Nonlinear regression model with complex survey design

[alan1]

Hi to anyone that can help

I have complex weighted survey data and I find that the data best fit a sigmoid curve, like a dose response curve. I am using the complexity of a doctor visit to predict the time that docs spend with patients.

In simplest form, for a linear regression I would use:

Proc surveyreg;

model time=complexity/ clparm adjrsq solution;

weight visitweight;

run;

But, to model a sigmoid response, how should I write the code?

Thanks in advance?

Alan

[SteveDenham]

Hi Alan,

I would start by looking through Example 45.18 Weighted Multilevel Model for Survey Data.  This would take into account the survey nature of what is going on.

For the response, I would consider rescaling so that the response fell on the open interval (0,1).  Then I would apply a logistic link using a beta distribution.

The alternative is a lot of coding in NLMIXED.  You may want to search the stackoverflow and SAS-L archives for approaches.

Steve Denham

[alan1]

Thanks!!

Alan