02-27-2016 05:58 PM
I like to estimate prevalence ratio from a complex survey design using SAS, but it appears that the SURVEY procedure do not support the link function that I need.
I will like to use either Poison distribution or log-binomial model, but Surveylogistic support only odds ratio, and the journal does not want odds ratio.
Doe any one know how to proceed with this?
02-28-2016 09:09 AM
02-29-2016 05:58 AM
If you dont have any other covariates you want to adjust for, then you can use PROC SURVEYFREQ and the option "OR" in the table statement will provide estimates and confidence intervals for the oddsratio as will as the risk ratio (prevalence).
In case have some other covariates then I dont think there is any procedure that can give the estimates directly. Perhaps PROC SURVEYPHREG may help you. I guess so because of the relation between Cox-regression and Poisson regression (conditioning on the intercept or a class variable in Poisson regression will give a likelihoodfunction similar to Breslows likelihood). Though, I dont know if this relation exist still exist when accounting for finite samplesize.
06-20-2017 04:54 PM
Thank you SWEETSAS for the question and JacobSimonsen for some direction. SWEETSAS, what did you end up doing? I have the same problem today and I don't want to learn Stata. I would like to compute adjusted prevalence ratios using a complex survey dataset that requires the use of replicate weights and balanced repeated replication. I'd like to use the Poisson distribution.
08-09-2017 03:13 AM
I used SURVEYPHREG because of the relation between Cox's regression model and Poisson regression model. One merely needs to manipulate the data so that the procedure can interprete time and the censor indicator correctly.
08-09-2017 10:39 AM
This should also be possible by using the STORE statement to save the fitted model from PROC SURVEYLOGISTIC, and then using the NLEstimate macro to estimate the risk ratio since it is just a nonlinear combination of the model parameters.