Publié le 5 décembre 2025 – Mis à jour le 11 décembre 2025
Lieu(x)
Pôle Tertiaire - Site La Rotonde - 26 avenue Léon Blum - 63000 Clermont-Ferrand
Salle 212

Séminaire de recherche. Multiple imputation of missing values in longitudinal economic data: is simpler better?


Anne Viallefont
CERDI, Université Clermont Auvergne

Résumé

Multiple imputation (MI) is an efficient and simple enough way to handle missing data and get unbiased estimates, even when there are numerous missing values, and when the process that generates missingness is not 'completely at random'. However, when dealing with economic panel data, the values that are generated by most available MI routines are often unrealistic, which could largely increase the estimated variances. In this preliminary work, I generate missing values in complete panel data, under various schemes, and I use three different data-completion algorithms, including a very simple algorithm that I propose. I then compare the quality of the results in terms of bias correction and variance inflation.