A Simple Information Criterion for Variable Selection in High‐Dimensional Regression
2024

A New Criterion for Selecting Variables in Complex Models

publication Evidence: high

Author Information

Author(s): Matthieu Pluntz, Cyril Dalmasso, Pascale Tubert‐Bitter, Ismaïl Ahmed

Primary Institution: High‐Dimensional Biostatistics for Drug Safety and Genomics, CESP Université Paris‐Saclay, Inserm Villejuif France

Hypothesis

Can the extended AIC (EAIC) improve variable selection in high-dimensional regression models?

Conclusion

The EAIC effectively controls the family-wise error rate in high-dimensional settings, outperforming traditional criteria like AIC and BIC.

Supporting Evidence

  • The EAIC controls the family-wise error rate in nearly all non-correlated settings.
  • Traditional criteria like AIC and BIC often yield many false positives in high-dimensional contexts.
  • The EAIC is designed to be user-defined for specific family-wise error rates.

Takeaway

This study introduces a new way to pick the best variables in complex models, helping to avoid mistakes when many options are available.

Methodology

The study proposes the EAIC and compares its performance with traditional criteria through simulation studies and real data applications.

Limitations

The EAIC's performance may not hold in settings with small sample sizes and strong correlations among variables.

Digital Object Identifier (DOI)

10.1002/sim.10275

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication