The test-negative design is a variant of the case-control study, which has been widely used to estimate influenza vaccine effectiveness and is increasingly being applied to other interventions. In these studies, patients meeting a particular clinical case definition are recruited and tested for the condition of interest. Vaccine effectiveness is estimated from the odds ratio comparing the odds of vaccination among those testing positive versus those testing negative. The design is purported to reduce bias associated with healthcare seeking behaviour and misclassification of outcome status. However, validation of the design has largely been done on simulated datasets that mimic a surveillance system rather than using administrative data, where clinical case definitions may be missing and testing practices may vary. Absent a common case definition, assumptions about the similarity of cases and controls may be violated. Similarly, when testing is performed by multiple laboratories that do not use a single unifying protocol, outcome misclassification may be both dependent and differential. In this paper, we use causal graph theory to explore the biases associated with leveraging administrative data versus collecting data through a controlled surveillance system. The models were tested in simulations to examine the direction of potential biases. This study underscores the importance of carefully considering the assumptions inherent in a study design, particularly with respect to control selection, before applying it to administrative data.