Hi all,

Does anyone know if it is OK to use an ordered probit model to predict a latent variable, e.g. self-perceived health; and then use the linear predictions as a right-hand side variable in another ordered probit to predict a different latent variable, e.g. life satisfaction. The right-hand side variables in the first ordered probit model do not appear in the second model and vice-versa.

More specifically, I intended to (1) regress self-perceived health (poor-fair-good-very good or excellent) on a set of health-related variables such as disability and co-morbidities using an ordered probit, (2) predict the linear health index from this model, (3) regress life satisfaction (in a scale of 0-10) on that linear health index and other non-health-related variables such as income, demographic characteristics etc., also using an ordered probit, and (4) predict the linear life satisfaction index from this model, which I need for some further analysis.

References are welcomed. Thanks in advance!