Ridderprint

Introduction 15 important decisions that have to be made, for instance, regarding construct operationalization, data handling, or modeling steps. These decisions will be led largely by the goal of the research and the type of modelling that is applied (e.g., descriptive, exploratory, explanatory, predictive; Shmueli, 2010; Woo et al., 2017). The stages of a conventional HRM research process – focused on building theory through explanatory modelling – will therefore frequently differ from those of a people analytics project – focused on organizational utility, regardless of the type of modelling. These differences can occur in many ways. For instance, construct operationalization in conventional HRM research is largely determined through theoretical justification and prior scientific validation. If scholars want to measure employee engagement , they turn to previous literature to examine how they may measure each of its theoretical dimensions with a validated scale. In contrast, availability, organizational legacy, stakeholder requirements, and predictive and benchmarking utility will largely determine how constructs are operationalized in a people analytics project. On the one hand, these factors have affected how HRM phenomena have been measured in the past and thus what data may already be conveniently available (e.g., archival data on employee engagement by the definition of the organization under study). On the other hand, new people analytics initiatives will have to make concessions in order to gain organizational buy-in, thus affecting what and how data can be gathered in any future studies. Similarly, complex (e.g., high dimensional, high volume, high velocity), unstructured (e.g., image, sound, text), and/or dirty data (e.g., missing values, errors) are often valuable for people analytics projects but are less easily leveraged in conventional HRM research contexts, due to the data’s unconventionality and its lack of theoretical foundation. Other potential differences between the statistical modelling processes of people analytics and conventional HRM research relate to the used methods, model evaluation processes, and model selection criteria (see Shmueli, 2010; Strohmeier & Piazza, 2013; Yarkoni & Westfall, 2017). In sum, the differences can be plentiful. 1.2.3.3 Potential Similarities Important to note is that people analytics and HRM research are not necessarily different. People analytics merely seems to follow a more inductive approach, starting with the purpose in mind, and is thus more flexible in terms of the procedure to best fulfill this purpose (Woo et al., 2017). The modelling process can be matched to the HRM issue at hand rather than necessarily conforming to the conventional procedures. Still, any people analytics project can be very much similar to conventional HRM research. For instance, a people analytics project can consist of a replication of an earlier scientific study in the own organizational context, in order to inform decision-making. Alternatively, a people analytics project with the purpose of informing organizational decision-making could demonstrate value for the academic community and be published scientifically. Increasingly, scholars and practitioners are teaming up to conduct people analytics research that holds both academic and direct practical value (e.g., Harter, Schmidt, & Hayes, 2002; Kryscynski et al., 2017; Van de Voorde, Paauwe, & Van Veldhoven, 2010).