UCPH Statistics Seminar: Bojana Milošević

Title: Missing data problems: a model specification testing perspective

Speaker: Bojana Milošević, Faculty of Mathematics, University of Belgrade

Time & location: June 16, 2025, 3:15pm, MATH 04-4-20

Abstract: Missing data problems are common across data-driven disciplines. While estimation methods have been widely explored, relatively little attention has been given to model specification testing in the presence of missing data. In this talk, we examine goodness-of-fit, independence, and two-sample testing in the context of missing data, and present a survey of recent methodological developments. We begin by outlining the main types of missingness and their mathematical implications. The focus is on multivariate settings, with particular attention to data missing at random and related advances in testing procedures. A central question addressed throughout is whether it is more effective to impute missing data or to adapt the test statistics themselves. We discuss the strengths and limitations of both strategies. Given the close connection between many test statistics and U-statistics in this setting, we present some general results for U-statistics. The talk concludes with a summary of current research directions and open challenges, highlighting opportunities for future work.