Out-of-Sample Fusion in Risk Prediction

Abstract

The probability that mortality from certain causes exceeds high thresholds is addressed. An out-of-sample fusion method is presented where an original real data sample is fused or combined with independent computer-generated samples in the estimation of exceedance probabilities assuming a density ratio model. Since the size of the combined sample of real and artificial data is larger than that of the real sample, the fused sample produces short confidence intervals relative to traditional methods. Numerical results show that the method maintains good coverage even for some misspecified cases.

AMS Subject Classification:

• Primary 62F40
• Secondary 62F25

Keywords:

• Mortality
• Density ratio model
• Threshold probabilities
• Tilt
• Semiparametric
• Coverage

View correction statement:

Corrigendum

Acknowledgments

The authors are grateful to the reviewers and to Don Malec for helpful remarks and suggestions.

Funding

Research partially supported by NSF DMS-1007647 and by USDA/NIFA. The findings and conclusions stated in this article are solely those of the authors. They do not necessarily reflect the views of the National Center for Health Statistics or the Centers for Disease Control and Prevention.