A multivariate strategy to measure and test global imbalance in observational studies

This paper presents the development of the data driven approach first introduced in Camillo and D’Attoma (2010) and D’Attoma (2009), which enabled one to obtain a global measure of comparability between treatment groups within a non-experimental framework. This paper points to better formalize the global measure of imbalance reported in Camillo and D’Attoma (2010) and D’Attoma (2009) and to introduce a multivariate imbalance test. We consider the global measure of imbalance and the multivariate imbalance test as tools for investigating the dependence relationship between categorical covariates and the assignment-to-treatment indicator variable within a more complex strategy whose final aim is to find balanced groups. We will show in simulated data how the strategy works in practice.

Research highlights
► We introduce a strategy for making causal inference from observational data without model dependence.
► The Global Measure of Imbalance represents a measure of comparability between-groups.
► The Multivariate Imbalance Coefficient expresses the importance of the measured imbalance relative to the total inertia.
► The multivariate test allows to determine if the measured imbalance is significant.