Contemporary debates about secondary education indicate that Colombian students are lagging behind in international benchmarks for school performance. At the country level, the results of standardised examinations display marked differences across regions, some of which have consistently faced the problem of school failure for (at least) the last three decades. This phenomenon has been researched using a variety of research paradigms (i.e. positivism, structuralism, poststructuralism), none of which have provided a convincing story as to why these schools continue to fail. This dissertation focuses on the study of the persistence of poor school performance introducing the ontology of critical realism to compensate for the failure of other research paradigms in informing explanatory scholarly work in this area of knowledge. One of the main methodological arguments sustained throughout this thesis is that any causal claim in social sciences requires an inquiry that recognises the autonomous, yet intervening, properties of educational structures and the agency of people (i.e. students, parents, teachers) in the making of education policy. Hence, its use of (mostly) qualitative methods, under the umbrella of a realist rationale, to analyse data collected in Northern Colombia to study complex structure and agency interactions to uncover causal mechanisms behind the persistence of educational problems. Margaret Archer’s morphogenetic approach plays a primary role in this thesis in helping to study fallible hypotheses about the historical and contextual drivers that have contributed to limiting the success of educational policy in some regions of the country. According to exponents of middle range realism, given the endless possibilities of conceptualisations of social events, the best-known call for abstraction in social research is setting up provisional theoretical constructs (or hunches) to guide researchers in the theorisation of causal mechanisms and their operation in different contexts. The purpose of empirical research is, hence, to collect empirical data to validate and/or refine those preliminary theoretical prepositions. In the absence of such theoretical (preliminary) constructs, researchers can contribute to knowledge by making progress in that early process of hypothesis building. In this spirit, this study uses Archer’s approach as a Domain-Specific Meta-Theory (DSMT) suitable for bringing together empirical data to build a middle-range theory (or that first theoretical construct) about the persistence of school failure in Northern Colombia. The middle-range theory built in this thesis points to the absence of a real national education project in Colombia because of the lack of empowerment of some subnational actors as stakeholders in the process. For example, evidence provided of the limited level of decentralisation in decision making in primary and secondary education policy in the case of the Department of Atlántico, helps to support the idea that local authorities are not prepared to design and execute educational initiatives that are both responsive to national standards and the needs of people in the region. One consequence of this is that students from the research site fail in exams partly because their socio-cultural environment obscures their perception of the benefits of studying. This explanation is different from the deterministic accounts given by mainstream traditions in school effectiveness research –such as, for instance, the economics of education - which emphasises the skills of teachers as the main driver of school failure. The thesis also exalts the active role of specific groups (i.e. teachers’ unions) in shaping contemporary education. Hence, the failure of some agents to transform educational structures to better suit their preferences and expectations does not reflect a lack of agential efforts to pursue change; rather it reflects the power of other social groups to protect their vested interests in reproducing the status quo. Ontologically speaking, such a causal narrative distances itself from frameworks that consider educational agents as passive recipients of education policy reforms.