(From left to right) Anabel Forte Deltell, Gonzalo García-Donato Layrón, Stefano Cabras, Mª Eugenia Castellanos Nueda, Alicia Quirós Carretero.

A research team from five Spanish universities, including the University of Valencia (UV), has developed a new statistical methodology that makes it possible to select relevant variables even when information is incomplete. The study has been published in Bayesian Analysis, one of the most prestigious journals in the field.

The work takes a novel approach to the problem of quantifying the uncertainty associated with the statistical models used to represent reality, taking into account the possibility of data containing missing values – a very common situation in scientific studies and real-world data analyses.

“This research represents a milestone in that it efficiently incorporates all sources of uncertainty that arise when data are collected only partially or lost in the process. The alternative, in many cases, is to completely discard the sample (for example, a patient) that had missing information possibly only in one of the recorded variables. Our methodology therefore makes it possible to make better use of the information that has been collected, avoiding the biases that could arise from eliminating it entirely”, explained Anabel Forte Deltell, researcher at the Department of Statistics and Operational Research at the University of Valencia and co-author of the study.

In contrast to the usual strategies, which involve disregarding data with missing values or using variable selection methods that produce false positives, this work proposes an objective Bayesian methodology that directly integrates the uncertainty about the missing data into the inference process.

This approach improves the robustness of the models, avoids common errors in variable selection and yields more reliable results, even when the datasets contain large proportions of missing values.

The article, “Model Uncertainty and Missing Data: An Objective Bayesian Perspective”, has been selected for public discussion by experts – a distinction reserved for the most significant contributions, according to the editor-in-chief of Bayesian Analysis. The discussion will take place in a webinar on 5 November.

In addition to Anabel Forte Deltell, the research team includes Gonzalo García-Donato (University of Castilla-La Mancha), María Eugenia Castellanos (Rey Juan Carlos University), Stefano Cabras (Carlos III University of Madrid) and Alicia Quirós (University of León). The study was funded by the Spanish Ministry of Science, Innovation and Universities and by the Spanish Research Agency through the Knowledge Generation project grants.

Reference:
Gonzalo García-Donato. María Eugenia Castellanos. Stefano Cabras. Alicia Quirós. Anabel Forte. “Model Uncertainty and Missing Data: An Objective Bayesian Perspective”. Bayesian Anal. Advance Publication 1 - 73, 2025. https://doi.org/10.1214/25-BA1531