A review of text analysis in human resource management research: Methodological diversity, constructs identified, and validation best practices

Abstract

Discovering and producing reliable and valid measures of psychological constructs are central aims for human resource management (HRM) researchers and practitioners. While HRM researchers have historically relied on traditional quantitative methods, increased accessibility of text analysis techniques enabled by advancements in machine learning make qualitative data more convenient to analyze and include in decision-making processes. In this review, we systematically analyze research in HRM, organizational behavior, strategy, and entrepreneurship that has used text analysis to uncover and/or measure constructs. Our goals are to 1) delineate types of text analyses (categorization, dictionaries, supervised machine learning, and unsupervised machine learning), 2) review what constructs can be derived from text data, 3) describe how those constructs have contributed to the core HRM functions, 4) provide guidance on validation efforts that are needed to trust inferences made, and 5) and identify future research opportunities to use text analysis by HRM function. We support these points by conducting two text analyses on the papers in our review: a hand-coded content analysis using an existing framework and building a topic model of the abstracts. We find that while there is convergence (triangulation), there is notable divergence such that the topic model revealed more nuanced and useful clustering in significantly less time, thus illustrating the value of different types of text analysis. We encourage HRM researchers and practitioners to use machine learning to increase efficiency, reduce subjectivity, increase replicability, and facilitate methodological diversity. We close with a brief discussion on the promise of large language models.