Table 1.

Steps taken in the realist analysis

Step 1: Identifying outcomes (description)	This involved reading and re-reading the papers, first to gain familiarity with the studies and then subsequently to identify events (i.e. outcomes) which occur as a result of decentralization, i.e. how decentralized governance changes the actions, decisions and relations of health system actors. The outcomes of interest are changes in equity, efficiency and resilience within health systems—or the actions, decisions and relations of health system actors that may result in such changes.
Step 2: Identifying contextual components of outcomes (resolution)	The articles were further reviewed to identify important contextual components (enablers and constraints) of the identified outcomes. These include the formal and informal rules or institutional that govern the actions, decisions and relations of health system actors, the socio-economic circumstances of individuals, groups, communities and of entire jurisdictions, and circumstances related to the physical geography of a community, sub-national or national jurisdiction. In addition, context included peculiar design features and characteristics of decentralization in each setting.
Step 3: Theoretical re-description (abduction)	This involved situating identified outcomes and their contextual components within theories to better understand what they represent. Three theories resulted from and informed our analysis: We situated decentralization within a multi-level framework which defines governance at three levels: constitutional governance (i.e. governments at different levels functioning at different distances from health service operations on the ground), collective governance (community-based groups such as local health boards and community health committees or close-to-community governments with significant community input) and operational governance (individuals and providers within the local health market) (Abimbola et al., 2014). This multi-level framework focuses on the rules that distribute responsibilities and determine the relations among health system actors within and across levels of governance. It also highlights the dynamic relationship between the levels of governance as failure at a level can be compensated for by health system governance actors at the same or another level (Brinkerhoff and Bossert, 2014; Abimbola et al., 2017). We applied the three conceptual options available to communities and jurisdictions in the face of poor, sub-optimal or costly services: Exit, Voice and Loyalty (Hirschman, 1970). We adapted these to decentralized health system governance, such that ‘Exit’ occur by health workers or patients moving across communities or sub-national jurisdictions or between health service providers within a community or sub-national jurisdictions. However, when ‘Exit’, for whatever reason, is not an available option, health workers and patients or other people in the community or jurisdiction are constrained to ‘Loyalty’ and therefore use their ‘Voice’ through accountability channels available to them to improve the quality or reduce the cost of services. And when ‘Voice’ fails, ‘Loyalty’ in the absence of the ‘Exit’ option constrains the local health system actors to invest in and govern their own healthcare services, seeking to provide public goods where governments have failed.
	3. The transaction costs theory of the firm predicts that economic agents will organize production within firms (i.e. centralize) when the costs of co-ordinating exchange through the market are greater than within a firm (Coase, 1937). However, the distribution of the costs and benefits of centralization (and, by extension, of decentralization) between providers and users vary in different settings. On the provider-side, while larger, centralized, providers incur higher internal transaction costs (thus reducing efficiency), they can also leverage size in external transactions as they reap economies of scale, (but central decision-making may ignore local context). On the user-side, with centralized provision, users may benefit from reduced transaction costs (in the form of search and information costs, and monitoring and enforcement costs) due to recognizable homogeneity in products, prices and quality across operating units, and as they avoid repeated transactions with different providers (Abimbola et al., 2015). But users may also incur costs if centralized firms control a large share of the market (thus charging higher prices), or as they ignore local realities.
Step 4: Identifying mechanisms (retroduction)	This involved examining the identified outcomes and their contextual enablers or constraints with the aim of arriving at the reasoning processes and system capabilities that resulted in the observed patterns across countries. The reasoning processes and system capabilities were identified by moving back and forth between the empirical data the theories applied in this review to develop explanation for the identified pattern of outcomes and their contextual components.