 Predicted value    Message/Author  Sanjoy Bhattacharjee posted on Wednesday, May 28, 2008 - 12:27 pm
Prof. Muthen

I have the following model

Y on X1, z1, z2, z2;
z1 on X2;
z2 on X3;
z3 on X4;

We are proposing an alternate to this model in the following way

Y1 on X1, eta;
eta by z1, z2, z3;
eta on X5;

X�s are vector and have enough unique elements. Y is binary, Z�s are ordered (3 categories) categorical. However, I actually am modeling Y* and z*s rather than Y and z�s and hence I am using your WLSMV estimator (based on your articles appeared in Journal of Econometrics, �83 and Psychometrika, �84, �87, �95 and �97)

I need to calculate Y* at the mean values of its explanatory variables.

I am using MPlus 4.1 and my questions to you are;

could you kindly suggest me -

Q1. how can I calculate predicted value for z*�s which I can later plug in Y* equation. I am NOT interested in, for example, Pr(z1=1) or pr(z1=2) or Pr(z1=3) and so on

Q2. how can I calculate predicted mean value for �eta�? Should it not be zero, I mean the mean value of �eta�?

Thanks and regards  Sanjoy Bhattacharjee posted on Wednesday, May 28, 2008 - 9:34 pm
Prof. Muthen,

In the context of the above question, should I consider the two thresholds associated with each Z's to estimate the respective predicted Z*s. The Z's and the respective Z*s are related through ordered probit.

Thanks and regards    Topics | Tree View | Search | Help/Instructions | Program Credits Administration