A prime example of evidence-led policies? The World Bank’s research and operations

International organizations (IO) have become highly productive, but often neglected, producers of scientific research. They often justify this investment by arguing that they use research and data to support their day-to-day operations. This claim is shared by proponents of evidence-based policy-making among bureaucrats, politicians and academics (Davies and Nutley 2000, Sanderson 2002, Pawson 2006, Cairney et al. 2016, Parkhurst 2017). And yet, this claim is studied very little empirically (Walter et al. 2003). In this paper I use a quantitative design to explore the link between projects of an international organization on the one hand and research output on the other. The World Bank may serve as an excellent case study to investigate the relationship, based on the methodology of ‘crucial case design’ as described by Eckstein (1975) and (Gerring 2007). Hypothesizing that international organizations use research to inform operations (Sanderson 2002, Greenhalgh and Russell 2009) or vice versa (Weiss 1972, Banerjee et al. 2010, Duflo et al. 2021), or that research and operations are co-produced (Gibbons et al. 1994, Jasanoff 2004), the World Bank should exhibit a close connection between the two: It has positioned itself as a research powerhouse in the development economics scene (Gavin and Rodrik 2005, Banerjee et al. 2006, Ravallion and Wagstaff 2010), and has set mechanisms in place to ensure that operations and research inform each other (Broad 2006, Broad 2007, Ravallion 2011, Stone 2013). To explore this association, I use a very rough approximation: the quantity of predefined themes in the operations and research output of the World Bank. I created a lexicon based on the World Bank Theme Taxonomy and apply it by means of computerized content analysis to all World Bank’s project descriptions and the title and abstract of all publications authored by World Bank researchers as recorded in the Web of Science in a twenty-year time frame. On the one hand, I plot the correlations in graphs and on the other hand, I analyze them using Granger causality. The paper thus contributes to current literature in two ways: It uses a novel dataset which combines bibliometric information, project information and a self-created lexicon. More importantly, it tests the ideas of evidence-based policy-making empirically using a crucial case design.