Supervisors: Frank Thinggaard & Niels Skipper
Discussants: Carsten Bergenholtz & Panos Mitkidis

Abstract
I investigate the degree of concordance between ten commonly use proxies for auditors’ workload and auditors’ severe stress-related treatment. Using two highly confidential datasets comprising medical prescriptions and interactions with the public healthcare system in Denmark from 2015 to 2020, I find that three workload proxies are positively associated with auditors’ severe stress-related treatment: (1) The natural logarithm of the number of clients audited, (2) the natural logarithm of the number of cross-industry clients audited, and (3) whether an auditor’s workload exceeds the 80th percentile in the sample. Three proxies, including the commonly used indicator for when clients possess fiscal year end in December month, are unassociated with auditors’ treatment whereas the remaining proxies provide mixed results. Next, I propose new proxies for capturing auditors’ severe stress-related treatment by combining insights from existing proxies with modern machine learning techniques. The proposed proxies outperform existing proxies by exhibiting stronger associations with auditors’ treatment, greater (adjusted) R2, as well as lower AIC, and BIC, and exhibits economically meaning improved classification abilities. The findings may provide guidance to researchers investigating the implications of auditors’ severe workload pressure in choosing a suitable proxy and constructing their research design.

Everyone is welcome!