Eliciting Subjective Survival Curves: Lessons from Partial Identification

Type d'activité: 
Séminaires du jeudi midi

Luc Bissonnette, CIRPÉE

26 Avril 2012
11 h 30 - 12 h 30
2225 DeSève

Résumé/Abstract :
When subjective expectations are elicited in surveys, researchers often rely on parametric models to characterize the underlying beliefs. We propose a new take on this type of data that does not impose a parametric form on expectations. Using the widely researched example of subjective survival expectations, we derive bounds on life expectancy. We show that these bounds contain interesting information and match them to official life tables released by Statistics Netherlands. While analysis that relies on parametric assumptions rejects consistency of expectations with objective forecasts for women, the non-parametric bounds show that subjective data are on average consistent with forecasts. Though expectations are in line with the life tables in the aggregate, large fractions of both men and women report survival probabilities that deviate from the official figures. These deviations are related to differences in education and health behaviors (smoking). We find a strong inverted U-shaped relationship between age and the fraction of deviating expectations as well as the size of the deviations. Our approach remains informative if we allow for rounding and can be combined with restrictions on expectations to provide narrower bounds.