). Therefore, we propose the following hypothesis: where responsibility for climate policy lies with the mayor or chief executive officer, the city is more likely to seek information from other cities.
A second element of the level of governance in transmunicipal network learning is the existence of a multi-stakeholder governing body to advise cities on climate change. External advisors could be experts or community members or representatives of non-government organizations. Expert knowledge can stimulate learning by providing different opinions on the policy field in question (Rose 1991) and has been influential in international environmental issues, such as ozone depletion and acid rain, where experts have gained timely access to the policy process (Haas 2000). Working with advisory committees provides government officials with experience in the coordination and learning from individuals and organizations outside of government. Experience working together facilitates individuals’ willingness to share and coordinate across positions and roles (Reagans et al. 2005). Local government experience working with city-scale stakeholder advisory committees on climate change could provide officials with skills and willingness to work with other actors beyond the city boundaries, such as other cities in the C40 network. Therefore, we propose the following hypothesis: cities with city stakeholder advisory committees are more likely to learn from other cities.
Identifying similarities that facilitate learning relates to the principle of homophily in network studies, where “similarity breeds connection” (McPherson et al. 2001). Individual attributes such as gender, age, class, or ethnicity have been investigated to determine whether they constrain or enable social ties, while at a global level, being located within the same geographical region has been identified as a basis for homophily (Zhou 2011). Geographical proximity was identified as influential in a study of inter-city transport policy transfer, where some cities considered that looking at cities outside of their region was an extravagance (Marsden et al. 2011). Homophily is used as an independent variable to examine network structure as increased learning ties within regions is related to convenience and ease of interaction, that is, it is easier and takes less effort to interact, either in person or more remotely, with actors located in the same geographical region. Additionally, perceived higher costs of interacting with cities in other regions may negatively influence learning (Marsden et al. 2011). This leads to the next hypothesis: cities are more likely to seek information from cities located within the same geographical region.
In addition to geographic region, geographic features of cities may facilitate learning activities among cities with similar attributes. Climate change has different impacts on different terrains and locations (Leemans and Eickhout 2004). That is, cities located in coastal areas have different climate change impacts compared to those located in mountainous territory since, for example, sea-level rise as a climate change impact is likely to be a serious concern for coastal cities, and reduced water resources are likely to be a problem in mountainous cities. Therefore, we propose the following hypothesis: cities seek more information from those cities that share a similar geographic terrain (coastal, plain, and mountainous terrain).
The final homophily variable investigated is similarity of language, which has been identified as a basis for increased learning activity (Child and Faulkner 1998); learning is facilitated by the ease of communication among cities which speak the same language. More effort is required to interact with cities that have different languages due to the need for translating information and other documents (Guo 2007; Wood and Parr 2005). Therefore, we propose the next hypothesis: cities are more likely to seek information from cities which share the same language.
The next nodal attribute used to understand network learning is “knowing” about information sources, that is an actor’s knowledge or awareness of another actor’s potential to hold the desired information (Borgatti and Cross 2003). Network members each have information stores which may be derived from knowledge or experience. An information seeker’s perception of and understanding of other network members’ information stores will influence their likelihood of seeking information from the network members. Good sources of information may be actors with expertise; however, the extent that the expertise is known across the network will influence whether these actors are considered to be potential information sources by those actors seeking information.
In the context of a transmunicipal network such as C40, it is difficult to measure whether cities know about other cities or not. Therefore, in this study, we measure “knowing” through a cities’ climate change policy performance. High-performing actors are sources of learning (Radaelli 2009; Rose 1991; Simmons and Elkins 2004) and are therefore likely to be identified as sources of information on policies and strategies for addressing climate change. While cities with failed policies might also offer lessons about what not to do, policy failures are less likely to be made public (May 1992), and therefore, we exclude policy failures from this analysis. In contrast to the homophily principle of network attributes of language, geographic structure, and region, knowing focuses on differences in performance among actors which are likely to facilitate learning. We define high-performing cities as those which have planned, implemented, and monitored climate change policies in accordance with the Cities for Climate Protection Program (ICLEI 2011). Information about high-performing actors may be distributed through the network, for example, through the provision of awards or case studies of successful projects which are publicized. Therefore, we propose the following hypothesis: cities seeking information are more likely to target high-performing cities.
The final nodal attribute is related to cities’ vulnerability to climate change. Cities facing greater vulnerability to climate change demonstrate greater commitment to climate mitigation and adaptation policy (Zahran et al. 2008) and therefore need to allocate more resources to meeting these commitments through climate policy development. Developing new policies for climate change mitigation and adaptation provides the stimulus for information seeking (Rose 1991). We therefore propose the following hypothesis: cities that face high risks due to climate change are more likely to learn from other cities.
Data and network description
To test the above hypotheses, survey data on policy learning was collected from city officials who are responsible for climate change policy and international affairs in 2010. To ensure a relevant person responded to the survey, the survey was sent to the department of environment or mayor’s office in all 58 C40 member cities. Given the survey focused on climate change policy and international affairs, it was considered appropriate to target these city officials. The individual city officials represented the city governments and as such, we attempted to measure learning behavior of city governments through their officials. As the focus of this study is the C40 cities network, we assume the survey respondents are representative of the city governments. We acknowledge that the climate change policy and international affairs sections of city governments may learn from other organizational departments or other stakeholders including non-government organizations (NGOs), companies, or individual citizens, but we designed the survey to focus on learning behaviors among city governments to provide a boundary to our study and consistency across the unit of analysis cities.
Policy-learning behavior was measured by asking each respondent to “list all of the names of C40 member cities that your city government learns best practice climate change policies from,” using a roster of C40 member cities.6 To minimize misunderstandings of the concept of learning, we provided a description of learning activities in the survey: “C40 cities share their best practices through conferences, the C40 and their own websites, and individual contacts. In the following questions, we would like to know of the partner(s) that your organization learns best practices from, and why.” A name generation survey using a roster and description of relational activities is a conventional way to collect relational data in social network analysis and is considered more reliable than surveys without rosters (Wasserman and Faust. 1994). Given the C40 network is global, and also that many city to city communications are informal, and information available online is variable in quantity and quality, we were unable to cross check the data reliability with a third data source.
Nodes in our analysis are city governments where the learning relationships were estimated based on individual surveys. Data was initially collected through email. Thirty-eight out of 58 surveys were collected, representing a 64 % response rate. Among thirty-eight surveys, thirty-three cities that identify their learning activities with other cities are analyzed. The number of total ties in the network is 74 and the network density is .07.7 Among these ties, three are mutual and 68 are asymmetric.
The data were transformed into a 33 × 33 adjacency matrix of links for each city, coding a value of 1 for the presence of a learning tie and a value of 0 for the absence of a tie. Learning ties are directed in the sense that a tie between city A and city B may indicate city A learns from city B, but city B may not learn from city A. Using directional ties also enables us to evaluate the direction of information flows by examining the points of origin and destinations of learning.
This study examines nodal attributes—the city climate change governance (political leadership, multi-stakeholder advisory committee), policy performance, similarities in geographical region and structure, and language, and perception of city climate change risk—as factors to explain the overall network structure.
To test the hypothesis of the impact of governance level on policy-learning ties, we use the survey data collected in 2010 from the Carbon Disclosure Project (CDP)’s Global Report on C40 Cities (KPMG et al. 2011). As a not-for-profit organization, CDP collects and reports C40 member cities’ GHG measurements, governance structure, and climate change adaptation policy. To measure political leadership, CDP asks the question, “Where is the highest level of responsibility for climate change in your city government.” We coded a value of 1 for the answer “governor or mayor”; and a value of 0 for “other officials.” The variable presents whether higher levels of political leadership (mayor or other chief executive of city) engage in and take responsibility for a city government’s climate change policy.
Collaborative advisory committees that consist of city officials and representatives of the public, NGOs, experts, and business sector representatives are likely to facilitate learning about climate change policies from other cities. CDP also asks whether a city government collaborates with stakeholders (science, business, and citizens) and forms a committee for local climate change governance and program development (a multi-stakeholder advisory body). We use dichotomized values (1 for existence of multi-stakeholder committee; and 0 for non-existence of such body.).
A city’s knowledge of climate change policy is measured by assessing the extent a city government has developed climate change policy based on ICLEI’s CCP five milestones: GHG inventory; climate change action plan; GHG reduction target; policy implementation; and monitoring and disclosing outcomes. Cities with higher levels of development in this five milestone process are likely to be information holders as they have obtained information about the feasibility and applicability of policy measures through their experience of planning and implementing the climate change policy milestones. Data is generated firstly from the survey responses and validated by checking ICLEI’s website.
Perception of climate risk drives city governments to act (i.e., joining local climate policy networks) and may facilitate learning from other cities’ best practices (Zahran et al. 2008). The self-reporting CDP survey of C40 member cities asks two questions on physical risks of climate change; “Do current and/or anticipated effects of climate change present significant physical risks to your city?” and “Have you identified the physical risks from climate change that your city may face?” A value of 0 is coded for those cities that answer “no” on both questions; a value of 1 for answering “yes” for one out of two questions; and a value of 2 for answering “yes” for both questions.
To test the hypotheses on homophily, we include three nodal attributes, language, geographical structure, and region, as independent variables. Learning about climate change policies in the C40 network potentially takes place across cities which speak different languages and are located in different continents and different terrains. In social network and graphic theory literature, homophily refers to the tendency of ties to be formed between nodes that share similar attributes (McPherson et al. 2001). Similarities in language and location might be a driving factor that facilitates learning since learning processes need cognitive understanding of substantive content. If city officials in two cities speak the same language, they are likely to learn from each other without encountering a language barrier. For analysis, five world languages (English, Spanish, Chinese, Arabic, and others) for each city are coded based on the CIA Factbook. To operationalize geographic structure, we categorize cities with three different types of geographic features: coastal cities; cities on plains; and mountainous cities. By looking at World Atlas and OECD research, we identify twenty coastal cities (such as Rotterdam, Sydney, Shanghai), eight cities on plains (such as Berlin, Changwon, Paris), and mountainous cities (such as Mexico City, Salt Lake City, Madrid where the altitude is over 200 meters).
Variable description and data source
Description and operationalization
Whether a multi-stakeholder climate committee has been established (1, existence of committee; 0, no committee)
Carbon Disclosure Project (KPMG et al. 2011)
Counts whether city government achieved (1) GHG inventory; (2) action plan; (3) reduction target; (4) implementation; and (5) monitoring (0–5)
Survey & The City Climate Catalogue (http://climate-catalogue.org/)
Highest level of responsibility for climate change (1 for governor or mayor; 0 for other officials)
Carbon Disclosure Project (KPMG et al. 2011)
Perception on climate change risk & physical risks from climate change (0 for neither risk perception nor physical risks; 1 for either risk perception or physical risks; 2 for both)
Carbon Disclosure Project (KPMG et al. 2011)
Asia, 1; N. America, 2; S. America, 3; Europe: 4
English, 1; Spanish, 2; Chinese, 3; Arabic, 4; Others, 5
CIA The World Fact book
Elevation (altitudes) and coastal city (Coastal city, 1; plain, 2; mountainous territory, 3)
Population (natural logged)
Population of municipality
UN-Habitat, Global Urban Indicator
Second, those cities that have achieved a higher level of climate change policy are located in the center of the network. Chicago, New York, Portland, Seattle and London play central roles in the learning network of the C40 cities. Cities that have conducted the initial stages of climate change policies, such as Basel, Changwon, Shanghai, are located in the periphery of the network.
Social network analysis: exponential random graph model
Learning can be viewed as relational behavior, especially in a network context. If there are those who learn, there must also be those who are learned from, and are benchmarks of good climate policy practice. While individuals or organizations can learn from past experience or internal guidelines, social actors mostly learn from other individuals or organizations. To examine this relational characteristic quantitatively, this study employs social network analysis, specifically the exponential random graph model (ERG model). Conventional regression models that assume independent relations of observations may be not an appropriate method to analyze relational data (Borgatti and Cross 2003).
The ERG model in the Statnet of R is a tool to test the likelihood of the presence of ties and formation of a particular set of network ties (Handcock et al. 2008). As such, the dependent variable of the ERG model is the observed network structure, which consists of nodes/actors (in this case, cities) and ties (in this case, learning linkages). The observed network structure (Fig. 2) as the dependent variable in this analysis is the one possible appearance of relations in the same number of nodes. Based on this observed network structure, the ERG model calculates a set of possible networks and probability distribution over them (Newman 2010). By simulating a number of networks using the Markov chain and Monte Carlo maximum likely estimation, ERG modeling uses stochastic approximation to estimate the parameters (Hunter et al. 2008). Independent variables of the ERG model can be magnitude (i.e., different degrees of values) or similarities (i.e., the same or shared attributes) of nodes. ERG modeling determines that the probability of the observed C40 learning network is not random; in other words, the observed network is associated with independent variables (nodal attributes and/or network configurations) and not formed by chance at the conventional (95 %) confidence level.8
ERG model outcomes for policy learning at C40 network
Network structure effects
To test the impact of nodal attributes on network structure, the model includes the level of policy performance and similarities in language and geographical region. Higher levels of information sources measured by the cities’ policy performance have a positive effect on network formation. This suggests that information seekers are likely to form learning ties with those cities which have already achieved high levels of climate change policy performance. This outcome corresponds to Fig. 2, which indicates that cities with higher performance are located in the center of the network. Given the nature of learning in which information typically flows from those who have more knowledge to those who have less knowledge (i.e., teacher to student), high levels of policy performance tend to attract more information seekers.
The leadership variable is not statistically significant at the 95 % confidence interval. This does not necessarily mean that leadership does not substantially facilitate learning. Rather, this outcome is based on the surveyed data since most cities (28 cities out of 33 cities, which accounts for 84 % of surveyed cities) responded to the leadership question saying that climate change responsibility lies at the highest level of city governance. Due to the lack of variation, the leadership variable may fail to be associated with the network structure. Thus, we interpret this outcome as not all cities where the top levels of government officials (governor, mayor, or chief executive) are responsible for climate change tasks seek information from other C40 member cities.
Nearly all C40 members perceive that some parts of cities are already subject to climate change risk such as more intense and frequent flooding, storm surges and sea-level rises. However, cities’ climate change policies and action plans so far have mainly focused on GHG mitigation policies, with few exceptions (e.g., New York City’s adaptation plan). Underdevelopment of adaption policy can be found in the CDP survey data, which shows that 35 responding cities “are in the process of identifying the specific risks they face from climate change.” Learning from others is unlikely to take place when adaptation policies of other member cities are an ongoing process. While perception of physical risks from climate change is high, cities are now working on how to respond to these risks. We expect that learning activities of cities will focus more on risk identification and adaptation measures in the future.
Both the language and geographical region homophily hypotheses are supported by the results. Homophily in language and region has a positive effect on network structure at the 95 % confidence level. Learning among C40 cities may take place because cities have similar characteristics which can reduce transaction costs. Cities using the same language do not need to expend effort to translate their partners’ language into their own. Location is also important as cities in the same region are likely to know the way close cities approach and solve shared problems. However, homophily in geographic structure (terrain) is not associated with the tie formation in the learning network. While there is an initial effort to share ideas among coastal cities, only a few cities (Rotterdam, New York, Jakarta, Tokyo, and London) are involved in Connecting Delta Cities, a C40 sub-network for coastal cities to tackle climate change adaptation collectively. Furthermore, such an effort to share ideas among cities with similar geographic attributes has not been found in those located in plains or mountainous terrain. Forming cohesive subgroups among those sharing similar problems could facilitate policy learning among cities.
A coefficient in the ERG model indicates the log-odds ratio of tie formation given that the dependent variable of network analysis is the presence or absence of ties in the logit regression model for a binary outcome variable. Thus, a positive coefficient indicates that tie formation in the network (network structure) is a function of city (nodal) attributes. In other words, an increase in the value of a nodal attribute, such as climate change policy performance, is likely to increase the log-odds ratio of tie formation in a given network (Henry et al. 2011).
Number of learning ties (in-ties and out-ties) and which cities teach and learn different policies
Cities learning the most
Cities being learned from the most
Examples of policy learning (from)
Examples of policy learned (by)
Action plan (Mexico City)
PlaNYC (Rio de Janeiro)
Rio de Janeiro
Action plan (London)
Waste management (Tokyo)
Transport (Mexico City)
Action Plan (Philadelphia)
Cycling (Rio de Janeiro)
Water management (Portland)
Action Plan (Austin)
As leading global cities, New York and London were ranked first and second, respectively in terms of being information sources (or sources of learning). These two cities initiated the C40 network and hosted the 2005 (London) and 2007 (New York) C40 Mayor Summits. The mayors of London (former mayor Ken Livingstone) and New York (Michael Bloomberg) have played a key role as the chair of C40 Climate Leadership Group. For example, New York’s comprehensive climate change action plan (PlaNYC) is modeled on Rio de Janeiro’s action plan and its green building policy is a learning target of Rome. Chicago, Copenhagen, San Francisco, and Paris all have five incoming learning ties from other cities. These four cities, two from the US and two from Europe, which lead the way in climate change policy, are benchmarked by other C40 member cities.
Our City [Sydney] is following the lead demonstrated by London in decentralizing its electricity generation to use lower carbon fuels to make use of “waste” heat through tri-generation. We have a plan to remove our city’s reliance on coal-fired electricity by 2030 through energy efficiency and using tri-generation (70 %) and renewable energy (30 %).
Both Sydney and London have established a climate change committee for facilitating planning and implementing climate change–related policy. London city government demonstrated higher levels of policy performance in energy-related climate change policy, which makes London a source of information and Sydney an information seeker. While they are not located in the same continent, they speak English as a common language. By learning about London’s decentralizing energy supply system, Sydney attempts to achieve its policy goals with some internal modification of policies.
Discussion and conclusion
In this study, we analyze the formation of learning links among cities around the world in the C40 climate leadership network. The C40 network was originally established to facilitate interactions including learning, collaboration, and collective procurement to combat climate change at the local level. However, little is known about with whom cities interact and which factors drive learning relations. Our social network analysis of information seeking suggests that learning ties are likely formed when cities create an internal committee of multiple stakeholders, have higher levels of experience and knowledge, and share a similar regional context and language. However, regional homophily was only observed among North American and European cities.
This study contributes to the policy-learning literature and local level climate change governance in several ways. First, theoretically, in addition to drivers of policy learning such as the multi-stakeholder governing body and knowing where to find information, similarities in language, geographical structure and region, have been identified by previous research as explanatory factors for policy learning. Our analysis has confirmed that these attributes are important for both cities that seek information and also for those who are sources of information. Integrating these actor attributes into the single analysis improves understanding of actor attributes in policy networks and learning which are not often specified. In addition, drawing on organizational, network, and policy-learning literature, we conceptualize learning as a three-step process comprising information seeking, adoption, and policy change. Our focus on information seeking as the foundation process of policy learning provides insight into this particularly relational part of the policy-learning process among multiple organizations and among different types of policy learning (i.e., instrumental, conceptual, and social learning). This conceptualization could be used in future studies to help analysis of the complex policy-learning process; we suggest examination of the adoption and policy change stages as opportunities for future research together with interactions among all three stages.
Methodologically, social network analysis, especially the ERG model, quantitatively tests hypotheses on how actors’ attributes influence learning network formation. Most learning literature has examined typologies of learning (Depledge 2006; Glasbergen 1996; May 1992), the learning process employing a single case (Beem 2006; Busenberg 2001; Flynn and Kroger 2003; Gabler 2010), comparative case studies (Dekker and Hansen 2004; Hall 1993), or process tracing (Fiorino 2001; Rashman et al. 2009). Given that information seeking is a relational concept, social network analysis enables researchers to explain black boxes of interactions among multiple actors in networks.
Empirically, this study attempts to enhance our understanding of learning activities among cities around the world. The burgeoning literature on multilevel climate governance (e.g., Betsill and Bulkeley 2004; Kern and Bulkeley 2009) has emphasized the role of cities and their networks in curbing GHG emissions and increasing adaptation capacity at the local level. Yet, comprehensive quantitative analysis of learning beyond state boundaries on climate change policies is greatly lacking. Our work expands recent empirical analyses concerning city interaction in the context of multilevel climate change governance. We specifically focus on learning within a network of cities, thus providing insight into factors likely to drive network learning in this and other transmunicipal networks related to climate change policy.
Three limitations of this study should be highlighted for future studies. First, the analysis of network formation in this study is static. Learning can be evolutionary in that learning ties change over time. Further studies of learning focusing on network evolution can refine the dynamics of learning relations. In addition, examining how characteristics of network structure, such as centrality, influence cities’ climate change policy outcome will be a valuable avenue for future research. Second, the existing learning literature suggests learning is a process of assimilating information (information seeking) and then updating behavior (policy change) (Henry 2009). To further understand learning activities, studies on how information seeking influences policy change should be pursued. Qualitative case study methods would likely improve our understanding of policy change within each city and the connections between individual cities and networks, and also how information flows between the cities and networks. Third, the network analysis in this study omits those cities that did not respond to the survey. Even though the surveyed cities did not indicate learning ties with these omitted cities, it is uncertain whether the omitted cities are learning from each other. Increasing survey responses and other probing techniques may solve this issue of omitted member cities.
In conclusion, we return to answer the question posed at the start of the paper: With whom do cities learn in a network like C40 and why? The aim of such networks is to facilitate the learning of best practices and also collaboration among cities to address common pool resource issues, such as climate change. However, learning ties in the C40 network are unevenly distributed, suggesting assumptions about horizontal network relations and mutual interaction for transmunicipal networks are incorrect. If the aims of climate change networks are to seek mutual learning about best practices, measures for effective learning between cities in Africa, Asia, South America, and cities in North America and Europe are imperative. Those cities lacking information and knowledge also need to pay more attention to and learn from other cities which successfully implement climate change policies. In addition, this study suggests that forming advisory committees with multiple stakeholders is likely to facilitate learning. Creating such a governing body could ensure participatory representativeness from citizens and relevant stakeholders to enhance climate change policy engagement and decision making.
We have used “actors” to refer to organizations and individuals in the generic sense of entities that have agency; “actors” is generally used in the theoretical section where concepts of policy learning and policy networks are described and discussed. We have used stakeholders to refer to individuals and types of organizations which include non-government, private sector organizations involved in policy learning. We use this term to emphasize the variety of organizations and individuals involved in policy learning.
We refer to transnational municipal networks as ‘city networks’ in this paper; the membership of these networks comprises municipal governments and representatives of these governments. Network membership typically does not include other organizations operating in the cities, such as non-government or private sector organizations.
While policy diffusion is another theoretical approach, we use policy learning as it is (1) analytically identical to policy diffusion through a network (Henry 2009), and (2) an easier concept for survey respondents to understand than policy diffusion and therefore more likely to return accurate results.
Hall (1993) refers to first-order learning as changes to the policy settings to fine tune the policy instrument as a result of previous performance; second-order learning as changes to the policy instruments used while maintaining existing policy goals; and third-order learning as changes to the hierarchy of the policy goals, and subsequent changes to the policy instruments and settings.
We use the terms information seeking, policy learning, and learning interchangeably in this paper given the similarity of their definitions.
The C40 network comprises the following cities: Addis Ababa, Amsterdam, Athens, Austin, Bangkok, Barcelona, Basel, Beijing, Berlin, Bogotá, Buenos Aires, Cairo, Caracas, Changwon, Chicago, Copenhagen, Curitiba, Delhi, Dhaka, Hanoi, Heidelberg, Houston, Ho Chi Minh City, Hong Kong, Istanbul, Jakarta, Johannesburg, Karachi, Lagos, Lima, London, Los Angeles, Madrid, Melbourne, Mexico City, Moscow, Mumbai, New Orleans, New York, Paris, Philadelphia, Portland, Rio de Janeiro, Rome, Rotterdam, Salt Lake City, San Francisco, Santiago de Chile, Sao Paulo, Seattle, Seoul, Shanghai, Stockholm, Sydney, Tokyo, Toronto, Warsaw, Yokohama.
The density of the network is calculated as (2 × 74)/2 × (33 × (33−1)).
Network analysis is to estimate how attributes of actors influence network structure. Here, network structure does not necessarily mean individual tie formation (in this case, who learns from whom as a dyadic relation) but the whole observed tie formation of the actors as one possible appearance of relations. Thus, examining how network structure influences attributes or behaviors of actors is a different domain to social network analysis. One example of such study can look at how centrality (one characteristics of network structure) influence cities’ climate change policies, using regression models.
This study investigates factors influencing learning within the network, and therefore, more detailed analysis of the types of climate change policies learned (i.e., the “what” was learned) is beyond the scope of this paper.
We would like to thank the four anonymous reviewers, the editor of Policy Sciences, Justin Robertson, and Bradley Williams for their helpful comments. Data collection for this manuscript is supported by City University Start-Up grant (7200233).
- Betsill, M. M., & Bulkeley, H. (2003). Cities and climate change: Urban sustainability and global environmental governance. London, NY: Routledge.Google Scholar
- Child, J., & Faulkner, D. (1998). Strategies of co-operation: Managing alliances, networks and joint ventures. Oxford: Oxford University Press.Google Scholar
- C40 Cities (2011). C40 cities: an introduction. http://www.c40cities.org/. Accessed 20 April 2011.
- Glasbergen, P. (1996). Learning to manage the environment. In W. M. Lafferty & J. Meadowcroft (Eds.), Democracy and the environment: Problems and prospects. Cheltenham: Edward Elgar.Google Scholar
- Handcock, M. S., Hunter, D. R., Butts, C. T., Goodreau, S. M., & Morris, M. (2008). statnet: Software tools for the representation, visualization, analysis and simulation of network data. Journal of Statistical Software, 24(1).Google Scholar
- Henry, A. D. (2009). The challenge of learning for sustainability: A prolegomenon to theory. Research in Human Ecology, 16(2), 131–140.Google Scholar
- Hunter, D. R., Handcock, M. S., & Butts, C. T. (2008). ergm: A package to fit, simulate and diagnose exponential-family model for networks. Journal of Statistical Software, 24(3).Google Scholar
- ICLEI (2011). ICLEI climate program. http://www.iclei.org/index.php?id=800. Accessed 18 April 2011 2011.
- Johnson, B., & Lundvall, B.-Å. (2001). Why all this fuss about codified and tacit knowledge? Paper presented at the DRUID Winter Conference January 18–20 2001.Google Scholar
- KPMG, Hendriksen, B., de Boer, Y., Copius Peereboom, E., Jansen, S., & Ballantine, J. (2011). CDP Cities 2011: Global report on C40 Cities. London, Amsterdam: Carbon Disclosure Project.Google Scholar
- Lipschutz, R. D. (1997). From place to planet: Local knowledge and global environmental governance. Global Governance, 3, 83–102.Google Scholar
- Newman, M. E. J. (2010). Networks: An introduction. Oxford: Oxford University Press.Google Scholar
- Scholz, J. T., & Stiftel, B. (2005). Introduction. In J. T. Scholz & B. Stiftel (Eds.), Adaptive governance and water conflict: New institutions for collaborative planning (pp. 1–11). Washington DC: Resources for the Future.Google Scholar
- UNPF (2009). State of world population 2009: Facing a changing world: women, population and climate. (New York, USA ed., Vol. United Nations Population Fund). New York, USA: UNPF.Google Scholar