Primary Submission Category: Generalizability/Transportability
Revisiting Representativeness
Authors: Haidong Lu,
Presenting Author: Haidong Lu*
The concept of representativeness has long been a key concept in epidemiology and other social sciences. Nonetheless, the interpretations of what constitutes representativeness and non-representativeness can differ among researchers. Some associate non-representativeness with a lack of generalizability, while others tend to conflate it with collider bias. In this paper, I aim to provide a comprehensive understanding of the underlying mechanisms of representativeness. Specifically, I will introduce and delineate two fundamental concepts: sample representativeness and estimate representativeness. Through the use of two distinct simulation studies involving causal diagrams, I will elucidate the reasons behind the occurrence of estimate non-representativeness when there is sample non-representativeness. Furthermore, I will discuss the settings in which estimate representativeness can be maintained even in the presence of sample non-representativeness.