Y2, Y3, Y4 = g(eta1) (g is a vector function) eta1 = h(X2) Y1 = f(X1, eta1)
Where eta1 is a continuous latent variable and Y2, Y3, Y4 are ordinal indicators capturing eta1. Y1 is BINARY and the variable of primary interest. X’s are exogenous variables.
We have sample selection issue. I was wondering whether it would be OK to use IMR (inverse mills ratio) as one of the explanatory variable in Y1 equation and then use WLSMV. I am calculating IMR from the selection model; Z = z(X4)
Could you kindly suggest me some related references?
I can think of 2 approaches that could be useful. One draws on UG ex3.9. The other draws on UG ex5.3 with CONSTRAINT = x, where the heteroscedasticity is a function of the covariate x. But you would have to explore.
Thanks for the advise. I had tried this but the saved file contained no data. Please suggest. Here is the command.
DATA: File is C:\mplus\batch1.dat Variable: names are ...(I have all the variables in the dataset written here including w1 which is used in useobservation command)...; Useobservation = w1 EQ 1; Savedata: File is C:\mplus\w1;
I got the file out of the run but there is no data in the file. It is supposed to be about 600 observations here. Thanks.
I am interested in using the NOBSERVATIONS option to select a subset (N) of a simulated dataset. The manual indicates that the first N cases in the dataset would be selected. Is it possible to select a random sample of cases using the NOBSERVATIONS option?
If all else fails I can create randomized groups the dataset and use the USEOBSERVATIONS option, but this does not seem any more appealing (to me) than just using the first N cases (in this case).