Mixed-effects Tobit model

[butter30]

Hi

Wondering if anyone knows how to run mixed-effect Tobit model?

Normally in cross-sectional data would use "proc qlim" but not sure if I can use this code in longitudinal data. My data consist of 4 fixed effects {x1, x2, x3, x4},  random effect intercept z0 and time varying random effect  z1 .

Example of code using proc mixed if this helps.

proc mixed data=data method=ml;

class gender treatment;

model y= x1 x2 x3 x4 /s;

RANDOM intercept time /s;

run;

If possible I would like to avoid Bayesian as my knowledge is minimal in that field.

Many thank,

[SteveDenham]

There may be some R code out there that could be translated, but this is an area I would really like to see developed, as timed measurements of biomedical parameters or ecological parameters often fall below a lower limit of quantitation, and it would be nice to see a method developed that doesn't involve biased imputation of these values.

Steve Denham

[lvm]

One approach is to use NLMIXED. Here are some code snippets:

SAS Code Fragments: random effect tobit model in nlmixed

This will require more work on your part, but this may get you started.

[SteveDenham]

Read an article today about the history of computing in the Soviet Union, and there is an anecdote:

As one joke tells it, Brezhnev is gifted with the latest in artificial intelligence, so he asks it “When will we have built communism?” The computer responds, “In 17 miles.” Brezhnev thinks, “There must be something wrong,” and repeats the question. The computer again replies, “In 17 miles.” Angered by the incomprehensible reply, Brezhnev orders a technician to investigate the machine. “Everything is correct,” replies the technician after some time. “You said it yourself: Every five-year plan is one step toward communism.”

I think those code snippets are the equivalent of a five year plan, one step on a journey of about 17 miles toward a repeated measures random effects tobit with a specifiable covariance structure.

Steve Denham