ESRA 2027 sessions by theme
Back to Overview of Sessions
Statistical Inference under Unknown Selection Mechanisms |
| Coordinator 1 | Dr Lisa Braito (University of Florence) |
| Coordinator 2 | Professor Emilia Rocco (University of Florence) |
Session Details
Statistical inference typically relies on assumptions about the mechanism governing the selection of observed data from a target population. In many empirical settings, however, this mechanism is unknown or only partially specified and may depend on observed or unobserved characteristics of the units, resulting in informative selection and potentially biased population inference. Such challenges arise, for instance, in the presence of missing data due to unit or item nonresponse, as well as when inference is based on non-probability data. The latter encompass a heterogeneous and rapidly expanding range of sources, including convenience and volunteer samples, opt-in panels, administrative records, electronic health records, social media and digital-platform data, and web-based or crowdsourced data, whose increasing availability creates substantial opportunities for statistical analysis while raising fundamental questions about selection, representativeness, and generalizability. Without attempting to provide an exhaustive account of the methodological developments in this broad area, this session brings together heterogeneous yet complementary contributions concerned with inference when the available data are non-representative of the target population, or when their degree of representativeness is unknown or cannot be reliably assessed. Rather than focusing on any specific data source or selection mechanism, the session adopts a broader perspective on the inferential challenges arising when the process generating the observed data is only partially understood. By bringing together different methodological perspectives and application domains, it aims to uncover common structures across problems often studied separately, clarify the assumptions required for identification and estimation, and discuss statistical strategies for drawing valid population-level conclusions under limited information about the selection and data-generating mechanisms.
