Placing PISA in perspective: the OECD’s multi-centric view on education

ABSTRACT The OECD has become a major driver of domestic education reforms, especially since the establishment of PISA. However, we know very little about the contextualisation of PISA within the publication output of the OECD, and what ideas of education the IO is spreading. In this article, we explore the entire thematic portfolio of the OECD’s education publications and show that it goes far beyond PISA. Utilising a mixed-methods approach, we examine whether the OECD’s view on education is shaped by considerations of human capital formation on the one side or by issues of equality, equity, and social cohesion on the other.


Introduction 1
The Organization for Economic Co-operation and Development's (OECD) view on education is considered to be one that emphasizes investment in human capital as a means to generate well-educated citizens who will boost the economy in a globally competitive market. This perception of education as an essentially economic tool can be reasonably deduced from the organisation's original economic purpose and has since been shown by various research (e.g., Hanushek and Luque 2003;Meyer and Benavot 2013). This economic orientation has also been criticised, particularly with regard to the influential and worldwide renowned Programme for International Students Assessment (PISA) ('Open Letter to Andreas Schleicher, OECD, Paris' 2014). With the position as an internationally approved expert organisation in education, the OECD has considerably shaped the debate around the goal of education and continues to provide extensive advice on how educational systems and education policymaking should be organised today. As a result, it departed from its envisioned role as a member-driven communication forum and successively furthered its autonomy as an active player in the field of education (Bloem 2016). Overall, International Organizations (IOs) such as the OECD have gained a considerable voice in public policymaking processes as the providers of education system evaluations, country reports, and policy recommendations.
The main policy tools of the OECD are characterised by soft governance: knowledge production, interpretation and dissemination, and discursive persuasion (Martens and Jakobi 2010). There are no funds to cut or redistribute, as the IO works solely off of its analytic and persuasive power without having the 'hard power' to support its agenda setting, as other IOs can, by redistibuting for example development aid funds (Armingeon and Beyeler 2004;Woodward 2009). Most importantly, the OECD holds the expertise to produce, analyse, and interpret large amounts of complex data on educational effectiveness. By far the most influential instrument for the OECD is therefore its data generating tool PISA. Since its inauguration about 20 years ago, PISA has compared the education outcomes of around 80 different states against defined proficiency standards. Numerous studies confirm the strong impact of PISA on national reform processes (among others : Engel 2015;Niemann 2016;Takayama 2008). This goes so far that PISA has almost become a synonym for the OECD: When discussing the OECD's influence in education, most scholars are discussing PISA.
While the development of the OECD's education policy has been described in detail (Centeno 2018;Henry et al. 2001;Papadopoulos 1994), some studies investigate aspects of the OECD's history and overall agenda to understand the breadth of its impact (Bürgi and Tröhler 2018). Most studies are concerned with the OECD's activities and policy impact of PISA (Breakspear 2014;Domínguez, Vieira, and Vidal 2012;Hopfenbeck et al. 2018). Others focus on the transferability of OECD standards and benchmarks to different contexts (Bulle 2011;Takala et al. 2018), and the OECD's function as a knowledge broker for evidence-based policymaking (Niemann and Martens 2018). Accordingly, the continuous process of developing ideas on education within the OECD is also strongly linked to PISA. Only a few studies try to dissect this process rather than its impact (e.g., Bürgi and Tröhler 2018;Ydesen and Grek 2020).
In fact, and despite a vast amount of research on the OECD as an actor in education, the underlying model of education within the whole OECD and how it evolved is comparatively underresearched. Surprisingly, the existing literature on the OECD treats the IO as a monolithic actor with a coherent position and overall strategy in education. Furthermore, when reviewing the existing literature on the OECD's education leitmotifs, the characterisation of the IO's approach toward education almost always refers to some core issues. It is often (implicitly) assumed that the OECD holds one belief system regarding education, namely an economic-centred view that solely emphasises the utility of education for the economic growth of national economies. In this article, we show that the OECD promotes a far broader spectrum of education views than the prominence of the PISA study and related publications suggest. We emphasize that the OECD's discursive output goes beyond the aforementioned economic-centred topics. However, economic ideas do dominate the discursive policy output of the OECD and how its policy preferences are externally perceived. Hence, we are interested in explaining two aspects of the OECD's education policy: why the OECD's take on education is equated with PISA, and why the IO holds a variety of views on education. By analyzing the OECD's publications on education over time, we contextualise how PISA fits into the IO's discursive production lines. We argue that the OECD is more discordant in its perception of education goals and purposes than what PISA and the related dictum of generating human capital through education may suggest. In fact, a multifaceted picture of the OECD's education topics shows that within the IO, different views on and positions toward education exist.
For explaining our main findings, our theoretical argument is twofold. First, we adapt Kingdon's multiple streams approach (Kingdon 1995), which allows us to assess why PISA has become the synonym for the OECD's work in education. Through the production of data on the effectiveness of education systems with PISA, the OECD presents the problem (varying student proficiency levels), sets the agenda by defining these problems within their ideological framework (defines the subjects worth assessing), and presents alternatives through underlining best practice models of policies. In a second step, we then conceptualise the OECD as a complex bureaucracy, as this will allow us to contextualise PISA within IO's organisational structure. We are then able to explain why the OECD is not a monolithic actor and why it has different approaches regarding education. In the empirical part, we first trace the OECD's success with PISA through a narrative and historical development perspective. We then highlight how the development of indicators and education measurements became increasingly institutionalised within the IO before we show how PISA became the 'favourite child' of the OECDand the prime reference point for external actors. In a second step, we contextualise PISA within the OECD's overall discursive framing of education using a topic modelling approach. This enables us to show that the OECD has emphasised multiple education topics over time and does not solely view education as something to enhance economic growth, but thus far, the focus of academia and policy-makers remained with PISA. Furthermore, when reviewing the whole thematic portfolio of the OECD in education, we find that the economiccentred focus associated with PISA is only one among many others, and topics like the development of higher education institutions were more prevalent over time.

Multiple streams and pathologiesunpacking the OECD theoretically
The following section provides the theoretical background for our interpretations, by first discussing why PISA is perceived as more influential by earlier as well as current education initiatives of the OECD. Then, we review how the OECD as a bureaucracy is prone to pathological behaviour and how this, in turn, can lead to a multitude of different ideas on the same topic within the IO. Kingdon's multiple streams approach, was initially developed in 1984 to explain US policy processes, it has now been extensively applied in various contexts and could also be used to examine the power of IO initiatives. Furthermore, the model is used in policy analysis to explain why some ideas make it to the political agendaand ultimately lead to political decisionsand why this is not the case for others (Kingdon 1995). Finally, the model emphasises that agenda setting is not so much a linear rational process preceded by a thorough problem definition as it is strongly influenced by chance.
Kingdon's approach is based on three independent streams and outlines that the policy process can be situated into problems, policies, and politics. During the problem stream, problems can become politically relevant issues, for example, through particular events or feedback. For example, the launch of the Sputnik satellite is often interpreted as the realisation that there was a problem with the US education system, which was not able to produce aerospace engineers as capable as the Soviet Union was. During the policy stream, ideas and solutions are formulated and compete for technical feasibility and normative acceptance. These ideas and solutions are usually developed by policy specialists, such as those working on the OECD's PISA Governing Board. Governments then act during the politics stream by launching relevant legislation based on the recommendations made earlier.
In any case, particular ideas influence problem definition, policy solutions, and legislative processes. While several ideas float in the 'primeval soup' (Kingdon 1995), only some succeed in becoming implemented policies. Both the agency and timing determine the ultimate success of an idea (Béland 2016, 230). According to Kingdon, a policy windowor window of opportunity for political change may open up due to an event, especially in the problem stream or in the politics stream, which leads to a political entrepreneur, such as the OECD, putting a particular issue or solution on the agenda. With PISA, the OECD made governments, education experts, and the media equally aware that education systems might have a 'problem' in the sense that they are not sufficiently equipped for preparing their students for the knowledge society and the global labour market (Martens and Niemann 2013). Thus, the OECD functioned as a transnational policy entrepreneur which deliberately presented problems and corresponding policies (such as best practices) at the same time.
The multiple streams approach is widely considered to provide conceptual insight into how, why, and under what circumstances policy entrepreneurs may exert influence on the political process. By applying the three streams, it helps to explain the gap between a policy idea getting attention and the adoption of a meaningful political solution. Kingdon claims that these three streams have to align at the same time to be successful in creating a window of opportunity for policy change. Thus, the OECD found the right moment to advocate problems, solutions, and implications with PISA, something that education initiatives of other OECD sub-departments as well as previous IOs and projects, such as the IEA and TIMSS, missed by other education initiatives of other OECD sub-departments.
The question then turns into why the OECD has different education sub-departments that might speak with several voices, and why they hold contrasting ideas regarding education. The Weberian concept of bureaucracies, however, stresses that IOs feature hierarchical organisation structures, impersonal and rational rule, and a defined thematic scope in which they operate. IOs resemble this concept because they are established by states which themselves are also organised according to this bureaucratic model. Conceiving IOs as bureaucracies implies that IOs are subject to a certain intra-organisational dynamic. IOs are fitted with autonomy and authority to fulfil the designated tasks. While autonomy is reflected in an IO's ability to produce and pursue their own ideas, aims, standards, and preferences as well as act accordingly, the authority of IOs is defined by their characteristics as bureaucracies: being impartial, hierarchical, and functionally specialised. Linking this conceptualisation of bureaucracies with the multiple stream approach above renders IOs as policy entrepreneurs that could put forward problems and policy solutions.
Autonomy, on the other hand, implies that IOs usually have some constitutional leeway in doing their work. Although IOs are set up by states and consist mainly of state delegates, they are able to develop their own positions, ideas, or dynamics because of intra-organisational networks and interactions that can neither be fully anticipated in advance nor comprehensively controlled by the member countries. In other words, IOs as agents are not subdued by their principals. This is because organisations tend to develop a life of their own and can address issues in a way that is neither planned nor foreseen by their founders. Conceptualising this behaviour as pathologies, the approach explains how IOs such as the OECD can become potentially 'dysfunctional' entities in the sense that IOs may initiate unexpected projects in education and exert more power over norms than what was originally delegated to them by states Finnemore 2004, 1999, Centeno in this special issue). The effect of this actorness of IOs sometimes produces undesired outcomes and unintended actions.
This pattern is not restricted to the relationship of the IO with its environment; it can also be applied to intra-organisational developments. Being structured as a bureaucracy, IOs are usually composed of different specialised sub-departments whereby each deals with predefined tasks. The work of Barnett and Finnemore tells us that these subunits 'may develop their own cognitive frameworks that are consistent with but still distinct from those of the larger organisation' (Barnett and Finnemore 2004, 39). Moreover, this implies that different subunits can hold dissimilar views when compared to each other. This contestation between internal cultures within an organisation can lead to dysfunctional behaviour within the entire IO, i.e., the IO holds and communicates different ideas regarding problem identification and appropriate solution strategies. The reason for intra-organisational ambiguity stems from incomplete hierarchical control, where different sub-departments could have different spheres of autonomy and compete with each other within the bureaucratic structure of the IO (Barnett and Finnemore 2004). Subdivisions may battle over material resources (budget, staff, or infrastructure) as well as over immaterial cultural values (Barnett and Finnemore 2004, 30). In sum, 'different constituencies representing different normative views will suggest different tasks and goals for the organisation, resulting in a clash of competing perspectives that generates pathological tendencies' (Barnett and Finnemore 2004, 41). Another example of this pathological behaviour is, for example, a counterproductive enthusiasm with upholding institutional procedures and rules, which can result in unresponsiveness to their surroundings (Barnett and Finnemore 1999). In addition, the OECD is a mandate-based IO and therefore subject to its members' service requests, which bears the potential of further scattering the IO's internal focus. The occasional alleged misfit between OECD's policy recommendations and national circumstances might be understood by this finding as well. Centeno (2018) describes how, despite the input of national experts in the agenda-setting process for publications, the OECD secretariats might still exercise control over the end result. When confronted with a conflict between members' requests and internal procedures, the sub-departments might choose different paths to follow. These findings suggest that the OECD not only has the potential for multiple perspectives being integrated due to the multitude of different sub-departments but also has the institutional structure through these multitude of departments that further predetermines this possibility.
In conclusion, our theoretical approach to contextualise PISA within the OECD's architecture aims to explain its perceived predominance within the IO and to also assess why its approach to education is more dispersed than unitary. By highlighting the developments of the OECD's work in education, we show how viewing education as something that fosters economic growth and generates human capital became the centre of the IO's interpretation of the purpose of education. It is important to note that the predominance of PISA (and its associated ideas) could not be solely attributed to processes within the OECD. External factors that could not be controlled by the IO substantially contributed to PISA's fame and the hegemony of the economic interpretation of education purpose. Eventually, the perception of PISA by the outside world made it so successful and boosted the idea of economic-oriented education purposes.

Contextualising PISA within the OECD
The field of education was part of the organisation's agenda since its commencement in 1961. However, the importance of education within the organisation increased dramatically over time. With the successively expanding significance of education, varying notions regarding the purpose of education were also developed and promoted within the OECD. Initially, a functional approach to support economic stabilisation through education was used (Ydesen and Grek 2020). Only with the establishment of the 'Centre for Educational Research and Innovation' (CERI) in 1968 and the 'Education Committee' in 1970 was education policy more formally institutionalised within the OECD.
From the early 1960s to the mid-1970s, the emphasis of the OECD's education activities progressively shifted toward social and equity objectives and became even more closely linked to issues related to the labour market and economic growth (Eide 1990;Klees 2008). In consonance with the leading economic paradigm of the time, Keynesianism, the OECD showed its support by advocating for increased state intervention in multiple policy areas (Armingeon 2004). Correspondingly, the independence of education within the OECD was disrupted since it became increasingly linked to other policy fields and was discussed under topics of economic purpose and efficiency.
In the 1980s, the OECD moved from a Keynesian to a neoclassical supply-side orientation which focussed on reducing government regulation (Lingard and Sellar 2013;Mahon and McBride 2008;Woodward 2010), and the 'free play of the market became the new magic formulas' (Zohlnhöfer and Zutavern 2004, 127). Thus, neoliberalism eventually constituted the leading paradigm in economics and also affected education policies (Dale 2005;Klees 2008). In addition, the focus shifted to a market-oriented approach for education, promoting privatisation and social equity as necessity for human capital and innovation. Through the 'New Public Management' (NPM) agenda, the national governments are to reduce their responsibilities to 'governing at a distance', while local actors are held accountable for producing the desired output (Lingard and Sellar 2016). Accordingly, a swing from input to output orientation occurred in the early 1990s (Papadopoulos 1994).
While the 1970s and 1980s were characterised by a philosophical and ideological debate about the nature and application of performance measurement in education, the OECD explored issues for educational reform, social equity, and innovation in more conceptual and philosophical rather than empirical terms. Introduced in the early 1990s, this perspective instituted statistical indicators for learning outcomes. This decade paved the way for a neoliberal agenda, with the redefinition of lifelong learning as a tool to allow for a more flexible adjustment of peoples' knowledge and skills, catering to the changing labour market (Rizvi and Lingard 2006). During this time, the OECD went from 'philosophical doubt to statistical confidence; from covering some countries to covering most of the world; from a focus on inputs to a focus on outputs; and from occupying an experimental status to being a central part of the Organisation's educational work' (Lingard and Sellar 2013). The most recent shift in the OECD's leitmotif is the strengthening of the neoliberal agenda, ensuring social cohesion, and fostering people's well-being and competitiveness in a fast-moving globalised world (Niemann and Martens 2021).
Shifting ideological paradigms were also reflected in the institutional setup of the OECD. Over time, the OECD's education organisation became multi-layered, more specialised, and diverse.
Education was organised from 1960-1961 in the OEEC directorate 'Office for Scientific and Technical Personnel ', from 1961-1975 in the 'Directorate for Scientific Affairs', from 1975Affairs', from -1991 in the 'Directorate for Social Affairs, Manpower and Education', and from 1991-2002 in the 'Directorate for Education, Employment, Labour and Social Affairs'. The initiation of yet another organisational layer in the 1980s, the 'Indicators of National Education Statistics' (INES) programme, prompted a transfer of responsibilities between the OECD and its member states (see contribution of Grek and Ydesen in this special issue). The OECD started shaping national indicators for education policy, thereby claiming its agency and developing a profile as a semi-independent actor. With INES, the organisation of education initiatives within the OECD also became more diverse because INES was not functionally equivalent to the Directorate for Social Affairs, Manpower, and Education.
While education was previously organised in OECD directorates that dealt with various social affairs, the establishment of the Directorate for Education in 2002 gave the area of education a distinct emphasis. After the establishment of the Directorate for Education and the Global Forum on Education in 2005, the inner-organisational status of education was strengthened, and it became clear that the OECD's work on education occupied a distinctive niche within the IO. The establishment of educational bodies created bureaucratic structures that enabled the OECD to become an increasingly independent producer and disseminator of knowledge (Morgan and Shahjahan 2014). The institutionalisation also ensured that education was interpreted as an issue that was able to have an impact on the development of member states and thus deserved its organisational status within the OECD.
Despite or maybe even due to the many different subdirectories, the OECD has only one flagship project. PISA is now seen as the major carrier of the OECD's education ideas and serves as a governance resource to influence its member states' policymaking. In combination with supplementary analyses, PISA gave implicit and explicit recommendations on national policy reforms. This has also contributed to how various national and international actors perceived education. In short, by continuously expanding its competencies in education, the OECD became one of the most important IOs concerned with education and the PISA project marked the IO's peak activity in education. With the well-thought-out broadcasting of its findings, especially with PISA, the OECD creates a competitive environment and a perceived collective consciousness, causing countries to implement suggested policy reforms and standards in the hope of presenting themselves in a favourable light in this group. Despite not all countries participating in PISA at the moment, doing so is already one of the aforementioned standards and the pressure to participate increases exponentially with the use of this index. Compliance can also demonstrate a country's willingness to be part of this Western-oriented community of well-developed countries and increase visibility on a global level, potentially drawing funds to boost the economy (Addey et al. 2017;Kijima 2015).
The OECD is greatly contributing to the 'pool of accepted ideas' of what form education policy should take, influencing and limiting the shape of education policies and reforms from which countries can choose. This creates homogeneity among countries worldwide. The OECD's recommendations therefore not only influence its member countries but also reach non-member states all around the globe. With PISA, the OECD established a common mindset among PISA-dedicated states, gave normative advice for policymaking, set soft standards, and helped actors to define and articulate their positions. Its extensive reach makes the exploration of the OECD's body of publications even more relevant, as there may be a pattern of interest or repeated discussions about certain views on education that are not obvious but still implicitly influencing education policies around the world.
In sum, the rather unspecific nature of the status of education (in the Convention and other official treaty documents) allotted the OECD some leeway, and the IO could develop its education approach and its working practices (Woodward 2008). This means that the loose reference to education allowed the IO to carve out its take on education without being bound by a detailed directive from the member states.
More than just PISAthe education topics of the OECD over time Taking the predominance of PISA and associated OECD activities and recommendations into account, it is easy to deduce that the OECD ideas in education circulate around human capital generation and utilitarian aspects of education. However, analysing the entire publication record of the OECD's education branches shows that the OECD is more than just PISA and that economicoriented education ideas are not uncontested within the IO. Using the methodological tool of topic modelling, we provide a denser picture of the OECD's output in the field of education.

Methods
The method of topic modelling allows researchers to quantitatively analyse large corpora of textual data, potentially bypassing long hours of reading and coding. In short, topic modelling is an unsupervised machine learning algorithm that is being trained to discover a set of topics in a latent space derived from a corpus of documents. The topics are essentially multiple lists of words, whose composition results from the probability of the word co-occurrences in the entire corpus of texts. The topics are shared between documents. Thus, some topics appear in multiple documents, some in just one, depending on their exclusive word composition. A topic, therefore, represents the words with the highest probability of forming one 'cluster', while being mostly exclusive to that cluster. This allows for the differentiation between different sets of vocabulary used or different discursive streams.
Practically, topic modelling can be seen as speed reading the corpus. The interpretation of the final topics is therefore partly dependent on the researcher's background, and the results can vary to a certain degree between model runs and interpreters. Not all information in a corpus will be extracted, but the topics used most often will be found even in the repeated analysis. It allows for the detection of underlying themes and the repeated discussion of certain issues. The advantage of this method over qualitatively coding the documents clearly lies in the simultaneous analysis of extensive amounts of documents. However, due to its deductive nature and the superiority of human coding to computer algorithms, and the intuitive understanding of word compositions which the algorithm is lacking, it is often recommended to combine topic modelling with a qualitative approach to verify or contrast results (Chang et al. 2009).
The analysis in this publication includes almost all documents between 1961 and 2018 found under the keyword 'education' in the OECD's online library, as well as further documents classified under this keyword collected from the OECD archive in Paris in 2019. 2 All documents were English, except for very few French abstracts that could not be removed and later formed their own topic. Key phrases, such as 'education policy', were combined and the text was lemmatised to remove a double count of pluralisation. The following section contains the results of the topic modelling analysis. First, after a model evaluation, the proportions of the topics in the corpus are discussed along with their distribution over time.
The structural topic model framework (stm) by  used in this publication allows for the joint estimation of topics and their relationship to document-level metadata, such as time or authors. Since this paper is meant to give an overview of the OECD's overall discussion stream, a model with 26 topics was deemed appropriate . A table of the top 20 words for each topic can be found in Appendix A. Most topics represent one concept that can be identified by the first 20 words (often even less) and can be confirmed through inspecting the 5 documents that comprise the topic most. The level of word incoherence or seemingly mismatch of topics and documents is as expected for the corpus. The topics are labelled according to their content and documents. Figure 1 illustrates the different topic proportions within the corpus. Topics on management and higher education as well as the labour market clearly dominate the corpus. Topics on PISA as well as students or children with disadvantages play a much smaller role, with only three topics dealing with PISA exclusively. In fact, the funding and management of education institutions is more frequently discussed than schooling in general. Figure 1 shows the different topic percentages in a stacked area chart, depicting their variation over time. An important note, the chart does not account for the number of publications per year, the topic percentages are shared throughout the entire corpus, adding up to 100% at all times. The distribution of colours shows that management, the labour market, and system evaluation have always played a role but can be further distinguished into multiple discursive streams with slightly different vocabulary or discursive orientations. Figure 2.
It becomes clear that there has been a substantial increase in topics over time and that the thematic scope of the OECD in education has widened. Up until the year 2000, few topics had higher percentages, whereas, from roughly 2015 onward, the percentages become much more evenly distributed throughout the publications.
Interestingly, the topics that are more prevalent earlier on are more concerned with the analysis and management of education systems, higher education, and especially the labour market. Between 1975 and 1995, the institutional management in higher education (topic 13) is discussed with extreme frequency. In recent years, however, the focus seems to be turning more toward schoolinga logical conclusion, since PISA only got started in 2000 and focuses on the secondary school level. There is still a recurring interest in labour market skills and vocational training (topic 7) starting from 2015. The focus on labour market-oriented topics (25, 26, 7) shifted from the analysis of labour market needs to the training of people for the labour market. This indicates a shift from evaluating the status quo toward reforming the system, mimicking the shift from input to output orientation (Papadopoulos 1994). Only three topics are found to be concerned with immigrant children in school, children with disabilities, or children at risk in general (topics 14, 22, 23), and all have a relatively low overall prevalence. Meanwhile, more topics are discussing higher education (topics 6, 11, 13, 17) or are somewhat work-related, either through analysis of the market structure or training (topics 3, 26, 25, 5, 7, 12). PISA is represented by only three topics each targeting specific aspects of the study: performance, questions, and data analysis (topics 1, 10, 20).

PISA, perceptions, publications, and pathologies
The interesting aspect of this result is the seemingly unequal prevalence of discussion on PISAor even secondary educationwhich stands in stark contrast to the previous image of the OECD painted by research. It appears that even though the OECD is mostly known for its influence through PISA, its main focus is not PISA at all. Even before the year 2000, one might have suspected reports discussing similar topics to PISA, the standardisation of education for example, but this has not been the case. Even with a limited model that only includes publications after the start of 2000, the prevalence of PISA-related topics only goes up from 6.98% (in this analysis) to approximately 12% despite the fact that PISA is what cemented the OECD's status in the field. This underlines our argument that the OECD's survival and PISA's influence are strongly dependent on the window of opportunity the OECD created for itself through their strategic marketing of PISA as a policy tool. The prominence of PISA is therefore in our opinion not only the OECD's doing: The frequent public discussion of PISA results by research and policymakers creates the overall impression that PISA is the OECD's 'favourite child'. PISA's success in contrast to other topics could be explained by Kingdon's framework, as stated earlier: The OECD problematises the (in)effectiveness of education systems on the grounds of their own analysis of the labour market structure, thereby creating a window of opportunity to act as a policy and knowledge broker for policymakers. This ensures the IO's relevancy and supports its hegemonic position within world society. The way in which PISA results are published and presented highly influences how they are perceived. While the results from country reports are not as contested, they are not presented with the same rigour as PISA results.
The results lead to further noteworthy observations: Education has been managed under different departments up until 2002, as previously discussed. Hence, the bureaucratic structure was more diverse and allowed for different (and competing) views on education. This coincides roughly with an increase in topic diversity, which could indicate that a change in management structures severely affected what was published previously. Since education was always integrated in departments relating to i) Science, and ii) Human capital (Manpower, Employment, Labour), not only the heavy focus on the labour market might stem from this setup, but also the discussion of seemingly similar topics with different orientations. The OECD still has a mandate from its members, the frequently commissioned reports on the quality of higher education might increase the prevalence of these topics. However, the seemingly uncoordinated, parallel discussion of similar education topics is a striking observation, especially from the standpoint of organisational theory. This demonstrates two things: first, the OECD is certainly not the unitary actor it is often perceived to be, and second, the organisational structure highly influences its output, potentially limiting its chances at establishing one coherent education idea. While it proves to be difficult to present one stream of OECD education ideas, it is possible to demonstrate that the OECD might not have just one education idea but many. This further leads to the suspicion that the people within the different departments of the OECD are crucial in influencing these ideas and shaping their agendas, ultimately leading to a diffuse picture of an IO internally fighting for dominance. This argument links back to Barnett and Finnemore's ideas on pathologies of different IO sub-departments leading to different ideas on the subject matter. This resulting image of the OECD as a rather inconsistent actor should not come as a surprise, given the number of sub-departments responsible for education.

Conclusion
In this article, we took a closer look at the OECD's publication output regarding education. The IO is often pictured as a monolithic and influential actor in education policymaking, dispersing an economic view on education reinforced by PISA data. What we found, however, is slightly less unified. The majority of the OECD's output does not focus on PISA or secondary education at all. Most publications on education are discussing finance, management, or the labour market connection. Ideological developments and finding solutions to the big problemssuch as how to relinquish class inequality, for exampleare rare and barely detectable in the overall corpus. In addition, multiple topics discuss similar issues but are nonetheless distinguishable from each other. This implies that the success of PISA, one of many topics the OECD is discussing, can be explained by drawing on Kingdon's multiple streams framework: PISA was published within the right time frame, subsequently presenting a problem (inefficient education systems identified through PISA scores) and the solution (policy recommendations based on country reports and PISA scores). While PISA has ensured the organisation's relevance, other topics are accepted more reservedly. The somewhat dysfunctional organisational structure of the OECD's education department can be partly blamed for this development. Various sub-departments publish reports on similar topics, causing the OECD to develop not one but multiple views on education. As such, the IO developed its distinct pathologies by featuring several organisational units concerned with overlapping education topics but with no overarching institutional architecture. Despite the discussion of similar topics from multiple perspectives, the heavy focus on management and the labour market connection remains.
We conclude that first, the OECD is not a homogenous actor and addresses much more topics than PISA, and second, within the IO itself, PISA as a subject is far less popular than expected. Due to the organisational pathologies, the internal view on education is, and has always been, diffuse and inconclusive, however, other topics do not seem to fit into the current problem stream as they are less influential than PISA.
Therefore, further research could take into account the development of the OECD's education ideas and the influence of these perspectives on OECD policy advice. Keeping the multitude of topics and sub-departments in mind, it might be interesting to assess the impact of the 'social factor' within a bureaucracy similar to the OECD and the development of such an IO. Andreas Schleicher, for example, has been the driving force behind PISA since its beginning. Looking into the individuals and social dynamics behind the different departments might shed further light on how certain topics could react on the policy window necessary for their public success. chapters that have been published by the OECD separately are excluded. Due to the fact that Working Papers do not reflect the OECD's own agenda but that of the author(s), as stated in most of the papers, this type of publication has been excluded. In the pre-processing procedure, title pages, headers and footers as well as tables, graphs, and bibliographies have been removed along with the usual stop-words.

Disclosure statement
No potential conflict of interest was reported by the author(s).