zoom link, opens an external URL in a new window

Meeting-ID: 280 519 2121
Passwort: 584190

Abstract:

International education statistics have long been based on official administrative data that was treated as observed fact. The demands of the indicator framework for Sustainable Development Goal 4 on Education (SDG4) have required the increasing acceptance of sample-based data sources, such as large-scale household surveys or learning assessments. Customary publication of "latest available" data points is no longer tenable given the variability and often contradictory signals from such sources. Facing similar challenges, the health sector successively developed and endorsed global estimates of infant and maternal mortality based on sophisticated statistical modelling. Building on these experiences, both in terms of methodology and acceptance by the global health statistics community, similar approaches have recently been adapted for education monitoring. In particular, in a first for the education sector, Bayesian estimates of school completion rates combining information from different survey and census sources have been endorsed as official SDG4 monitoring data, despite the complexity of the model and its lack of transparency for government stakeholders. I will discuss the interaction of statistical, institutional, and domain-specific challenges from an inside perspective, such as questions of information sharing between countries' estimates within a Bayesian framework via hyper-parameters, the lack of gold-standard "ground truth" data for calibration of estimates of source-specific bias.