Transforming Text into Big data: Leveraging Government Documents to Enhance Theory-Driven Research in Public Administrations

A significant part of public administration is based on creating written documents, which are used to develop policies, laws, and strategies, document compliance with requirements and regulations, and inform and collaborate with the private sector and the public. These documents contain a wealth of information on public administration but are only used in systematic quantitative analyses to a
limited extent and in limited contexts. Therefore, there is significant untapped potential in how large quantities of public documents can be better used in research on public administration.

This dissertation investigates how technological advancements and the use of machine learning and artificial intelligence enable the quantitative analysis of large volumes of public documents. The dissertation refers to this as big data analysis and focuses empirically on contracts between the public and private sectors. Public procurement constitutes a significant portion of the public sector spending and is based on written contract documents. Systematic big data analyses of these contract documents across products and services can provide important insights into public procurement At the same time, they can serve as an example of the potential of big data analysis in public administration.

The dissertation consists of five articles. The first article Is a review of the existing
research that uses big data analysis and identifies a series of challenges for the widespread use of such analyses in public administration research. The existing research only uses text documents to a limited extent and rarely long and unstructured documents, which are characteristic of public administration. Theory is also underprioritized, and the analyses focus narrowly on predicting specific outcomes. This challenges its application in public administration, which quantitatively focuses on estimating how factors influence outcomes based on a theoretical foundation.

Therefore, the dissertation presents two theory-driven approaches to big data analysis in public administration. One approach is used in two articles, where big data analysis is used to measure theoretically developed concepts, which are then used in an estimation analysis. One article develops one of the first quantitative measures of employee protection in outsourcing contracts. The article finds that prioritization of price and financial capacity influences how municipalities protect employees. The other article develops a quantitative measure for green public
procurement and shows how political interests, procurement complexity, and procurement capacity affect how green municipalities purchase. The other theory-driven approach to big data analysis in public administration is based on even stronger inspiration from big data analysis, where the analytical approach is based on prediction. This approach is used in an article that attempts to predict measurability and activity specificity. Previous research has struggled to measure these concepts across many different types of products and services. The article succeeds in measuring these concepts across 178 types of products and services. The article is preregistered to ensure its theory-driven anchoring.

The last article of the dissertation shows that although there Is great potential for big data analysis with text data, this is only realized through theoretically grounded research and qualitative insights about the data material closely related to the domain research field. Overall, the dissertation contributes to how big data analysis and quantitative text analysis can be used to measure and analyze key concepts in public administration, which previous research has struggled
to quantify. The dissertation outlines two theory-driven approaches to qualify future big data analyses of text documents, which can provide new insights into public administration research and contribute to more systematic use of text data in practice.