All entries for Thursday 11 August 2005
August 11, 2005
(1) Example 4.1 of Wooldridge (2002) shows that E(y|x) can be
m(x,theta), a nonlinear function. Also, the model (distribution) of Y|X can be a member of the linear exponential family yielding the quasi log-likelihood.
(2) In this light, there seems to be four components: distribution of Y|X, model for E(Y|X), distribution of X and a model of R.
(3) In the case where we have to estimate selection probabilities, Y and X are required to be observed only when S =1. Thus, X can be missing with Y???
Normally, MAR => R depends on X which is fully observed. In this case, it is still MAR but R depends on the completely observed Z and X can be missing ???