A life history can be regarded as a random process that evolves with age th
rough various states of health before terminating with absorption into the
state of death. Health expectancies are the occupation times of the non-abs
orbing states and their estimation is of interest. A continuing major probl
em has been the lack of satisfactory longitudinal data on which to base est
imates and as a result standard inferential techniques may not be relevant.
Supposing only cross-sectional data available, we propose a method that is
generally applicable and first estimates a logistic parametrization of the
probabilities of the various states. A large sample approximation is obtai
ned for the distribution of age specific log (odds). Parameters are estimat
ed by weighted least squares, and this in turn leads to estimates of cohort
health expectancies. A result of Liang and Zeger is used to find standard
errors. The method is illustrated by application to Australian data from th
e health surveys of 1981, 1988 and 1993. Copyright (C) 2001 John Wiley & So
ns, Ltd.