Mean Reverting Measurement Error in Survival Expectations*

Recent literature found that subjective survival expectations in the Health and Retirement Study (HRS) appear to be overly optimistic among older respondents compared to life table values and actual survival rates. This paper argues that a significant portion of this bias is due to measurement error as opposed to biases in survival beliefs. HRS asks about the subjective probability of living ~12 more years. The life table 12-year survival probabilities are decreasing with age and they are close to zero among the oldest respondents. As the used probability scale in the HRS (0-100%) is bounded, measurement error is likely to be biased away from the boundaries. The paper shows that the extent of this bias is likely to be large. I use a simple measurement error correction model, which assumes that measurement error is mean-zero in the logit of the probabilities. The logit transformation leads to an unbounded scale between minus and plus infinity, and the meanzero measurement error assumption is more reasonable in this metric. The estimated corrected survival beliefs are much closer to actual survival probabilities than the raw survival reports. I discuss the limitations of the simple measurement error model used in the paper, and I discuss a more promising alternative which will be estimated in the future using to-be-collected data in an experimental module of the 2014 wave of HRS. ____________________________ *Support from National Institute on Aging program project grant 2-P01-AG026571 is gratefully acknowledged. Please send comments or questions to [email protected].