10.2.1 What are Positive Deviant Cases?

Positive deviant cases are outliers that exhibit superior performance than the predictions of a model would hold. In technical terms, this refers to cases selected to maximize |Y_i – Ŷ_i|, or the difference between the actual value and the fitted value in a regression (Reference GerringGerring, 2007: 89; Reference SeawrightSeawright, 2016: 16). In order to qualify as cases of positive deviance, however, cases must be more than ephemeral outliers operating in an environment conducive to good performance. Rather, examples of positive deviant organizations or other types of collective actors must exhibit sustained high performance in a context in which good results are uncommon.

Writing from the perspective of organizational behavior in the management literature, Reference Spreitzer and SonensheinSpreitzer and Sonenshein (2004) present a somewhat distinct definition and operationalization of positive deviance that is especially useful when thinking about the role that deviant cases can play in the formulation and implementation of development policy. Their definition emphasizes a normative component of positive deviance, depicting the construct as behavior that departs from established norms in a referent group “in honorable ways” (Reference Spreitzer and SonensheinSpreitzer and Sonenshein, 2004: 832). This understanding differs from the more neutral statistical perspective in which deviance – whether positive or negative – represents behavior that diverges from average or normal experiences (Reference Spreitzer and SonensheinSpreitzer and Sonenshein, 2004: 830). In their analysis of positive deviance in agricultural development, Reference Pant and OdamePant and Hambly Odame (2009: 160), too, describe positive deviants as those who break from prevailing norms to promote positive change, which in their study is measured by the facilitation of agricultural knowledge creation and application in the instances they highlight. As I hope to show, an understanding of positive deviance that incorporates a technical, statistical definition while including attention to deviation from prevailing norms and practices in a given sociopolitical context is essential if we view positive deviant cases in both methodological and inspirational terms.Footnote ¹

10.2.2 Methodological Justifications

The literature on research design in the social sciences identifies a range of case selection strategies and highlights the pros and cons of each type for distinct research goals (Reference Collier and MahoneyCollier and Mahoney, 1996; Reference Eckstein, Gomm, Hammersley and FosterEckstein, 2009; Reference GeddesGeddes, 1990; Reference GerringGerring, 2007; Reference Przeworski and TeunePrzeworski and Teune, 1970; Reference SeawrightSeawright, 2016; Reference Seawright and GerringSeawright and Gerring, 2008). What, if anything, can be learned from positive deviance in the context of development policy?

The consensus view on the role of deviant cases is that they are most useful for exploratory purposes or discovery. Anomalous cases, whether positive or negative, are valuable for theory building exercises and to search for new but as yet unspecified explanations (Reference OdellOdell, 2001: 166; Reference SeawrightSeawright, 2016; Reference Seawright and GerringSeawright and Gerring, 2008: 302). They can also generate insights into the identification and operation of causal mechanisms not examined in the existing literature (Reference George and BennettGeorge and Bennett, 2005: 20; Reference GerringGerring, 2007: 89; Reference Seawright and GerringSeawright and Gerring, 2008: 303). In the case of positive deviance, then, these outliers can show how specific causal mechanisms propelled high performance, indicating how specific values on a previously unexplored variable propel superior outcomes.Footnote ²

Deviant cases are well suited to pinpoint sources of causal heterogeneity (Reference SeawrightSeawright, 2016: 21, 25). This goal is particularly relevant for development policy because it is increasingly recognized that social context affects the implementation and results of the same types of interventions and policies (Reference Pritchett and SandefurPritchett and Sandefur, 2013). Policy interventions that have been validated through experimental research in one country or subnational unit often do not yield the same results in another place, limiting the broader value of studies based on experimental designs for development policy (Reference Pritchett and SandefurPritchett and Sandefur, 2013; Reference Rodrik, Cohen and EasterlyRodrik, 2008: 26–27). An advantage of a qualitative approach to studying positive deviance is that contextual factors can be identified and integrated into an analysis of what has or has not worked in a given unit (Reference Bradley, Curry, Ramanadhan, Rowe, Nembhard and KrumholzBradley et al., 2009: 3). As a result, it is possible to study the way that social norms within an organization or in the community where it is embedded may affect the behavior of relevant actors and, hence, the performance of the organization as a whole.

At the same time, deviant cases have inbuilt limitations. By virtue of their status as outliers, they are obviously not useful for identifying the causal pathways of average or on-the-line cases. In addition, as Reference SeawrightSeawright (2016: 21) argues, they may not be as useful for identifying omitted variables as some have suggested (Reference Bennett and ElmanBennett and Elman, 2006). Furthermore, as is true for other types of case studies, deviant cases cannot on their own refute a theory. However, when a theory is based on a deterministic proposition, deviant cases can also be used to disconfirm a theory (Reference Seawright and GerringSeawright and Gerring, 2008: 302), although this is a less common causal premise in mainstream social science research.

It is also important to stress that identifying cases of positive deviance by selecting cases with exceptionally high values on the outcome of interest is subject to the usual pitfalls of selecting on the dependent variable (Reference King, Keohane and VerbaKing, Keohane, and Verba, 1994). If we only study cases of high performers, we are liable to falsely ascribe their success to factors that may also be present in less successful cases as well. For this reason, the convention in qualitative research is to select on the independent variable whenever possible. But analyses of positive deviant cases among a field of otherwise similar cases that operate in the same context or area can be a valuable way to identify potential explanatory variables for exceptional performance. The hypothesized explanatory variables can then be incorporated in subsequent quantitative or qualitative studies in which their effects are evaluated more generally.

Despite these words of caution, taking a positive deviance approach has the potential to change the way researchers or development practitioners think about the phenomenon or outcome in question because the goal of analyzing outliers is to identify factors that explain why it does not fit the general pattern. From a statistical perspective, the ultimate goal of analyzing deviant cases is to aid in the development of a new model or the revision of an existing model so that the variable responsible for the anomalous value of a case is incorporated. The insights generated from a close analysis of the positive (or negative) deviant case can then be can tested more generally in a broader set of cases. To the extent that this model works, the deviant case is no longer deviant (Reference Seawright and GerringSeawright and Gerring, 2008: 302–303). Deviant cases, then, can play an important role in revising the conventional wisdom on the determinants of a particular development outcome.

Positive deviant cases can yield especially valuable insights when situated in a larger mixed methods approach. In particular, if a qualitative analysis of high performers precedes a quantitative study, this affords the possibility of identifying previously unconsidered hypotheses or of devising measures that can tap into potentially important constructs. Process tracing is a valuable technique for identifying the factors that explain deviant performance (Reference George and BennettGeorge and Bennett, 2005, p. 215), which can then be assessed in a larger sample. For example, in their discussion of learning from positive deviance in health care delivery, Reference Bradley, Curry, Ramanadhan, Rowe, Nembhard and KrumholzBradley et al. (2009) present an approach that entails (a) the identification of organizations that have consistently demonstrated high performance on a clear set of indicators, (b) intensive analysis using qualitative methods to generate hypotheses about the specific practices and policies that led to exceptional performance, (c) the use of statistical tests in a larger, representative sample of organizations to assess the broader applicability of the hypothesized explanatory factors, and (d) the dissemination of best practices to other organizations operating in the same field. The discussion of a multistep research project on the determinants of quality in social service provision in Jordan, which was led by the World Bank, is another example of this approach (see Section 10.3.1).

10.2.3 Inspirational Justifications

A less technical perspective highlights a distinct potential benefit of focusing on deviant cases, and especially positive deviant cases, with notable benefits for development policy. The identification and exploration of high performers can be motivating in and of itself through a signaling effect. In the context of low-resource environments, which development policy tends to target, cases of success against the odds indicate to others that there is hope. When the prospects for advancement seem dim, examples of high performers signal that it is possible to break, or at least start to weaken, development traps.

Actors embedded within resource-constrained communities may regard the experiences of other actors or organizations in similar contexts as applicable to their own circumstances, increasing the probability that they will seek to emulate their practices. This dynamic may facilitate the diffusion of best practices, in the same way that activists in the Arab region or in other regional waves of mass mobilization emulated the strategies and tactics of first movers because they perceived that they faced similar opportunities and constraints in their own countries (Reference Patel, Bunce, Wolchik and LynchPatel, Bunce, and Wolchik, 2014). In this sense, positive outliers can help to stimulate a sense of “cognitive liberation” (Reference McAdamMcAdam, 1999), whereby practitioners and policy-makers come to believe that change is not only desirable, but possible. The mere act of shedding feelings of resignation and hopelessness can propel positive change by motivating greater effort and stimulating collaboration toward shared community goals.Footnote ³

In the day-to-day design and implementation of development policy, highlighting examples of positive deviance is likely to play a far more constructive role than pinpointing instances of underperformance or negative deviance. Positive incentives can be a source of motivation, potentially inducing greater effort and commitment to problem-solving among stakeholders, whereas sanctions may have a deterrent effect by increasing the sense of resentment and hostility to change (Reference Ryan and DeciRyan and Deci, 2000). Thus, just as emphasizing instances of success may galvanize positive change, policy-makers and development practitioners must tread lightly when advertising success stories lest others in the community resent the fact that they, too, were not the objects of praise. As I note below, this is precisely the reason why Moroccan officials were hesitant to include identifying information in a short case study of high-performing public health centers included in a World Bank report (Reference Belkâab and CammettBelkâab and Cammett, 2014; Reference Brixi, Lust and WoolcockBrixi, Lust, and Woolcock, 2015). For these reasons, how the lessons of positive deviance are shared can affect the probability that other organizations will adopt them. If the management of organizations with less-distinguished records feels threatened or marginalized when learning of the exceptional performance of others, then it will be difficult to foster acceptance of the identified best practices. If, on the other hand, others can be made to feel as if they are partners in the identification and scale-up of effective strategies, then dissemination may be more likely.

In short, in development practice it is imperative to distinguish between the empirical identification of positive deviant behavior for methodological purposes and the inspirational role of positive deviant actors or organizations in driving beneficial outcomes. Positive deviants in the context of development policy and outcomes may be more than statistically anomalous cases. From the standpoint of effecting tangible change in public goods provision or other development policy goals, they may be actors or groups who drive innovation and bring about solutions to problems that are not well addressed under the status quo; alternatively, they may just be those who, on an everyday basis, seek to find ways to manage existing resources in a way that yields strong performance against the odds. As such, an analysis of deviant cases can serve at least two roles. First, a close analysis of the actions and behaviors of deviant actors, and especially the identification of causal pathways linking these to outcomes, provides explanations for why a handful of cases exceeded expectations. Second, in the design and implementation of development policies and programs, positive deviants can play an important signaling role by demonstrating to others who toil under equally challenging circumstances that success (or at least improvement) is possible.

10.2.4 Selecting Cases of Positive Deviance

A systematic approach to the selection of positive deviance cases is important to avoid biases that can result in missed lessons or misleading interpretations of the causal factors leading to exceptional performance. Under ideal circumstances, case selection must occur with reference to a broader population of cases that are identified based on a general causal model. This approach, however, is contingent on the availability and validity of information on the population – a condition that is often hard to meet in some countries, whether due to lack of information or lack of transparency – and on the quality and nature of the model (Reference GerringGerring, 2007: 106; Reference SeawrightSeawright, 2016).

Given a conventional understanding of the determinants of the outcome of interest, then, positive deviant cases can be selected because they are off the regression line in the direction corresponding to high performance, or outcomes that imply beneficial or honorable behavior or outcomes (Reference Spreitzer and SonensheinSpreitzer and Sonenshein, 2004). Yet positive deviant cases are more than just outliers. Rather, to qualify as examples of positive deviance, high performers must excel in a context in which most other comparable units do not perform well. To pinpoint such cases, case selection might proceed in a two-stage process in which the research first highlights low performers in a high-deviance region and then identifies high performers in an area characterized by low overall results (Reference WoolcockWoolcock, 1998). At the same time, such exceptional performance in a field of otherwise poor results must be sustained lest it arises due to chance, ephemeral factors, or particular circumstances.

As with any model, the utility of a regression line approach to selecting positive deviant cases also depends on the ways in which the outcome of interest is conceptualized and measured. Many goals central to development policy are multidimensional. A case or unit may perform exceptionally well with regard to some aspects of an outcome but may exhibit average or inferior values with respect to others. For example, in analyses of the quality of health care, the empirical focus of the examples discussed in the next section, dependent variables can focus on the infrastructural, process-oriented, or health outcome aspects of quality (Reference DonabedianDonabedian, 1988). Which aspects of quality are emphasized may be self-evident if there is a professional consensus on which dimensions are the key drivers of human development outcomes, or the choice may depend on the objectives of the researcher or development program.

In practice, it may be impossible to employ such a systematic approach to case selection due to a lack of data or restrictions on access to data. Should case studies of positive deviance then be abandoned? Not necessarily. It may be possible to identify actors or organizations that are deemed to be high performers in a less rigorous way by drawing on insights from local officials or community members. These actors know the terrain exceptionally well because they are embedded in the communities where policies or programs are implemented and have implicit or explicit rankings of local actors with respect to performance on development outcomes. Feedback from local actors can be elicited through interviews or by the administration of a survey, depending on resource availability.

That said, the measurement of positive deviance based on the reports of key informants is obviously vulnerable to a host of biases, even when respondents have the best of intentions. For example, officials or other stakeholders may inadvertently provide invalid information due to recall biases or because they wish to highlight favored actors or organizations at the expense of others. At a minimum, when cases are selected purposively, it is important to choose cases for in-depth analysis from a diverse sample that includes organizations varying in size, resource endowments, geographic location, and other factors that might affect performance (Reference Bradley, Curry, Ramanadhan, Rowe, Nembhard and KrumholzBradley et al., 2009: 3). As an alternative or supplemental approach, researchers may draw on findings and insights derived from studies in other, comparable settings to use as a benchmark for the determination of high-performing organizations or actors.

10.3.1 Jordan

In March 2014, I was part of a team of researchers that visited six health centers in four of Jordan’s twelve governorates, collecting data for a case study of positive deviance in the delivery of primary health care in the Jordanian public health system. The first order of business was to identify a set of indicators against which high performance would be assessed. A growing body of research emphasizes that several process-oriented aspects of health care quality, such as provider effort and human resource management at the facility level, are key factors contributing to the poor quality of health care in developing countries, with consequences for health outcomes (Reference Das and HammerDas and Hammer, 2014; Reference Dieleman, Gerretsen and van der WiltDieleman, Gerretsen, and van der Wilt, 2009; Reference Harris, Cortvriend and HydeHarris, Cortvriend, and Hyde, 2007: 450–452). At the same time, the quality of physical infrastructure may enhance or inhibit the ability of health workers to fulfill their professional obligations, and therefore it was also imperative to collect data on the availability and condition of medical equipment and supplies.Footnote ⁴

In order to identify appropriate cases, we aimed to benchmark the performance of the selected centers in the past three to five years against others within the same urban or rural region with similar socioeconomic profiles. Given the limitations of time, resources, and existing data on service quality, case selection faced significant challenges. In the end, the choice of facilities relied heavily on insights provided by the Ministry of Health and referrals made by the Health Care Accreditation Council (HCAC),Footnote ⁵ a nonprofit organization in Jordan that implemented an accreditation program funded by USAID. As a result of its role in identifying appropriate health centers to undergo the accreditation process, HCAC staff members had collected a database on a sample of high-performing facilities in the public health system and were willing to advise us on case selection even though they were not permitted to share the full database with us. Input from the HCAC was especially valuable for the purposes of carrying out case studies of positive deviance because the participating facilities had been selected in the first place due to a longer record of high performance, which was enhanced after completing the accreditation program.

During site visits, the team carried out a series of open-ended interviews with the chief medical officers, doctors, nurses, administrative staff, and, where applicable, the members of local health councils to gather information on multiple indicators of health care quality and to probe the underlying drivers of the quality of care. In conjunction with administrative data provided by the Ministry of Health and the centers themselves and with interviews with local, regional, and national government officials, the data collected during these visits provided the information used to write the case studies. Given that the accreditation process itself was an important driver of quality, the team was careful to collect data illuminating performance on quality indicators prior to and following participation in the program.

The findings of in-depth studies of the selected health centers yielded a variety of insights about the determinants of high performance in the delivery of primary health services in Jordan. The fact that the accreditation program was implemented in a nonrandom fashion and the design of the study prevented an analysis of the independent effects of participation in the program. However, in-depth interviews at the facility level clearly demonstrated that the act of preparing for and undergoing the evaluation for accreditation drove major improvements in the quality of care, even at facilities that already performed above their peers. In particular, the findings indicated that preparation for accreditation in and of itself provides one mechanism for improving administrative procedures and the quality of service delivery. The procedures and requirements of the accreditation process helped to establish clear rules and regulations, increase transparency to clients and staff, develop more effective staff monitoring, and give greater voice to and participation for staff within facilities as well as communities.

Beyond adherence to externally imposed standards and practices, leadership emerged as a key factor improving the operation of facilities. Indeed, strong, proactive chief medical officers at the health centers multiplied the positive outcomes from reforms implemented for accreditation or for other purposes. The head of one health center emphasized that he set clear expectations for his employees, fostered a collaborative work environment, and ensured that adequate feedback mechanisms existed for staff members to convey their concerns. Conversely, administrative reforms were less effective under weak leaders. For instance, the chief medical officer in one center noted that staff meetings were not held, since people came on different days and the assumption was that “everyone knows their job.”

A third finding pointed to the role of social networks in facilitating voice and participation and potentially for improving the extent and quality of service provision. Personal ties, whether among family or friends and neighbors, are particularly valuable for establishing priorities, extending public health outreach in the community, and mobilizing resources to support the activities and development of health centers. Interviews with staff and members of local committees at the selected facilities indicated that shared identity, especially a common tribal affiliation, was especially valuable in rural areas but somewhat less relevant in urban centers, where people from diverse regions and backgrounds intermingle and many residents do not come from the major Transjordanian tribal families. In particular, when members of local health committees and local residents hail from the same tribes, the staff and governing board of a given center had an inbuilt channel through which to reach the community with vaccination and other health campaigns and to encourage greater compliance with medical advice. At one center, members of the local health committee claimed that the center’s public outreach initiatives are relatively successful because they come from the same families as the targeted beneficiaries.Footnote ⁶ Beyond family ties, we also found that social networks and linkages to important local social institutions such as mosques, youth centers, and local schools also assist health workers to accomplish their tasks. This insight provides a clear example of the ways in which case studies of positive deviance can generate new variables and causal pathways.

The information gathered through the case study of successful facilities helped to inform a subsequent quantitative study that the team undertook in Jordan. Based on a national sample of 100 health centers, the analysis aimed to explain variation in standard measures of provider effort, including absenteeism, adherence to clinical practice guidelines, rights-based practice, and time spent with provider. The main independent variables captured various aspects of within-facility governance, a neglected variable in the public health literature on quality and, more generally, in research on governance and the quality of service delivery. These included indicators to capture the presence of various management practices at the health center, including monitoring of professional staff, sanctions for underperformance, and financial and nonfinancial rewards for good performance. The main control variables tried to address both top-down monitoring and a limited component of citizen engagement, notably the presence of a community health committee within the local primary health center (PHC). Many other factors were not measured, however, due to a lack of resources and time. The findings indicated that monitoring is the most consistent predictor of improved provider effort, including adherence to clinical practice guidelines, the provision of rights-based and responsive care, and time spent with patients in clinical examinations. When considered independently, sanctions were either not associated with provider effort or were associated with worse provider effort, a finding that fits with existing studies in the management literature on the importance of positive work environments (Reference EdmondsonEdmondson, 2003) and accords with what we saw at high-performing facilities in the case studies. The use of nonfinancial rewards to recognize good performance was not associated with provider effort, but this result likely obtained because the usage of this approach was too infrequent to be able to detect an effect.

Due to resource constraints, the study could not incorporate much attention to the impact of social context on the quality of service delivery. However, insights on the role of social ties in driving higher-quality health care derived from the case study of positive deviance facilities in Jordan have informed subsequent projects that my collaborators and I have undertaken. Thus, while we were not able to fully capitalize on the findings from deviant cases in the quantitative study in Jordan in order to build a new, more comprehensive model, we aim to do so in future research in relevant contexts.

10.3.2 Morocco

In May 2014, I traveled to Morocco to carry out a parallel case study of positive deviance primary health centers for the aforementioned World Bank report (Reference Brixi, Lust and WoolcockBrixi et al., 2015). Initially, the strategy for case selection adopted a systematic approach by identifying positive outliers from data collected through a quantitative, nationally representative sample of the quality of primary health care in public centers carried out by the World Bank and the Ministry of Health (MOH). Given time constraints, it was not possible to visit all the centers I initially requested and, instead, I selected centers in several provinces with guidance from national and regional MOH officials. Although the sampling procedures were nonrandom, several systematic criteria guided the choice of facilities. First, half of the facilities participated in the Concours Qualité (CQ), a program introduced by the MOH to set up province-level competitions between public health facilities to be recognized for exceptional performance.Footnote ⁷ Second, they exhibited notable and sustained improvements in one or more key dimensions of the quality of health care provision in the last five years. Third, the centers were not located in major urban areas, where quality tends to be somewhat higher.

A consistent array of factors was observed at well-performing Moroccan public health centers, including larger patient loads; reasonable wait times despite high demand; good management and availability of stocks, consumables, and equipment; and detailed and regular maintenance of patient medical records, among other factors. Based on interviews with staff members at the facilities and MOH officials from multiple administrative jurisdictions, at least five factors seemed to be associated with improvements in the quality of care. First, the presence of dynamic, energetic, and visionary leadership helped to motivate staff members to carry out their duties competently and thoroughly, introduce new procedures and management systems, institute a “culture of quality” among staff members, inspire confidence in the community and local government officials, and attract additional resources to the facility. Second, a sense of a shared mission and collaborative ethic helped to motivate the staff and ensure that all staff members know and fulfill their responsibilities. Like leadership, a team spirit and a relatively flat organizational culture are especially important in the context of resource scarcity, in which staff members are required to make do with less.

Third, meetings with staff members from facilities that had and had not previously competed in the national CQ program demonstrated that the initiative has had a clear, positive effect on the management and administration of participating centers. The mere act of enrolling in the program generates significant transfer of knowledge and the adoption of new procedures within participating health centers. At the same time, it became clear that the program may be unsustainable because it requires a major investment on the part of staff members, taking them away from their primary professional duties, and has the paradoxical effect of increasing pressures on the successful facilities by boosting their patient load.

Fourth, effective coordination between the head doctor and local health officials was critical for the health centers to meet the needs of the populations in their catchment areas. Regular exchanges between the administrators of facilities and officials from the provincial delegation help to ensure that stock-outs of medications and equipment do not occur, that facilities receive resources when available, that local solutions are developed for local problems, and that good administrative and management practices are disseminated. Finally, partnerships with nearby groups, organizations, and prominent individuals help health centers to meet the needs of their surrounding communities more effectively. These findings were elaborated in a case study of positive deviance in the Moroccan primary health sector (Reference Belkâab and CammettBelkâab and Cammett, 2014). It is important to note that MOH officials explicitly requested that we refrain from naming specific health centers that performed exceptionally well in order to avoid resentment among staff members at other facilities not featured in the case study.

10.3.3 The Contributions of Positive Deviance Case Studies in Jordan and Morocco for Mixed Methods Research Designs

Case studies of high-performing health centers in two countries in the Middle East and North Africa suggest that, even in different sociopolitical contexts, some common factors at the facility and community levels affect the quality of service delivery. For example, leadership and management practices were important proximate determinants of quality in both contexts. This finding helped to inform a subsequent quantitative study in Jordan centered on the factors within facilities that incentivize greater provider effort.

Perhaps more interesting are the findings that facility health committees composed of elites, who hold either formal or informal positions of influence at the local level, and the extent and nature of social ties may have important causal effects on the quality of social services. From the perspective of the large body of social science research on clientelism and elite capture (Reference Bardhan and MookherjeeBardhan and Mookherjee, 2012; Reference Dasgupta and BeardDasgupta and Beard, 2007; Reference KhemaniKhemani, 2015; Reference PlatteauPlatteau, 2004), the potentially constructive role of elites in driving improved process and outcome-based measures of health care quality may be surprising and deserves greater investigation. This finding raises questions about the conditions under which elites are likely to exert efforts to improve the quality of services and, more generally, to play a positive role in promoting inclusive access to services for local communities and to encourage citizens to adopt more health-seeking behavior. They also call for additional research into the nature of social ties and how they may mediate the effects of different types of interventions on the quality of social services. In both Jordan and Morocco, at least some local elites appear to have leveraged their influence to improve the quality of care delivered and to ensure that nonelites take greater responsibility for their own well-being.