Time to Include Time to Death? The Future of Health Care Expenditure Predictions

Abstract

Government projections of future health care expenditures - a great concern given the aging baby boom generation - are based on econometric regressions that do not explicitly control for end-of-life expenditures. Because expenditures increase dramatically on average at the end of life, predictions based on regressions that omit time to death as an explanatory variable will be biased. Specifically, not controlling for time to death is an omitted variable problem that biases the estimated effect of age upwards. Although health care expenditure predictions for a current sample will not be biased, predictions for future cohorts with greater longevity will be biased upwards, and the magnitude of the bias will increase as the expected longevity increases. We explore the empirical implications of incorporating time to death in longitudinal models of health expenditures for the purpose of predicting future expenditures. Predictions from a simple model that excludes time to death and uses current life tables are 10 percent higher than from an expanded model controlling for time to death. The bias increases to 16 percent when using projected life tables for 2020. The predicted differences between the models are sufficient to justify reassessment of the value of inclusion of time to death in models for predicting health care expenditures.