Coarsened data in theory and practice: challenges, considerations and opportunities

Leiden Repository (Leiden University)

Abstract

The focus of this dissertation is on coarsening. Coarsening means that instead of observing the true value of a variable, a strict subset of the variable's possible levels is observed. The most simple example is a variable with three levels: a, b and c, where one observes the combination \{b, c\}. The true variable value is unknown (it is either b or c), only the value a can be excluded. The cause of coarsening can be manyfold, for example, limitations in the measurement process, but can also be induced intentionally to reduce model complexity. When the goal is to include the coarsened observations in an analysis, a statistical approach is required to properly take the coarsening information into account. Two categories of approaches are often used: Multiple Imputation and a problem-specific Expectation-Maximization algorithm.

Authors 0

1. Author list not loaded yet.