Abstract
This article introduces the Unintended Consequences Knowability Classification Schema, a typological tool to support evaluators in classifying unintended consequences based on their knowability. Drawing on gaps in the literature, the Unintended Consequences Knowability Classification Schema addresses the need for a field-ready framework that distinguishes among anticipated, foreseeable and unforeseeable unintended consequences. It offers conceptual clarity and practical utility across the programme cycle, from design to evaluation, helping practitioners surface blind spots, guide proportionate responses and support adaptive management. The schema complements complexity-aware approaches by integrating contextual insight and actor-dependent foresight without presuming linear causality. Application examples, including in fragile and conflict-affected situations, demonstrate how the Unintended Consequences Knowability Classification Schema supports reflexivity, ethical accountability and inclusive learning. While inherently subjective judgements remain, the schema strengthens transparency and evaluative reasoning around what could or should have been known, and by whom. Further field testing is encouraged to refine its use in diverse contexts.
Keywords
Introduction
There are known knowns: there are things we know we know. We also know there are known unknowns: that is to say, we know there are some things [we know] we do not know. But there are also unknown unknowns – the ones we don’t know we don’t know. And if one looks throughout the history of our country and other free countries, it is the latter category that tends to be the difficult one. (Rumsfeld, 2011: XIII)
Despite decades of scholarship and investment, evaluating the success of international development (ID) projects remains fraught with complexity (Bourguignon and Sundberg, 2007; Dreher et al., 2024). Central to this challenge is the problem of unintended consequences (UCs): outcomes that fall outside the original intervention logic but nonetheless shape project effectiveness. While project evaluations often focus on intended outcomes, UCs can undermine, distort or even reverse them (Jabeen, 2016; Koch and Schulpen, 2018). However, they are rarely consistently classified, limiting learning and accountability (Davidson et al., 2022; Jabeen, 2018). This article begins with a persistent challenge in development evaluation. Although complexity is now widely acknowledged, practitioners and policymakers still lack reliable methods for assessing how project effectiveness is influenced by UCs, especially those that were foreseeable but went unrecognised.
While considerable attention has been paid to understanding what makes aid effective (e.g. Clements, 2020; OECD, 2005), much less attention has been paid to the role of UCs in shaping project success or failure. The difficulty lies not in recognising that UCs exist; they are well documented across disciplines (Baert, 1991; Koch and Schulpen, 2018; Merton, 1936) but in understanding whether and to what extent they could have been known in advance (Morell, 2005, 2010). For practitioners and evaluators, knowability is critical. If certain UCs were foreseeable but unanticipated, poor foresight or weak contextual analysis may be to blame. If others were genuinely unforeseeable, then different forms of risk management and evaluation are warranted. However, the lack of a common language or framework to distinguish between types of UCs, let alone classify their knowability, hampers practical efforts to improve evaluation practice.
Several scholars have contributed to this field. Morell (2005, 2010, 2018) advanced the field by conceptualising UCs along a foreseeability continuum and advocating adaptive, systems-oriented evaluation. Burlyuk (2017) developed a typology to support UC classification. Jabeen (2018) proposed an unintended outcomes evaluation approach, and Koch et al. (2021) outlined operational categories of UCs in development cooperation. These works enriched the conversation but also expose gaps. While some frameworks engage with the notion of foreseeability, few place knowability at the centre of their classification logic. And none, to our knowledge, offer a parsimonious, field-ready schema grounded in clear terms that help evaluators distinguish UCs by their knowability. This remains a missing link between evaluation theory and development practice.
Building on these traditions, this article introduces the Unintended Consequences Knowability Classification Schema (UCKCS): a typologically precise, field-ready framework that enables evaluators to classify UCs by their knowability at the time of programme design. Rather than displacing existing models, UCKCS complements them by standardising terms and offering practical logic for retrospective and prospective evaluation.
Foreseeability is often conflated with terms such as anticipation, prediction and knowledge, yet these distinctions matter for evaluative reasoning. Merton (1936) linked unanticipated consequences to ignorance, error and short-term interests. Morell’s (2005, 2010) foreseeability continuum builds on these foundational ideas through systems thinking and iterative learning. While foreseeability reflects what evaluators could have known, knowability refers to what was knowable given available evidence and context. UCKCS adopts knowability as its organising logic, offering a more transferable and practitioner-ready typology.
In doing so, the article makes three contributions. First, it clarifies persistent terminological ambiguities in the literature and explains how they obstruct evaluation. Second, it synthesises prior frameworks through a structured gap analysis to demonstrate why a knowability-specific schema is necessary. Third, it presents and explains UCKCS, showing how it can differentiate among types of UCs, assess their implications and improve the quality of evaluation findings. While developed for international development evaluation, the schema is designed as a flexible tool that can complement established evaluation frameworks and remain relevant in other sectors facing complexity and uncertainty.
The remainder of the article is organised as follows. The next section reviews relevant literature across three domains: the definitional ambiguity surrounding UCs in development, existing frameworks for classifying or anticipating UCs, and the conceptual evolution of knowability in social action and evaluation theory. Special attention is given to the convergence of insights from sociology, behavioural economics and project management that inform the understanding of UC predictability. This is followed by a structured gap analysis of nine prominent models and frameworks developed to support UC identification in development contexts, highlighting their respective contributions and limitations with respect to knowability.
Building on this foundation, the article introduces UCKCS, explaining its logic, structure and terminology. A discussion section situates its theoretical contribution, practical implications and future applications. The article concludes by arguing that classifying UCs by their knowability is not simply an academic exercise but a necessary step towards more adaptive, accountable and context-aware development practice.
Literature review
Defining knowability in the context of unintended consequences
The knowability of UCs refers to how the outcomes of human actions, whether social, economic or political, can be anticipated (Merton, 1936; Weber and Runciman, 1978). The knowability spectrum ranges from fully predictable outcomes to entirely unforeseeable due to complexity, uncertainty or incomplete information (Merton, 1936; Morell, 2010). Understanding where a consequence falls on this spectrum can inform better decision-making and policy development.
Sociology-economics intersection and the knowability of unintended consequences
Ankersmit (2021) argued that UCs are foundational to historical processes, exemplified by events such as American independence following Britain’s taxation of tea (Moore, 2023). Early thinkers like Giambattista Vico and Max Weber laid the groundwork for understanding the causality behind human actions, highlighting the challenges in predicting social phenomena (Gane, 2006; Weber and Runciman, 1978). Vico’s distinction between natural sciences and the study of human society initiated a critical conversation on whether UCs can ever be fully knowable, given the complexity of social systems.
The evolution of economic thought and unintended consequences predictability
Adam Smith’s concept of the ‘invisible hand’ acknowledged that self-motivated actions could yield beneficial public outcomes, even if not deliberately designed (Ferguson and Oz-Salzberger, 1995; Nozick, 1994; Smith, 2010; Smith and Seligman, 1910; Ullmann-Margalit, 1997). However, critics (Mill, 1965; Nescolarde-Selva et al., 2019) have noted that Smith’s theory oversimplifies human motivations, suggesting that not all UCs are beneficial or predictable. Hirschman’s ‘hiding hand’ further complicated the notion of knowability by proposing that individuals’ ignorance of future challenges may inadvertently foster creativity in overcoming unforeseen obstacles (Hirschman, 1967).
The optimism underpinning Hirschman’s ‘Hiding Hand’ principle that creativity and serendipity would offset initial underestimation of risk has come under increasing scrutiny. Flyvbjerg and Sunstein (2016) introduced the ‘Malevolent Hiding Hand’ as a countervailing pattern in which both costs and benefits are systematically misestimated, resulting in compounded failure. This critique was later expanded through a large-N empirical rebuttal by Flyvbjerg (2018), who concluded that the benevolent version was the exception rather than the rule. However, Room (2018) counters that Flyvbjerg’s analysis neglects the developmental logic of the Hiding Hand, which centres on emergent learning and strategic experimentation rather than predictive certainty. This article builds on that middle ground, acknowledging the Hiding Hand’s limitations while recognising its conceptual value in high-uncertainty settings.
Behavioural economics and rationality limits
Behavioural economists such as Thaler and Sunstein (2008) highlight how cognitive biases disrupt rational decision-making, contributing to some UCs’ unknowability. Specific biases, such as optimism and confirmation bias, systematically undermine attempts to foresee UCs by distorting risk perceptions and overestimating positive outcomes. The interaction between sociology and economics, as confirmed by Aydinonat (2008) and Portes (2010), suggests that UCs emerge from both rational and irrational behaviours, influenced by social, psychological and cultural factors beyond traditional economic models.
Frameworks for enhancing the knowability of unintended consequences
Weber’s social action theory (Weber and Tribe, 2019) provides a framework for understanding UCs by interpreting individual intentions within social structures. Merton (1936, 1968) classified the mechanisms through which UCs arise, such as a lack of adequate knowledge or immediate interests, offering pathways to anticipate them better. However, de Zwart (2015) clarifies that Merton’s framework primarily applies to deliberate actions, leaving spontaneous behaviours largely unpredictable. Probabilistic forecasting and scenario planning are emerging as practical tools to bridge this gap, allowing for partial foresight in inherently uncertain environments.
Morell’s (2005, 2010) continuum model is a key contribution in this space, which recognises UC foreseeability gradations and urges evaluators to respond adaptively to emergent outcomes. His work laid the necessary conceptual groundwork by highlighting that not all UCs are equally predictable and that evaluative strategies should reflect this variation. However, the primary emphasis of his framework remains on evaluative responsiveness and how to adapt once UCs are identified rather than on typological precision.
While the literature offers valuable insight into how UCs arise and might be interpreted, the specific concept of knowability remains under-theorised and inconsistently applied in evaluation models. Some frameworks nod to foreseeability, but few define it precisely, let alone offer practical tools for classification. This conceptual and operational gap continues to limit the field’s ability to assess whether and how UCs could have been anticipated.
Challenges in predicting unintended consequences in development projects
Development projects, especially in fragile and conflict-affected situations (FCAS), are particularly susceptible to UCs due to systemic complexity and uncertainty (Davidson et al., 2022). These risks can be partially mitigated through improved design, coordination and monitoring (Ika et al., 2012). However, their effectiveness remains limited by the unknowability inherent in sociopolitical systems. An example is where aid agencies hire and upskill local staff, which can, in turn, unintentionally weaken government institutions by exacerbating the ‘brain drain’ (Bizhan, 2017; Honig, 2018).
The intersection between project success or failure and the knowability of UCs is critical. Studies by Ika et al. (2010), Khang and Moe (2008) and the World Bank Group (2017) show that success often hinges on factors such as the provision of adequate resources, stakeholder engagement and support, monitoring and evaluation, and project coordination. However, UCs can undermine these factors, so they must be specifically considered (OECD, 2021).
The OECD (2021) further highlights how UCs can unevenly affect different groups, leading to project failures even when initial objectives are met. For example, the unintended support of antigovernment elements in FCAS (Davidson et al., 2022) demonstrates how UCs can contribute to project failure, especially when the sociopolitical context is not adequately understood.
Despite a lack of clarity, Ika (2012) highlights that development projects often fail because development project managers are caught by one or another of four development project traps: (1) the one-size-fits-all technical trap, (2) the accountability for-results trap, (3) the lack-of-project-management capacity trap and (4) the cultural trap. Ika suggests that project sponsors and implementers can avoid these pitfalls by adopting agile management, focusing on long-term objectives, providing greater aid agency supervision and ensuring awareness of cultural norms are embedded in implementation modalities. Principal-agency top-down management has also been associated with development project failure. Following a review of 14,000 development projects, Honig (2018) found that too much and too little control from aid agencies often led to poorer outcomes. He proposed that better results occurred when implementers were empowered to make real-time decisions within a delegated authority framework. These findings illustrate the empirical complexity of control dynamics in development practice.
Complexity, risk and uncertainty
Development projects have long been recognised as complex (Hirschman, 1967; Ika et al., 2012; Morell, 2005). UCs are more likely to occur with increasing complexity and heightened uncertainty (Bamberger et al., 2016; Davidson et al., 2022; Koch and Schulpen, 2018; Morell, 2010; Oliver et al., 2020). However, traditional management of development projects has often relied on top-down, linear-based approaches (Michael, 2004; Morell, 2005; Rondinelli, 1982), designed to mitigate quantifiable risks (Al Hasani, 2018; Rihani and Geyer, 2001), including the use of contingencies for project milestones and budget variances (De Meyer et al., 2002). These approaches are poorly suited to managing uncertainty linked to unforeseeable events that cannot be predetermined or quantified (Al Hasani, 2018; Al Hasani and Regan, 2017).
Notably, Samset and Haavaldsen (1999), in their review of 249 development projects, found that 63 per cent of uncertainty-related problems stemmed from poor planning, flawed design or limited environmental awareness. They concluded that, with hindsight, 70 per cent of these issues were predictable. Given this finding, managing and evaluating development projects with complexity-sensitive tools and approaches is likely to improve outcomes (Al Hasani, 2018; Al Hasani and Regan, 2017; Larson, 2018; Michael, 2004; Mowles et al., 2008; Oliver et al., 2020; Ramalingam, 2013; Rihani and Geyer, 2001; Ward and Chapman, 2003).
Hindsight and foresight evaluation
Within the discourse on development project success and failure determination, the OECD’s Development Assistance Committee (DAC) evaluation criteria are the most well-known and extensively used criteria for evaluating aid (Chianca, 2008; Patton, 2021; Picciotto, 2020) and underpin other global evaluation standards and principles (Mountfield, 2015; World Bank Group, 2019). While there are notable distinctions and focal points, many common elements can be found in donor and recipient country monitoring and evaluation standards (Boehmer and Zaytsev, 2019).
However, despite almost ubiquitous use, Clements (2020) argues that the DAC criteria cannot provide an adequate foundation for learning and accountability in ID. His argument includes three elements: (1) the DAC criteria are vulnerable to positive bias, (2) they do not contain a consistent measure of results such as an economic rate of return and (3) several definitions are ‘deeply problematic’ (Clements, 2020: 3). Moreover, Patton (2021) and Picciotto (2020) add that the DAC criteria are not equipped to evaluate complex development projects as they are based on a traditional evaluation framework built on linear causation methodologies. As such, the relationships between cause and effect are only evident retrospectively (Hummelbrunner, 2011) and unable to appropriately deal with unpredictability associated with complexity (Patton, 2021).
Complex projects, such as those implemented by ID actors, require the use of prospective systems-based evaluation methodologies that capture social and environmental externalities, interdependencies and variable relationships and unpredictability (Hummelbrunner, 2011; Morell, 2005, 2010, 2018; Patton, 2021; Picciotto, 2020) that often result in UCs (Davidson et al., 2022).
Unintended consequence terminology
Over time, terms such as ‘unanticipated’, ‘unexpected’, ‘unforeseen’ and ‘unplanned’ have all become synonymous with ‘unintended’ and ‘consequences’ is often replaced by ‘effects’, ‘outcomes’ and ‘impact’ in ID evaluation reporting (Davidson et al., 2022). Regardless of the language used, over the decades that followed since Merton’s (1936) essay, several clarifications and/or refinements to the definition of UCs have been added to the body of literature; these include:
‘unintended outcomes are the side effects of governmental policies and programs, not the outcomes these actions are supposed to produce. Unintended outcomes are anticipated when accurately predicted in advance and are unanticipated when they occur unexpectedly. Whether they are anticipated or not, they are either positive, neutral, or negative in value’ (Sherrill, 1984: 28);
‘unintended consequences are a particular effect of purposive action which is different from what was wanted at the moment of carrying out the act’ (Baert, 1991: 201);
‘end-states that are qualitatively distinct, and sometimes the opposite of, those originally intended’ (Portes, 2000: 10);
de Zwart (2015), who explored the conflation of unanticipated and unintended, argued that the conflation of the two terms had effectively hidden a subcategory of outcomes, those being unintended but anticipated. He termed these as ‘permitted outcomes’ (de Zwart, 2015: 295);
‘Unintended outcomes refer to the effects of an intervention other than those it aimed to achieve. Such effects could be positive – producing additional benefits, negative – causing harm to those involved directly or indirectly, or neutral’ (Jabeen, 2016: 144);
‘Unintended consequences/effects are particular outcomes of purposive social and/or political action which are different from the outcomes that were expected at the moment this action was undertaken’ (Hoebink, 2017: 3);
Specific to development projects, Lemon and Pinet (2018: 254) defined UCs as ‘any effect outside of the logical framework or going against the direction of the original theory of change’;
Burlyuk (2017: 1012) argued that the ‘concept-and-phrase “unintended consequences” is self-explanatory and refers to outcomes of purposive action(s) which are not directly intended by an actor’;
In citing Baert as the source, Koch (2024: 15) defined an unintended effect as ‘the consequence of an action that differs from the consequence that was aimed for when starting it’.
These conceptual, methodological and linguistic ambiguities underscore the need for a classification schema focused explicitly on the knowability of UCs, an area that remains underdeveloped in current evaluation frameworks.
Conceptual framing of the UCKCS
UCKCS is grounded in long-standing theoretical efforts to understand the nature of purposive action and its often unpredictable outcomes. Drawing on Weber’s theory of social action (Weber and Tribe, 2019), Merton’s analysis of the sources of UCs (1936) and Hirschman’s (1967) framing of development uncertainty, the schema also directly engages with Morell’s continuum of foreseeability (Morell, 2005, 2010). However, whereas Morell’s contribution focuses on responsive strategies once UCs emerge, UCKCS introduces a structured typology that enables evaluators to classify UCs by their knowability during programme design, implementation and completion.
The schema operates across three levels. At its broadest, an outcome may be identified simply as a UC, without further disaggregation. Where additional information is available, UCs are classified as anticipated or unanticipated. Anticipated outcomes are further distinguished as expected or unexpected, while unanticipated outcomes are categorised as foreseeable or unforeseeable. This hierarchical and definitional structure underpins the methodological approach that follows. In addition to Morell’s emphasis on adaptive response to emergent UCs, UCKCS supports ex-ante planning and anticipatory diagnostics, bringing greater analytical sharpness to evaluative practice in complex development environments.
Method
Building on the conceptual limitations identified in the literature, this section uses gap analysis to systematically assess how existing frameworks address the knowability of UCs. The scholarship has offered several methods and models to assist evaluators with identifying UCs, each with specific strengths and weaknesses. However, despite incremental improvement, current models and frameworks still do not entirely codify UCs in terms of their knowability. To address this weakness, gap analysis, as described by Kim and Ji (2018), guided our assessment of the current state of each evaluation model and framework regarding knowability classification. Gap analysis assesses the current state and contrasts that assessment with the required state, identifying the gaps in the process (Kim and Ji, 2018). Merton (1968) was the first to refer to gap analysis when theoretically assessing the disconnect between an individual’s culturally induced goals and institutional norms. The technique has been applied widely, including biodiversity and conservation (Jennings, 2000; Scott and Jennings, 1998), business intelligence and management (Einstein, 2024; Kim and Ji, 2018; Molensky et al., 2010), defence system engineering (Langford et al., 2007), evaluator competency assessment (Galport and Azzam, 2017), humanitarian and foreign aid (Reaves et al., 2008; Schneider, 2020) and project management (Økland, 2015; Winch et al., 1998).
Our method follows the stepwise approach for gap analysis widely applied in previous studies (Chaves, 2021; Einstein, 2024; Galport and Azzam, 2017; Jennings, 2000; Kim and Ji, 2018; Langford et al., 2007; Molensky et al., 2010; Økland, 2015; Reaves et al., 2008; Schneider, 2020; Scott and Jennings, 1998; Winch et al., 1998). This adapted approach entails the following steps:
Reviewing the current state of existing models and frameworks regarding the knowability of identifying unintended consequences of ID interventions;
Defining the future state needed for these models and frameworks concerning unintended consequences knowability;
Summarising the strengths and weaknesses of each model and framework and identifying gaps between the current and future state of each regarding unintended consequences knowability;
Outlining a proposed knowability classification schema that bridges the gap for each;
Providing a core set of terms consistent with the knowability classification schema so unintended consequences can be codified.
A search of peer-reviewed ID papers from Scopus and Google Scholar identified nine models and frameworks developed over the past 20 years specifically to assist with identifying and classifying UCs. A gap analysis of each, in chronological order, was completed.
Models, such as the one proposed by Sherrill (1984), were excluded because identifying UCs focused on government programmes and not specific to ID. We also omitted the paper from Oliver et al. (2020) as they highlighted the need for adaptive, inclusive and context-sensitive approaches within ID evaluation methodologies rather than offering a new model.
Results
This section presents the gap analysis findings of the nine models and frameworks developed over the past two decades to assist with identifying and classifying UCs from ID interventions. Table 1 chronologically presents models and frameworks, including description, strengths, weaknesses and gaps in knowability classification. A narrative of the gap analysis results of each model and framework is then provided.
Chronological summary of models and frameworks.
Current state assessment
Morell (2005) recognised the distinction between unforeseen and unforeseeable UCs and advanced the aetiology of each category as part of his original evaluation model. He concluded that improved analytical tools, lesson documentation and systems thinking would likely improve the identification of UCs. Five years later, Morell (2010) offered further refinement of his model. He recognised that UCs appear along a continuum encompassing foreseen, foreseeable and unforeseen. In highlighting high levels of complexity, he explicitly incorporated systems thinking, advocated for agile and adaptive evaluation methods, and offered a framework to help evaluators categorise and anticipate different types of UCs.
Bamberger et al. (2016) advocated using mixed-method evaluation approaches that integrated qualitative research methodologies with quantitative evaluations such as randomised controlled trials (RCTs) and coined the term RCT+. Using case studies, their Mixed Methods Evaluation Framework (MMEF) identified additional UCs beyond those found through conventional methods.
Burlyuk’s Classification of UCs (BCUC) (Burlyuk, 2017) makes a valuable methodological contribution by offering a typology for ID UCs. Her work separated intended from unintended outcomes from development interventions and provided the ability to break down and classify UCs in an orderly manner, including unanticipated, anticipated and expected, anticipated and unexpected, all important subcategories of knowability. The BCUC provides the ability to further classify UCs by value, relationship to the initial intention, type of action, type of outcome, strength of causal links, who was impacted, modes of acknowledgement, temporality and the reason/cause.
Jabeen (2018) developed a structured classification framework, the Unintended Outcomes Evaluation Approach (UOEA), to identify, categorise and analyse UCs arising from social development programmes. UOEA recognises that, regardless of whether intended outcomes were achieved, interventions may also produce UCs. UOEA offers a typology of UCs based on knowability, value, distribution of effects and temporality. Jabeen suggests that UOEA should be part of a three-step process: (1) a clear outline of intended programme outcomes, (2) forecasting foreseeable UCs and (3) using UOEA to map and analyse UCs to enhance understanding.
In Morell’s (2018) later work, he acknowledged that UCs are often unavoidable and that evaluation models must remain flexible and continuously evolve to capture emergent effects, terming this the Systematic Iteration Approach (SIA). His earlier work described UCs and explained their surprise appearance, whereas SIA provided evaluators with a methodological approach that could systematically anticipate and adapt to UCs in real-time.
The use of Ripple Effects Mapping (REM) to identify UCs was explored by Peterson and Skolits (2019). The authors argued that in addition to traditional, linear-based methods, systems-oriented and RCT evaluation methods were also unsuitable for highly complex situations, such as development programmes, and unable to capture all UCs. The authors integrate grounded theory into the REM qualitative evaluation method and report it as a powerful tool for understanding complex programme dynamics.
Koch et al. (2021) developed an operational typology of 11 types of UCs, categorised across micro, meso, macro and multi-level effects. These include (1) price, (2) distortions, (3) behavioural shifts, (4) migration patterns, (5) conflict triggers, (6) governance impacts, (7) nationalist backfire, (8) marginalisation, (9) leakage, (10) catalytic spillovers and (11) poor process effects. Their typology provides a framework for identifying and assessing ID UCs, assisting practitioner decision-making.
In his book published in late 2024, Koch (2024) refines the operational typology provided by Koch et al. (2021) in which UCs can be categorised into (1) backlash effects, (2) conflict effects, (3) migration and resettlement effects, (4) price effects, (5) marginalisation effects, (6) behavioural effects, (7) negative spillover effects, (8) governance effects, (9) environmental effects and (10) ripple effects.
Required future state
We argue that development practitioners require the removal of overlapping terminology and poor categorisation initially identified by Jabeen (2018). Our view is supported by Davidson et al. (2022), who identified 90 different terms for UCs in a systematic literature review of 121 project documents. The lack of agreed definitions has permitted stylistic variation, a well-known literary device to avoid repetition using synonyms (Saraireh, 2001), further confusing the ID evaluation field and its users, many of whom work in English as a second language.
Davidson et al. (2022) also point out that regardless of framework used, UC codification and terminology need improvement to remove ambiguity and enhance understanding. This is particularly the case for UC knowability.
Framework and model strengths and weaknesses
Morell’s (2005, 2010) early work advanced the understanding of UCs. However, including contemporary systems-oriented approaches into traditional rigid evaluation methodologies will likely limit practical application and may be resource-intensive.
Importantly, Bamberger et al. (2016) distinguished between anticipated and unanticipated UCs. While well suited to complex operating environments, the MMEF requires more time, funding and specialist expertise.
Burlyuk (2017) developed a typology for classifying UCs. Despite this progress, the principal weakness of her work is an apparent conflation of the terms ‘unanticipated’ with ‘unforeseeable’ and ‘anticipated with foreseeable’, leaving ID practitioners to manage ambiguous language around these four critical aspects of knowability. The 10 primary and 41 secondary categories also add complexity to UC classification, and given the nature of UCs, there are likely difficulties in establishing robust causal links.
Compared to the BCUC framework, Jabeen’s (2018) UOEA offers methodological improvements and a less complex classification procedure. Weaknesses of UOEA include terminological conflation and ambiguity on knowability relating to ‘anticipated’, ‘foreseeable’ and ‘expected’, and ‘unanticipated’, unforeseeable’ and ‘unexpected’. Despite recognising terminology inconsistency, Jabeen uses ‘unintended outcomes’, ‘unintended effects’ and ‘unintended consequences’ interchangeably.
The strength of Morell’s (2018) SIA model is the emphasis on continuous revision, keeping evaluations adaptable to emerging consequences. However, we posit that its reliance on iterative model refinement may be impractical in rigid funding or organisational contexts that resist methodological flexibility.
The primary strength of the model offered by Peterson and Skolits (2019) lies in its ability to map causal pathways to reflect lived experiences and support deeper stakeholder engagement. However, there are challenges with attribution, sampling bias and generalisability.
The typology offered by Koch et al. (2021) organises UCs into distinct types, enabling practitioners to identify UCs. However, it categorises them after the fact and does not provide predictive tools to anticipate them in ID programme design.
The book by Koch (2024) contributes to understanding ID UCs. It provides case studies for each category of UC and actionable recommendations. However, the typology is limited to categories of effect.
These results show that while existing frameworks have advanced UC understanding, significant gaps remain, particularly in consistently classifying their knowability. These limitations stem from conceptual overlaps, terminological ambiguities and practical implementation challenges across different contexts. The analysis further demonstrates that no existing model fully addresses the nuanced classifications required to accurately assess UCs’ predictability, thus reinforcing the need for a more comprehensive and standardised schema.
Unintended Consequence Knowability Classification Schema
This gap analysis informs the development of UCKCS, designed to resolve ambiguities and inefficiencies identified in prior models. By establishing a logical, evidence-based framework, UCKCS aims to provide a systematic approach for evaluators to better anticipate, classify and manage UCs across diverse development contexts.
It is recognised that there are two primary categories of UCs: anticipated and unanticipated (Baert, 1991; Burlyuk, 2017; Jabeen, 2018; Morell, 2018). Each has two secondary categories. The two secondary categories for an anticipated UC are expected and unexpected. For unanticipated UCs, secondary categorisation consists of foreseeable and unforeseeable. UCs can also be classified into tertiary classifications, such as those offered by Burlyuk (2017), Jabeen (2018), Koch et al. (2021) or Koch (2024).
Building on the frameworks reviewed earlier and summarised in Table 1, the tertiary classification concept within UCKCS bridges the new schema and existing typologies. It signals that evaluators may, where appropriate, nest additional subcategories drawn from established frameworks, such as those of Burlyuk (2017), Jabeen (2018), Koch et al. (2021) and Koch (2024), beneath the secondary levels of UCKCS. This enables finer-grained coding and continuity with prior approaches while preserving the schema’s parsimony and field-readiness.
Figure 1 shows UCKCS, applicable across the full project cycle. Some decision points use past tense, such as ‘was there an unintended outcome?’ for readability and not to imply retrospective use only. The schema itself is structurally neutral and supports prospective (ex ante), real-time (in situ) and retrospective (ex post) application. For instance, in a forward-looking application, evaluators may ask, ‘can any unintended outcomes be anticipated?’ The framing used is aligned with the definitions introduced in Section 5. This flexibility enables designers and evaluators to map potential and confirmed UCs, regardless of emergence, thereby enhancing risk identification, adaptive course correction and reflective learning, particularly in fragile and complex development contexts.
Unintended Consequence Knowability Classification Schema (UCKCS).
While UCKCS is presented as a single decision tree, it is intended for iterative application and does not assume that development interventions produce only one UC. Projects in complex and fragile settings often generate multiple UCs, favourable and adverse, across domains and stakeholder groups (Davidson et al., 2022). The schema is therefore applied independently to each identified or anticipated UC, enabling nuanced classification and cumulative analysis across spatial, temporal and institutional dimensions. UCs may also occur in chains, where an expected outcome triggers further unintended ones, or in branching forms, where the same intervention produces divergent effects across groups (e.g. an expected outcome for urban women but an unforeseeable one for rural women). In these cases, UCKCS can be re-applied recursively, with each link or branch treated as a separate node for classification. This maintains the schema’s operational clarity while acknowledging that consequences often unfold in cascades and distributions rather than along a single, tidy pathway. 1
Definitions
This section defines the seven classifications in UCKCS.
Unintended consequence
When applied to ID, an ‘unintended consequence’ is defined as any intervention effect, outcome or impact outside of the intervention logic (adapted from Lemon and Pinet, 2018). This definition aligns with those provided by other scholars (Burlyuk, 2017; Jabeen, 2018; Koch and Schulpen, 2018; Mica, 2015). Our definition also aligns with the original definition of UCs Baert (1991), provided and adapted by Koch (2024). It also considers the original understanding of Merton (1936) that UCs are not necessarily undesirable.
Davidson et al. (2022) identified 43 stylistic variations of ID UCs in a systematic review of published and grey literature. These include terms such as ‘detrimental consequence’, ‘spillover effect’ and ‘unintended benefit’. Despite the diversity, they are all variants of the primary term, lacking adjectives to refine the UC type.
Anticipated unintended consequence
As its name suggests, an ‘anticipated unintended consequence’ arises when development actors have prior awareness of a UC and consider it possible. De Meyer et al. (2002: 61) termed this ‘foreseen uncertainty’, where UCs are known, but actors are unsure if they will occur. Regardless, as these types of UCs are known to occur, contingencies can be planned and implemented, and risks mitigated (Cagliano et al., 2015; De Meyer et al., 2002). These UCs can then be further subclassified as either expected or unexpected.
Anticipated and expected unintended consequence
We acknowledge that this form of UC should never exist. If the consequence was anticipated and expected, then surely those responsible for implementing a project cannot argue that it was unintended – a viewpoint also held by Suckling et al. (2021). Nevertheless, an ‘anticipated and expected unintended consequence’ is a subcategory of anticipated UCs that a development actor thought of (anticipated) and determined that a UC was also probable. Despite this, negative UCs would be tolerated as long as they were at an acceptable scale relative to the intended outcome. Merton (1936: 895) described this as ‘the lesser of two evils’.
For expected favourable UCs, mitigation strategies would likely not have been implemented, and the result may even have been encouraged, especially where the intervention logic failed to identify these as intended. These situations are explained by Portes (2000, 2010), who points to Marx’s ‘hidden abode’, in which the ‘announced goal is not what it seems – that is, it is not what the actor or those in authority in a collectivity actually intend’ (Portes, 2000: 7). Thus, we concur with Ritzer (2011) and argue that given that the actual outcome can be hidden, it is often difficult for everyone to know the actual intended outcome. In ID practice, this translates to those responsible using the current inconsistency in terminology as a retrospective excuse for permitting outcomes outside of the intervention logic, or as Merton (1936: 897) more formally described, ‘unintended consequences are post facto declared to have been intended’, with development actors seizing the opportunity to rewrite history and claim departures as intended (Merton, 1936; Portes, 2000; Tilly, 1996). Regardless of the positive result, these outcomes were not intended as part of the intervention logic.
Anticipated and unexpected unintended consequence
An ‘anticipated and unexpected unintended consequence’ is a subcategory of anticipated UCs that a development actor thought of (anticipated) but incorrectly predicted. In these circumstances, the probability was incorrectly assessed to be negligible, or mitigation strategies implemented were ineffective.
Merton (1936) identified ‘immediacy of interest’ as another underlying cause of these UCs. Development actors are so determined to accomplish the intended outcomes that they blindly choose to ignore the possibility of UCs and, therefore, are surprised when they arise.
Stylistic variations of an anticipated and unexpected UC identified by Davidson et al. (2022) included ‘unexpected benefit’ and ‘unexpected outcome’, along with several other terms.
Unanticipated unintended consequence
An ‘unanticipated unintended consequence’ is a primary UC category that a development actor did not consider. Morell (2010) points to a continuum; at one end, unanticipated UCs could have been anticipated with more scrutiny, that is, ‘unanticipated but foreseeable unintended consequences’. On the other hand, no matter how thorough the planning was, the complexity level meant that correctly anticipating outcomes was impossible. These are known as ‘unanticipated and unforeseeable unintended consequences’.
Stylistic variations of an unanticipated UC, identified by Davidson et al. (2022), include ‘unanticipated effect’, ‘hidden outcome’ and ‘unforeseen impact’. UCs termed ‘unforeseen’ remain at this higher level unless clarity on whether the UC was foreseeable or unforeseeable is provided.
Unanticipated but foreseeable unintended consequence
As the terminology used to describe this category suggests, an ‘unanticipated but foreseeable unintended consequence’ is a UC that was unanticipated because, as Morell (2005: 446) states, ‘analytical frameworks and experience were not considered when projecting what might happen when a programme is implemented’. These UCs may have been apparent ex-ante if actors had access to more resources (time, funding and staff) or were availed of greater knowledge and understanding of potential UCs. Regarding prior knowledge, Merton (1936) notes that no formula exists for determining the exact amount of knowledge necessary for adequate foreknowledge to avoid UCs. Further to Merton’s point, actors may lack the knowledge needed to anticipate UCs because it was not shared. This situation can arise when a phenomenon known as the MUM (keeping Mum about Undesirable Messages) effect (Rosen and Tesser, 1970) occurs, whereby individuals are reluctant to share negative or undesirable information.
Unanticipated but foreseeable UCs can also result from what Pinto (2014) describes as the normalisation of deviance. Here, UCs occur because incremental, small steps that individually appear rational ultimately result in a negative outcome. Thus, despite being unanticipated, given the graduality of processes and decisions taken, these forms of UCs are often foreseeable.
Finally, another pathway to unanticipated but foreseeable UCs occurs when a project sponsor intentionally conceals intended outcomes from the implementer, making them appear unanticipated to some actors but foreseeable to others. This practice, known as knowledge withholding (Connelly et al., 2012), involves deliberately concealing information. It results in outcomes that appear unanticipated to those excluded, but foreseeable to those withholding knowledge.
This dynamic may also emerge more subtly through epistemic asymmetries between actors. In development practice, donors, implementing partners, local government agencies and community groups frequently hold different types and depths of contextual knowledge. A UC that appears unforeseen to a funding agency may have been entirely foreseeable to local stakeholders had their knowledge been elicited or heeded. Conversely, community actors may be unaware of broader system dynamics influencing intervention outcomes. These differences in perspective can result in differential anticipation of UCs, where what is unanticipated for one actor is foreseeable for another (Chambers, 1998; Eyben, 2013).
As Koch and Rooden (2024) demonstrate in their study of aid localisation, such asymmetries may lead to UCs that reflect not a lack of available knowledge, but the exclusion or marginalisation of that knowledge, often shaped by cognitive distance, normative framing or power differentials. Their work underscores that foreseeability is not fixed but actor-contingent, reinforcing UCKCS’s emphasis on interrogating who (un)anticipated the UC.
Once identified, these UCs can retrospectively move to anticipated UCs, and appropriate risk management strategies can be applied to mitigate them (Cagliano et al., 2015).
Unanticipated and unforeseeable unintended consequence
An ‘unanticipated and unforeseeable unintended consequence’ arises because, even with perfect information, some events cannot be predicted, and humans make unpredictable decisions. Merton (1936: 899) termed these ‘chance consequences’ as their prediction is not possible due to numerous complex circumstances and, therefore, ‘step beyond the boundaries of knowledge’ (Vernon, 1979: 61) and, accordingly, they cannot be managed (Suckling et al., 2021). Within the ID context, Morell (2005: 445) describes these as outcomes that ‘stem from the uncertainties of changing environments combined with competition among programmes occupying the same ecological niche’. Given the inherent difficulties associated with this class of UCs, prospective evaluation methodologies that use complexity-sensitive tools and approaches and systems thinking are required to minimise their occurrence.
Discussion and conclusion
This article introduces UCKCS as a conceptual and practical contribution to development evaluation, addressing the absence of a consistent, field-ready framework for classifying UCs by their knowability. Earlier models advanced the field but remain partial; none provide a typologically precise tool anchored in knowability.
UCKCS makes three theoretical contributions. First, it clarifies conflated terms such as anticipated, foreseeable and expected. Second, it synthesises prior classifications into a nested hierarchy of seven UC types. Third, it elevates knowability to a central organising principle. Beyond theory, the schema offers practical utility for designing, managing and evaluating interventions, particularly in FCAS, where flux and uncertainty heighten the risk of UCs.
Practically, UCKCS can be applied across the programme cycle to help practitioners identify outcomes outside the intervention logic. During design, teams can use the schema to anticipate blind spots. For example, a women’s cooperative project might foresee domestic tension due to shifting household power. At the implementation stage, UCKCS supports agile management: if a value chain programme displaces women from markets, the schema helps determine whether this was foreseen and guides proportionate responses. In ex-post evaluation, the framework sharpens accountability, such as judging whether governance reforms that eroded customary institutions were foreseeable but ignored, or genuinely unforeseeable. Finally, retrospective reclassification is possible, strengthening learning loops and early warning systems.
A critical factor shaping what UCs surface is evaluative boundaries. Choices about scope (economic vs sociopolitical impacts) and inclusion (beneficiaries vs households, neighbours or institutions) determine which UCs are revealed or obscured. In FCAS, boundaries may deliberately or inadvertently exclude groups aligned with opposing sides, masking how UCs are distributed (Davidson et al., 2022). Inclusive approaches such as Most Significant Change can surface emergent outcomes that logframe methods may miss (Dart and Davies, 2003). More broadly, systems thinking emphasises the importance of deferring rigid boundaries and capturing emergent outcomes (Hummelbrunner, 2011). By prompting evaluators to reflect on whose perspectives are included or excluded, UCKCS aligns with principles of boundary critique and complexity-aware evaluation (Ulrich and Reynolds, 2010; Williams and Van’t Hof, 2014).
Development contexts are inherently unpredictable: outcomes emerge through non-linear interactions, contested perspectives and shifting systems (Patton, 2017). While UCKCS does not capture every dimension of complexity theory, it complements complexity-aware approaches. Complexity may manifest through project design, stakeholder dynamics, systemic interdependencies or non-linear causal pathways (Hummelbrunner, 2011), each shaping whether UCs appear foreseeable. Features such as emergent behaviours, diverse perspectives and volatile contexts challenge linear assumptions and call for evaluative tools that tolerate ambiguity and dynamic change (Patton, 2017; Williams and Hummelbrunner, 2010). By focusing on knowability, UCKCS enhances diagnostic insight without imposing linearity and can be used alongside systems mapping, outcome harvesting and other complexity-sensitive methods to support reflexive, adaptive evaluation.
Temporality further complicates UC identification. Many UCs, especially those related to sustainability, exclusion or institutional disruption, emerge only after projects end and are missed by short, time-bound evaluations (Bamberger et al., 2016). While not centred on temporality, UCKCS supports phased and longitudinal application by enabling UCs to be revisited and reclassified as new evidence emerges. Its knowability-based logic aligns with Morell’s (2005, 2010, 2018) call for prospective approaches that use systems thinking and scenario planning to anticipate UCs before implementation. Used prospectively, UCKCS helps surface potentially knowable risks; used retrospectively, it distinguishes between avoidable blind spots and genuine unknowability, supporting proportionate accountability and learning.
In addition to the nine UC-focused frameworks reviewed in the gap analysis, UCKCS can be situated alongside broader evaluation models. For example, Stufflebeam’s (2000) Context, Input, Process and Product (CIPP) framework and the Promoting Action on Research Implementation in Health Services (PARIHS) framework (Kitson et al., 1998) are designed to address programmes operating in complex contexts and have been applied in international development practice (Harvey and Kitson, 2015; Molope and Oduaran, 2020). UCKCS adds diagnostic clarity by focusing specifically on the classification of UCs by their knowability. The CIPP framework has long structured assessments across planning, process and outcomes, while the PARIHS framework highlights the role of context, facilitation and evidence in implementation. UCKCS builds on this tradition by offering a diagnostic logic applicable prospectively, in situ and ex post. Used alongside CIPP, it sharpens analysis of contextual blind spots, input assumptions, process disruptions and spillovers. For PARIHS, it adds precision by asking whether outcomes were knowable given existing evidence and clarifying how institutional or cultural dynamics may have masked foreseeability.
Another consideration is the uneven distribution of information and foresight across actors. The knowability of an outcome depends on who holds relevant knowledge, whether it is shared and how it is interpreted (Chambers, 1998; Connelly et al., 2012; Eyben, 2013; Rosen and Tesser, 1970). A UC that appears unforeseeable to donors may be foreseeable to national agencies or community groups, particularly where ‘shadow’ organisational structures operate alongside donor-imposed systems. By asking ‘could this have been known?’ and ‘to whom?’, UCKCS surfaces the politics of knowledge distribution and encourages more reflexive evaluation that acknowledges positionality and power.
UCKCS also has limitations. Its value depends on users’ willingness for honest reflection and contextual understanding, which institutional incentives or power dynamics may constrain. Judgements about foreseeability and anticipation remain subjective and contested (Bamberger et al., 2016; Merton, 1936; Morell, 2010). While iterative modelling remains a theoretically robust way to anticipate UCs, organisational rigidity and resource limits often preclude its use. UCKCS is therefore proposed not as a rejection of iterative modelling but as a complementary, field-ready tool that preserves conceptual clarity while remaining usable in practice.
These limitations highlight the need for empirical testing. Evaluators and programme teams should trial UCKCS across varied contexts, assess its usability and refine it with field feedback. Future testing should also examine how unequal access to information among donors, implementers and local stakeholders shapes UC classification, since what is unforeseeable to one group may be foreseeable to another. Further research could explore integration into donor evaluation standards, training curricula and adaptive management protocols.
In sum, UCKCS provides a typologically precise schema for classifying UCs by knowability and highlights how power, positionality and knowledge distribution shape what is seen or overlooked. By surfacing these dynamics, it supports a more inclusive and reflexive evaluation practice that challenges assumptions about what could, or should, have been known. In this way, UCKCS fulfils its purpose: addressing the long-standing gap in development evaluation while offering a flexible tool that complements established evaluation frameworks and has potential relevance for other sectors facing complexity and uncertainty.
Footnotes
Acknowledgements
The authors thank Professor Emeritus Michael Young, Centre for Global Food and Resources, School of Economics and Public Policy, Faculty of Arts, Business, Law and Economics, The University of Adelaide, for his expertise and guidance with this paper.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The authors received no financial support for the research, authorship and/or publication of this article.
Compliance with ethical standards
No human subjects were involved in this study, as the framework focuses on theoretical constructs and analysis of existing data, thus eliminating the need for separate ethical review board approval.
Data availability
The conceptual framework presented in this study is a theoretical model and, therefore, does not involve directly accessible data.