Study eligibility

We included studies focusing on displaced populations, including refugees, asylum-seekers, forcibly displaced persons, or other IDPs. We considered studies whose participants were at least 18 years old, and who were willing and able to engage in a livelihood activity. We excluded studies focusing on economic migrants or host community members who were not themselves displaced.

We included studies that explicitly described the intervention as the “Graduation approach” or “Ultra-Poor Graduation model,” or that directly referenced BRAC’s “Targeting the Ultra-Poor Programme” as a conceptual template. We only considered programs that included at least three of the following four elements: 1) an asset transfer, 2) consumption support, 3) savings support, and 4) a coaching intervention. We chose these four elements after conducting preliminary scoping reviews, as our findings suggested that most Graduation programs include a combination of these four “pillars” rather than BRAC’s original six core components [9]. We excluded studies of programs with fewer than three components; studies that did not explicitly reference the “Graduation approach” or BRAC’s model; and studies that gave inadequate detail for us to determine what the program entailed.

Only quantitative studies were required to have explicitly defined comparator groups, which we defined flexibly (e.g., similar populations not receiving the specified intervention, participants’ own baseline measures, or formally randomized control groups). We imposed no restrictions regarding comparison groups for qualitative studies.

Our primary outcomes of interest were food security, employment, living conditions, social wellbeing, poverty, dependence on humanitarian aid, and mental health, as measured at the end of program participation or at later follow-up. We intentionally applied broad definitions of these primary outcome domains, considering the substantial heterogeneity of reported outcomes we identified in scoping reviews. As such, we did not exclude any otherwise relevant studies based on specific reported measures of effect. Secondary outcomes included barriers and facilitators (which are factors that hinder or help program implementation, respectively), and programmatic feasibility and sustainability with displaced populations. Feasibility is an intervention’s relevance, suitability, or practicality—typically most salient in early implementation—while sustainability refers to an intervention’s maintenance over time, integration into standard practice, and sustained routine use—typically most salient in later implementation stages [20]. We extracted data for all compatible measures at all available timepoints from each study.

We excluded documents that were summaries or alternate publications of included studies; purely theoretical articles; articles summarizing evidence from multiple countries; program promotional materials; and articles not describing person-level results. We placed no further restrictions on the types of studies we considered for inclusion.

Search strategy

We searched eight databases (EconLit, Academic Search Ultimate, Family & Society Studies Worldwide, APA PsycINFO, PubMed, Web of Science, Global Health, and the Cochrane Library). We formulated a search strategy for PubMed (see Table 1) and subsequently adapted it for each database. Our complete search strategies are provided in S1 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Database search strategy.

https://doi.org/10.1371/journal.pstr.0000252.t001

When reviewing the grey literature, we hand-searched websites of the following key implementing agencies: Innovations for Poverty Action (IPA), Trickle Up, Poverty Alleviation Coalition, BRAC, Concern Worldwide, World Bank Group Partnership for Economic Inclusion (PEI), Mercy Corps, UNHCR, World Vision, Association of Volunteers in International Service (AVSI), and CARE. We also reviewed the reference list of a key overview publication [21]. We restricted our searches to documents published in 2002 or later. We conducted initial literature searches from March-April 2023, with further grey literature searches in April 2024 after learning of more recently published documents relevant for our review. We conducted a final updated database search in October 2024.

Study selection and data extraction

Two authors (LY and YW) independently screened document titles and abstracts and removed duplicates using the web-based software platform Covidence [22]. Studies were included for full-text review if both reviewers agreed upon their eligibility. During full-text review, LY and YW independently reviewed the articles and decided upon their eligibility for inclusion. All discrepancies were resolved through discussion.

LY and YW extracted data independently using identical Covidence templates with distinct columns for each outcome domain of interest before comparing extraction results and resolving them into a final document. We recorded study characteristics (full citation, program name, study type, comparison group type, country of implementation, year(s) of implementation); study context (implementation setting, notable details); participant characteristics (number of participants, type of displacement, targeting or inclusion/exclusion criteria, household characteristics); intervention characteristics (program components, duration and sequencing of components, format of implementation, adaptations made); primary effectiveness outcomes (measures of food security, employment, living conditions, social wellbeing, poverty, and humanitarian dependency); and secondary implementation outcomes (barriers, facilitators, feasibility indicators, and sustainability indicators). Where data were missing or unclear, we recorded them as such in the data extraction template.

We did not pre-specify effect measures of interest for each outcome, due to anticipated heterogeneity between studies and accounting for the mixed-methods nature of this review. To facilitate cross-comparison of outcomes, we used our data extraction template to organize each study’s reported outcomes by domain (i.e., food security, employment, etc.). Missing information about any outcome was marked as such during data extraction and noted in our narrative synthesis. We assessed the quality of each study using the corresponding measure from the Mixed Methods Appraisal Tool (MMAT) for systematic mixed studies reviews [23].

Analysis

We reviewed and compared findings across studies using our data extraction template. We narratively synthesized all data from qualitative and quantitative studies as the level of anticipated heterogeneity in reported outcomes precluded meta-analysis. Our unit of analysis was the study, rather than publication (multiple publications sometimes reported on the same participant cohort at a given project site; when referring to a study, we have triangulated findings from all available publications about a given cohort/project). In Covidence, related publication records were grouped together and indexed as a single study.

Characteristics of included studies

Out of 575 records screened, five distinct studies (reported in six publications) met inclusion criteria. Fig 1 shows a PRISMA flow chart of the screening and assessment process, specifying the number of records assessed at each step [17]. After screening out irrelevant references, most publications failed to meet key inclusion criteria; they often lacked any measure of participant-level results or their participants were not displaced. Notably, no studies meeting our inclusion criteria were indexed in databases of peer-reviewed literature. All non-duplicate studies identified in the literature search, including those excluded from the analysis, are described in S2 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flow chart.

https://doi.org/10.1371/journal.pstr.0000252.g001

Several notable studies were excluded at the full text review stage, despite their strong apparent relevance for the study questions. We excluded brief reports and case studies about UNHCR’s pilot projects in Ecuador, which provided insufficient detail about the components [24] and failed to report any results or comparators [14,25]. We also excluded a mid-term evaluation of UNHCR’s graduation program in Egypt due to the lack of a clearly identified consumption support pillar [26]. Finally, we excluded a case study [27] and a multi-national learning workshop report [28] relating to UNHCR and Self-Help Africa’s Graduation project in Zambia that did not report participant outcomes.

The five reviewed studies included two projects in Uganda [29,30], one in Colombia [16,31], one in Mozambique [32], and one in Costa Rica [15]. Key characteristics of these publications are described in Table 2. See S3 Table for the final data extraction document, which includes all specific quantitative results available from the included studies alongside summaries of qualitative findings.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Key characteristics of included studies.

https://doi.org/10.1371/journal.pstr.0000252.t002

The reviewed publications’ quality is described in Table 3 according to the MMAT framework [23]. Each publication included at least some elements that were either of poor quality or impossible to assess; quality determinations as recorded are applicable to all outcomes that were possible to assess in the indicated study. The study from Costa Rica did not feature clear research questions, so it was not appropriate to proceed with further quality evaluation using the MMAT [15,23].

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Quality assessment using MMAT.

https://doi.org/10.1371/journal.pstr.0000252.t003

Program effectiveness results

Food security. Three of the five studies reported improvements in participants’ food security, with broad heterogeneity in outcome measures. The first study in Uganda (representing the first cohort of a continually supported project) reported highly significant increases in food security index scores (comprising measures of food consumption, food insecurity, and children’s length/height-for-age) for all treatment arms relative to controls [29]. The second study in Uganda—a midline evaluation of the second cohort, whose participants received adapted intervention components—did not report food security or related measures [30]. In Colombia, eight participants (out of 11 total) reported in qualitative interviews that they consumed more food, ate more frequent meals, and had more diversified diets during the program [16]. In Mozambique, participants reported “lower food insecurity” and had a 40% lower probability of skipping meals than the control group [32]. UNHCR’s program report from Costa Rica did not assess food security outcomes [15].

Employment. The reviewed studies reported few employment-related outcomes except for the first Uganda cohort study, which found that program participants in the most comprehensive treatment arm spent significantly more time working than controls [29]. In Colombia, seven participants (out of 11) reported diversifying their productive activities during the course of the program, while three others did not; four households who previously engaged in “dependent activities” (e.g., wage employment) increased their time spent on “independent activities” (e.g., self-employment) [16]. Three participants also managed to reinvest their profits by the end of the program [16]. The Costa Rica study reported endline unemployment rates of 4.6% in the 2014 cohort (versus 37.9% at baseline) and 4.3% in the 2015 cohort (versus 30.4% at baseline) [15]. Neither the second-cohort midline study in Uganda [30] nor the study in Mozambique [32] reported employment outcomes.

Living conditions and quality of life. Four of the five studies described improvements in participants’ quality of life during the Graduation programs. The first-cohort study in Uganda reported significant improvements in participants’ access to basic drinking water and in handwashing with soap for the most comprehensive treatment arm (with significant improvements in handwashing only for the other treatment arms) relative to controls [29]. In Colombia, six participants (out of 11) reported making refurbishments, extensions, or renovations to their houses during the timeframe of the project [16]. In Costa Rica, 91.2% of displaced respondents with complete case data finished the project with a valid form of refugee documentation [15]. In Mozambique, over 10% of participants bought or constructed a new house, and over 10% invested in electrical grid connections [32]. The second-cohort study in Uganda did not describe specific living conditions or quality-of-life improvements [30].

Social wellbeing. Studies reported various improvements in participants’ social wellbeing according to heterogeneous assessment measures. The first-cohort study in Uganda reported a highly significant increase in participants’ subjective wellbeing for all treatment arms relative to controls, measured as an index derived from participants’ Kessler 6 (K6) score (a short screening measure of psychological distress) [33] and overall life satisfaction [29]. In the second-cohort study in Uganda, host community participants receiving the full intervention (modified to include a group therapy component) scored lower on optimism and social wellbeing indices than their counterparts receiving the same support without the therapy component [30]. All participants in Colombia reported qualitative improvements in self-esteem, self-value, and self-confidence; some also reported changes in their “aspirations and expectations” [16] or “increases in agency” [31]. The study in Costa Rica reported a 9% decrease in the number of participants not engaging in “organized groups” over the course of implementation [15]. Of all participants in Costa Rica, 57.9% reported improved emotional wellbeing by the end of the program, with 25.4% reporting no change, 12.3% reporting worse emotional wellbeing, and 4.4% declining to answer [15]. In Mozambique, participants reported “significant increases in levels of trust in the other group” (i.e., refugee vs. host community members) by the end of the program, compared to controls [32]. The authors also reported that “more participants believ[ed] that both communities [i.e., Mozambiquans and refugees] should be equally prioritized for employment” in circumstances where few jobs are available [32].

Poverty. Measures of poverty varied widely between studies, generally finding positive results. In the first-cohort study in Uganda, participants in the all intervention arms had highly significant increases in monthly income, the value of their productive assets, and monthly consumption relative to controls [29]. The second-cohort midline evaluation study in Uganda found that host community participants receiving the most comprehensive treatment (including therapy) scored somewhat lower than participants in the regular intervention on an “economic activity index” (comprising information about labor supply, investment, and intention to expand their enterprise); these differences were driven by disparate levels of investment in income-generating activities [30]. The study in Colombia reported qualitative improvements in savings and reinvestment, debt management, spending, and domestic assets for most or all participants [16]. Among Costa Rica’s participants with full-case data, 67.5% reported increased income since program baseline, while 18.4% had decreased income, 9.6% maintained their income level, and 4.4% did not answer [15]. By the end of the project, 78.9% of participating households reached a monthly income level at least equal to the national minimum wage, while 16.7% did not reach that threshold, and 4.4% did not answer [15]. In Mozambique, participants’ household income was 1.3 times greater than controls’ by the end of the program [32]. Participants also reported saving an average of $14 per month by the end of the program, while controls reported no savings at all [32].

Dependency on humanitarian aid. While dependency on humanitarian aid is often cited as a key problem of protracted displacement that justifies the need for livelihood interventions [24,26,34,35], none of the reviewed studies assessed reliance on aid.

Mental health. Few studies assessed mental health outcomes. The second-cohort study in Uganda measured mental health via an “aggregate measure of psychological distress” including the Kessler-10 [33], Patient Health Questionnaire 9 (PHQ-9) [36], and Generalized Anxiety Disorder 7 (GAD-7) questionnaires [30,37]. Host community members receiving therapy scored higher on this measure of distress than those in the regular intervention arm, but no significant effect was found among refugee participants [30]. UNHCR’s study in Costa Rica simply reported that 26.5% of participants or their household members received counseling [34]. Other studies did not mention mental health impacts of the Graduation programs.

Program implementation results

Implementation barriers. Barriers to implementation were most extensively addressed in the qualitative research in Colombia. Participants in Colombia experienced multiple livelihood-relevant shocks and obstacles throughout the course of their lives, including illness, employment problems, government corruption, household conflict, regional insecurity and migration shocks, enterprise failure or inability to expand, child school dropout, police involvement with the family, debt, climate shocks, and family deaths [16,31]. Participants’ physical isolation from urban centers in Colombia limited the extent of their market integration [16]. The authors reported that an “adjustment phase” during implementation had interrupted program activities and communication, leading to a “sense of abandonment and mistrust” that hindered participants’ further progress in the program [31]. The first-cohort study in Uganda highlighted infrastructure problems, including poor roads, heavy rains, distance between participating households, and COVID-19-related restrictions such as evening curfews that limited staff working time [29]. Other studies did not identify implementation barriers.

Implementation facilitators. Qualitative research in Colombia described specific facilitators of project implementation in greatest detail. Coaches in Colombia highlighted that having a profitable business was essential for participants’ ability to pay off their debts [16]. In households where both members of a couple were formal project participants, families achieved higher levels of productive assets [16]. The authors concluded that the highest levels of perceived wellbeing were typically found among participants facing less severe poverty [16]. The use of a tablet-based app was viewed favorably in this context, and was more expedient and practical than paper-based materials for participants with low literacy [16]. Finally, factors thought to improve participants’ education and work prospects included road quality, distance from educational centers, and quality of available education in the municipality [16]. The first-cohort study in Uganda additionally highlighted how group coaching (rather than one-on-one) facilitated cost savings in the treatment arm and was estimated to be the most cost-effective version of the intervention [29].

Feasibility. Studies in Uganda, Colombia, and Costa Rica commented on feasibility. In the first-cohort study in Uganda, key indicators of feasibility were participants’ stability (i.e., likelihood of remaining in their current location rather than repatriating) and their level of interest in the intervention activities [29]. In Colombia, the feasibility of establishing livelihood activities was negatively affected by climate shocks, which caused participants to lose livestock and income [16]. The authors advocated for flexibly adapting participant recruitment criteria and program components to improve the program’s suitability for diverse participants [16]. Finally, because coaches were considered crucial for the program’s success, the authors highlighted the importance of managing coaches’ caseloads to ensure they could maintain a practical workload [16]. In Costa Rica, program attrition revealed feasibility issues among certain participant subgroups. A total of 156 participants (27.5%) were dropped from the program due to non-commitment, leaving the country without notice, or losing contact with organization [15]. In this context, more refugee participants completed the program (39.6%) than local community participants (22.3%); the authors attributed this difference to local women’s distinct vulnerability statuses and to differences in the intervention [15]. A feasible time frame for implementation in Costa Rica was estimated to be 13–24 months, during which time 72.6% of their “graduated” participants were able to complete the program [15].

Sustainability. Only the qualitative research in Colombia provided direct recommendations for program sustainability. Land tenure and environmental shock preparedness were important for sustaining agricultural activities in this context [16]. The authors also suggested that the program could better support sustainable enterprises by emphasizing participants’ inter-reliance—ideally, by promoting a range of complementary productive activities to minimize redundancies and competition in restricted markets [16]. Finally, the authors suggested that government management of Graduation programs as poverty reduction schemes could ensure continuity and consistency of implementation [16].

Effectiveness

Programs’ reported outcomes were generally positive, but only the “Graduating to Resilience” projects in Uganda utilized randomized controlled trial designs to support causal interpretations of measured improvements (i.e., attributing them to the effect of participating in the programs) [29,30]. Specific outcomes and measurement approaches for each outcome domain varied substantially across studies. While most studies lacked the ability to rigorously attribute observed changes to the interventions, their results nevertheless illuminated specific points of interest that implementers and researchers can consider in future work.

While food security outcomes were generally good, few studies provided quantitative estimates, and only the first-cohort study in Uganda included well-specified measures [29]. Several food security measures are commonly used in humanitarian programming, including the Household Hunger Scale [38], Food Insecurity Experience Scale (FIES) [39], Food Consumption Score [40], or Reduced Coping Strategies Index [41]; future Graduation program studies may consider using such measures to enhance comparability. It may be worthwhile to also consider nutrition-related indicators of child growth such as stunting, wasting, and underweight [42], as did the first-cohort study in Uganda using a “food security index” Z-score that combined information from these domains [29]. However, while such indices have the advantage of drawing from multiple domains, implementers may have difficulty constructing and interpreting them. From an implementer (and donor) perspective, reporting specific food security-related domains independently may be more interpretable and actionable.

The qualitative research in Colombia highlighted the diversity of factors affecting participants’ food security, including incidental improvements in the food security landscape (e.g., new grocery stores) and other (unspecified) co-existing interventions [16]. This raises an important question about how simultaneous support from multiple projects or sources—sometimes reflecting apparent redundancies in service provision—truly affect program participants. Previous projects have strictly excluded potential participants who already receive comparable support [27]. This is a common approach in displacement settings marked by limited short-term funding for lifesaving aid. However, these findings suggest that apparent “redundancies” could potentiate the program’s effectiveness regarding food security and possibly other domains as well. This possibility needs to be explored further, and its implications could differ substantially by context (a theme to which we will return frequently). At the very least, these findings caution against excluding participants a priori because they receive concurrent support. Participating in multiple programs may allow participants to experience greater synergistic poverty reductions than in “standalone” programs aiming to reach many people with a minimum level of support. In fact, this concept engages a similar logic to that of the multifaceted Graduation model itself—which presumes that concurrent support across multiple domains is more effective than intervening on only one.

Surprisingly few employment outcomes were reported, given that the Graduation approach is fundamentally a livelihood intervention. The exception was again the first-cohort study in Uganda, which quantified participants’ total working time [29]. The other reviewed studies had limited interpretability due to a lack of nuanced employment indicators [15]. For example, some studies focused on the difference between “independence” (e.g., self-employment) and “dependence” (e.g., wage employment) as a key livelihood outcome, with “independence” presumed to be superior [16]. However, previous research has demonstrated that this distinction’s meaning is highly context-specific; in some cases, local wage labor represented a more economically attractive opportunity than “independent” employment [43], and “dependency” indicators thus became uninformative or misleading when used as key outcomes. This finding suggests that “independence” of livelihood activities should not be adopted uncritically or assumed to be a measure of superior employment outcomes. The reviewed studies also raised questions about whether participants’ enterprise success was due to program elements such as coaching or to external factors [16]. Although the question of attributing improvements to the intervention was only addressed rigorously in the first- and second-cohort studies in Uganda, previous work has highlighted the critical importance of addressing counterfactuals. At least one earlier Graduation project documented substantial contextual effects (unrelated to the program) on labor market opportunities [43]. In future evaluations, rigorously addressing counterfactuals must become part of standard practice for assessing Graduation programs’ effectiveness at improving participants’ employment prospects.

Four of the five studies reported positive metrics on improved living conditions for at least some participants, but most (except the first-cohort study in Uganda) did not account for contextual factors that could have affected these outcomes [8]. One important insight was that formal documentation is often critical for improving displaced participants’ living conditions; UNHCR’s project in Costa Rica prioritized documentation, recognizing its profound implications for participants’ lives [15]. Other quality-of-life factors that Graduation programs can plausibly impact likely differ across contexts [8,43].

Results for social wellbeing demonstrated further measurement limitations, as this construct is difficult to operationalize. As the wellbeing index used in the first-cohort study in Uganda included a negative score of the K6 screening tool [29,44]—which was intended to measure serious mental illness in the general population [33], rather than positive wellbeing—the validity and interpretability of the resulting index may be limited. Qualitative research from Colombia presented authors’ and coaches’ conclusions about why wellbeing improved, but did not highlight participants’ own perspectives on this point [31]. Participant perspectives will be crucial for future work that seeks to validly assess Graduation programs’ impact on wellbeing. Additionally, quantitative wellbeing metrics such as “group participation” rates in Costa Rica [15] had unclear meaning or practical significance in the absence of further contextual details. As previous research has documented distinct trajectories of participants’ holistic wellbeing—that is, trends of improvement, decline, or stagnation over time—it will be crucial for future studies to interrogate such trends rather than relying on “snapshots” of a single point in time [45]. To accomplish this, detailed subgroup analyses and long-term follow-up studies are needed to clarify how participants may benefit (or decline) through and beyond the duration of the program.

Poverty outcomes were highly heterogeneous, as each study used different indicators. Notably, no programs apart from “Graduating to Resilience” in Uganda reported basic indicators such as household consumption, assets, finances, or income/revenue, all of which have been used previously in seminal evaluations of Graduation programs and elsewhere in the economic literature [10]. Some studies reported dichotomized or categorical outcome measures, which offer limited information about important poverty-related outcomes such as savings and income [15,16]. While the use of dichotomized and categorical indicators often aligns with donor reporting requirements, it seriously limits the inferences that can be drawn from the data. Quantifying impacts more precisely may produce more informative data in future studies. It is also important to rigorously interrogate the contextual validity of the poverty indicators used in each program. Earlier studies of Graduation programs highlighted the importance of doing so, finding differences in the meaningfulness of savings practices [43], savings group participation [46,47], and quantities saved [10,48] across contexts. Finally, tracking participants’ trajectories of change in poverty measures over time could offer deeper insights about their experience in this domain as well. The one reviewed study that did so found mixed results. In Costa Rica, almost one-third of participants’ poverty measures stagnated or even dropped below baseline levels by the end of the program [15]—a finding which should be flagged for serious follow-up efforts to interrogate potential issues with the program’s effectiveness or implementation. To our knowledge, few studies have followed participants longitudinally after the end of the program, or have planned to do so, in displacement settings.

Surprisingly, it appeared that explicitly assessing dependency on humanitarian aid was not prioritized in these studies. None of the reviewed studies explicitly assessed whether dependency was reduced, although this is a central issue that motivates Graduation programming in humanitarian contexts. Indicators of reduced dependency such as self-reliance [49–52] and resilience [53–57] are stated aims of most UNHCR Graduation programs [24,58,59]. It is worth attending to these distinctive outcomes, as this intention to reduce dependency reflects a fundamental departure from traditional Graduation programming—which sought to promote entrepreneurial livelihoods in settings of rural poverty rather than enact a shift away from a baseline of existing subsistence aid. Future projects should explore valid, meaningful metrics to address this issue, engaging in careful qualitative work with participating communities to understand what such a shift would truly entail.

Finally, mental health outcomes were only vaguely reported in most studies, if at all. It may be advisable for future projects to attend to mental health indicators; related programs have found promising results regarding children’s mental health, suggesting that multi-faceted interventions like the Graduation approach could plausibly affect a range of social, emotional, and psychological outcomes at the family level [60]. The exception was the second-cohort Uganda project midline report, which focused specifically on adding a group therapy component to the intervention package [30]. Again, the project’s composite index measure of mental health may have limited contextual validity and interpretability. Future studies should consider measuring mental health outcomes with locally adapted and validated measures designed to assess each distinct mental health construct of interest.

Implementation

Although implementation is a critical focus of many Graduation program reports, our reviewed studies offered limited insights regarding contextually specific concerns, adaptations, or implementation criteria pertinent to these interventions. These issues were discussed most extensively in the qualitative work from Colombia, which emphasized the complexity of both barriers and facilitators of program implementation; for example, facilitators of program success consisted of distinct factors with similar themes across contexts [16,31]. They also found that family-based social capital was a particularly important facilitator of recovery from shocks [16]. This insight draws into question the fundamentally individualistic approach of many Graduation programs, and lends further credence to the finding that additional external support may facilitate greater improvements [16]. These findings suggest that more comprehensive, community-level support could be useful if not necessary for Graduation programs’ success.

Our findings suggested that program feasibility is best supported by flexible implementation in which programs adapt to participants’ needs, provide additional supports as needed, and pivot strategies in response to shocks [16]. The program does not seem practical to implement among communities who move during the course of the program activities, as noted in UNHCR’s Costa Rica program [15], in “Graduating to Resilience” in Uganda [29], and in World Vision’s recent program experience in Zambia. While the time needed to feasibly implement the full program varies by context, it seems unlikely to succeed in less than 13–24 months [15]. Funders should seriously consider the implications of such a timeframe, particularly in the context of short humanitarian grant cycles. The program’s multiple components each require substantial time to prepare and implement in these challenging contexts.

Finally, studies suggested that program sustainability is sensitive to factors such as land tenure, which is particularly important for many participants’ livelihoods [16]. However, land availability is limited in many displacement settings and is likely to present a challenge for the program’s long-term success. Importantly, the Colombia research team also identified the importance of market inter-reliance for sustaining positive livelihood outcomes; in contrast, many current Graduation programs present a high risk of creating markets that are both redundant and highly competitive [16]. It is crucial to design livelihood strategies that have sustained communal impact, focusing on complementary value chains to promote long-term generative economic activity. Finally, studies suggested that government actors are uniquely well positioned to sustain Graduation-type programs [16]. Indeed, Graduation interventions have been integrated into government social safety nets in multiple contexts [14,55,61,62]. This option could powerfully support Graduation programs’ long-term sustainment—although it is unclear whether displaced non-citizen participants could easily benefit from a government-administered version of the program. Once again, these findings underscore the urgent need for more rigorous evaluation of sustainment strategies and how they may differ between displacement scenarios characterized by distinct legal contexts and degrees of protraction.

Gaps and opportunities for future investigation

Our review highlights a notable gap in research regarding sustainment and longitudinal evaluations of Graduation interventions in displacement contexts. Sustaining Graduation interventions over time requires deep understanding of the interplay between interventions, socio-cultural and community contexts, and organizational factors [63]. This understanding begins early in the intervention’s lifecycle—even during the planning phase—and develops with ongoing monitoring and collaborative adaptation to address evolving contextual needs [64]. Addressing this gap requires a committed effort to conducting longitudinal research and developing frameworks that effectively capture long-term outcomes to assess these interventions’ sustainability. Doing so can ensure their continued impact and relevance for addressing the many challenges that displaced populations face.

We also found that many studies included few concrete details about the contextual realities of programming. As such information may be particularly valuable to implementers, it is important for future work to prioritize rigorous analysis of both outcomes and processes to account for the heterogeneity of implementation contexts and intervention components. To analyze these complex contextual issues, it may be helpful to use structured frameworks to systematically identify implementation barriers, facilitators, and required adaptations—all of which can help implementers understand and address challenges [65]. Numerous such frameworks are available and commonly applied within the field of implementation science, including the Consolidated Framework for Implementation Research (CFIR) [66,67], while additional tools such as the Framework for Reporting Adaptations and Modifications to Evidence-based Implementation Strategies (FRAME-IS) [68] or Expert Recommendations for Implementing Change (ERIC) [69] can facilitate documentation and assessment of implementation strategy adaptations. Comprehensively documenting these points may not only improve understanding of complex implementation landscapes but also enhance the generalizability and applicability of findings across various contexts—potentially facilitating development of clearer “standard operating procedures” that implementers value. In this way, Graduation programs offer unique opportunities to test and refine existing theories of implementation science in highly dynamic and challenging environments.

Limitations

Most reviewed studies had shortcomings in terms of research rigor and provided insufficient detail to fully assess the quality of their work. However, it is important to recognize that none of these documents was intended for peer-reviewed publication. Rather, they provided a full cost-effectiveness report [29], a midline implementation report [30], interim findings clearly intended for more comprehensive write-ups at a later time [32] or general overviews for implementation stakeholders [15,16,31]. The studies in Uganda [29,30] and Mozambique [32] clearly had high levels of research support, but described their study procedures as suited their document formats (a long report, a brief report, and a blog, respectively). The qualitative studies in Colombia [16,31] provided more transparent descriptions of their study processes, but had shortcomings related to the mismatch of their analytic approach with the nature of their data. All studies were also limited to some degree in the contextual detail they provided, which is a crucial element for properly interpreting the reported findings and understanding their implications for implementation. As such, they do not fully address the importance of heterogeneity across displacement contexts or identify how site-specific factors can profoundly influence program processes and outcomes.

We could not directly assess any of the studies’ risk of reporting bias (e.g., due to missing results), as the level of available detail precluded such an assessment. Furthermore, we could not quantitatively synthesize our findings due to the absolute inconsistency of outcome indicators; no two studies reported any outcomes using comparable indicators. Only the studies in Uganda [29,30] assessed statistical or measurement uncertainty. Only the first-cohort Uganda study [29] presented subgroup-specific results for refugee participants (if relevant, i.e., if the participant population included non-displaced host country citizens). We have only reported this study’s combined-sample results to align with other studies that did not distinguish between these groups.

Finally, the low number of studies included in the review presents a very limited picture of the status of Graduation programming in global displacement contexts. The robustness of these findings could be enhanced in the future by exploring additional sources of evidence—for example, by directly consulting the study authors or implementers to elucidate reasons for data limitations.