Organizational Reputation and Regulatory Communication: An Analysis of EU Agencies’ Communications Using Machine-Learning Techniques

In view of increased transparency demands, public scrutiny, as well as the blame games, public organisations have to increasingly manage their organizational reputation, as they are perpetually being assessed by manifold audiences possessing conflicting expectations. Recent scholarship has demonstrated that organisations attune their communications to respond to the most relevant reputational threats by emphasising their technical capacity, performative conduct, procedural appropriateness, or/and moral image. However, we still have limited knowledge on: (a) how public organisations manoeuvre to diffuse potential risk of losing a good organisational reputation in the eyes of relevant stakeholders and (b) what explains the substantial variation of reputational repertoire on which organizations draw to legitimise their conduct. This study relies on a bureaucratic reputation account (Carpenter, 2010) to enhance our understanding of the strategic behaviour of public organisations and their endeavours to communicate about their outputs and processes in view of potential reputational threats. Reputation-based explanations originate from the assumption that organizations actively pursue multiple reputation-management strategies to influence the judgements of audiences that monitor and assess their behaviour. Organizations are careful in choosing which audiences they want to please and which signals they want to send in order to craft audiences’ perceptions about their undertakings. To influence audiences’ judgements about specific nuances of their activities (and thereby cultivating their reputation), organizations may choose to send strong signals about: (a) their scientific rigorousness and expertise of their staff (technical reputation), (b) effective performance and their ability to take decisive action (performative reputation), (c) legality of their processes (legal-procedural reputation), (d) values and ethical implications of their activities (moral reputation). Recently, scholars have started to use machine-learning techniques to study signals sent by public organisations to manage their reputation (Anastasopoulos & Whitford 2019). Despite those efforts we still know little of how and why public organizations send out those communications. To fill this research gap, we studied EU regulatory agencies’ communications. With web-scraping, we compiled a large dataset containing communications from EU agencies. In order to deal with the size of the data set (approx. 20 000 texts), we used machine-learning methods to label the dataset. To this end, we manually labelled part of the dataset with the four reputational dimensions. This resulted in a high-quality data set that could be used to train and rigorously evaluate different machine-learning models. The best performing model was then used to categorise the remainder of the communications. Our preliminary results suggest that EU agencies exclusively focus on emphasising the technical dimension of their activities, however, we observe that communication patterns are affected by agencies’ regulatory tasks and the degree of reputational threats that they face.