Ridderprint

Chapter 7 154 provides a way to assess the real-world value of our HRM models through replication at little to no extra cost (Yarkoni & Westfall, 2017). On the other hand, cross-validation becomes increasingly important with the rise of algorithmic methods in the HRM domain (see Strohmeier & Piazza, 2013) which are capable of modeling complex interaction and non-linear effects, and are thus prone to overfitting (e.g., Breiman, 2001). In order to build accurate predictive models – estimating the likelihood that employees will perform and be retained, or the likelihood that HRM interventions will be successful in specific contexts, for specific employees (e.g., personalized HRM) – the implementation of machine learning techniques such as cross-validation is critical. In sum, cross-validation can help HRM scholars in building explanatory and predictive models that do not overfit the coincidental patterns in research samples, but rather generalize to the wider population, and thus have value for real-world applications. 7.3.4.2 Exploration-Exploitation Tradeoff A second, related learning for HRM and people analytics involves the exploration- exploitation tradeoff. This tradeoff considers the competing needs to, on the one hand, acquire new knowledge and information (exploration) and, on the other hand, optimize decisions based on existing knowledge and information (exploitation). By focusing resources on exploration, one gains information on the payoff of different options but retains less time to benefit from these options. By focusing resources on exploitation, one gains direct benefits but may potentially overlook alternatives with a higher payoff, resulting in suboptimal benefits on the long term. Although the exploration-exploitation tradeoff has been discussed previously in relation to organizational learning and survival (e.g., Gupta, Smith, & Shalley, 2006; March, 1991), the rise of evidence-based HRM and people analytics emphasizes its importance. The HRM function is inherently oriented towards exploitation, but may want to explore more. Indeed, some exploration is central to the HRM design phase as practitioners may consult stakeholders concerns, scientific papers, and organizational metrics when designing and implementing HRM policies and practices (Rousseau & Barends, 2011). On top of this, progressive HRM departments may even conduct formal pilot studies (i.e., people analytics) to validate that the implementation achieves the right outcomes. However, once these steps are completed, policies and practices often remain unchanged for as long as they (seem to) benefit the organization. The knowledge that an HRM policy or practice works is exploited whereas ongoing, organized exploration of alternatives and modifications is rarely conducted. While this focus of knowledge exploitation may have been viable for the traditional management of personnel in organizations, the current speed of change and digitalization requires and allows for more systematic exploration. For instance, let us revisit the selection decision-support system discussed earlier, recommending job candidates based on their predicted likelihood of being a successful hire. Ultimately, the effectiveness of such a system depends on the continuous exploration of new information and data patterns. For instance, imagine that the skills and competencies that jobs require change over time. A system focused purely on exploiting an established (i.e., trained) model will