Machine Learning­based Quantitative Grounded Theory: A New Paradigm for Management Research

The advent of machine learning (ML) has and will continue to reshape practice in many sectors, including management research. ML will yield new insights and theory develop­ ment through advanced analytical capabilities. This thesis focuses on Machine Learning­based Quantitative Grounded Theory (ML­based QGT), an emerging interdisciplinary approach that facilitates a nuanced understanding of complex management phenomena. This work contributes to management sciences by allowing for the possibility of applying abductive reasoning and generating data­driven theories that illuminate novel relationships. 

The power of ML­based QGT is best witnessed when it is applied to large complex datasets. These datasets are not suitable for traditional quantitative approaches like ordinary least square regressions that make stringent assumptions about data distribution, homoskedasticity, and multicollinearity. These limitations, coupled with the fact that complex datasets are also prone to missing values, creates an environment for potentially biased results and theoretical implications. Leveraging ML’s inherent capabilities, ML­based QGT adeptly handles missing values in these datasets, yielding more robust and novel insights. The results generated by ML point to likely explanatory variables for the dependent variables. These potential relationships can be tested and thereby either supporting exiating theory or steering theory in new directions. 

Chapter two demonstrates ML’s prowess by analyzing a comprehensive dataset from the Household, Income, and Labour Dynamics (HILDA) survey, which was previously heavily pruned in past literature due to missing values. Using ML, we no longer need manual data deletion of observations, leading to a paradigm shift in analysis. Whereas previous findings emphasized job insecurity as the strongest job­related predictor of neuroticism, the ML analysis pointed towards employment status, reigniting debates about the roles of sociological variables in predicting neuroticism. 

Chapters three to five utilize the World Values Survey (WVS) data, a large multi­ country multi­wave dataset with about 75% missing data. Unlike previous studies constrained by the data’s complexity, we employed ML to identify shifting attitudes, values, and beliefs over the last four decades across 83 countries. This analysis spotlighted life satisfaction, Protestant work ethic, and prosociality as significant drivers of cultural change over time, variables hitherto overlooked in the cultural change literature. 

ML’s ability to facilitate abductive reasoning, as shown in Chapter four, opens new vistas for theory generation. By building an ML model, the results identified optimism about humanity’s future as a novel predictor of ethical behavior, a result that both surprised and expanded the existing literature on optimism and ethicality. Confirmatory studies established that fostering optimism about humanity’s future promoted ethical behavior, thereby providing a fresh direction for research on ethicality predictors. 

Chapter five showcases the potential of retraining ML models for different purposes on a given dataset. The model built in Chapter four was repurposed to predict respondents’ attitudes towards income inequality, identifying the belief in ’hard work’ as a significant predictor, findings subsequently supported by correlational and experimental studies. 

The final chapter ponders the implications of ML for management sciences, discussing how ML­based QGT can lead to a paradigm shift in the way we apply logic to research questions and conduct data analyses. However, the accuracy of ML findings and their implications for theory are contingent upon the quality and robustness of the dataset, underscoring the need for stringent data handling protocols. 

ML­based QGT offers a fresh lens to interpret management phenomena, enabling the generation of novel insights, refining existing theories, and driving the development of new theories. This thesis points to new research era in which ML becomes an integral part of the methodological toolkit in management research, triggering a shift from purely deductive reasoning towards more inductive and abductive approaches in theory development.