I have run a proc surveylogistic model using link=glogit on sas. The idea is to run a multinomial regression model.
However, whenever I run the model with it covariates, SAS provides an output with betas and corresponding ORs. My colleague has run the same model on STATA, and he got RRRs estimates. Does anyone know how to get RRRs from SAS outputs?
Thanks!
Hello Vitor. The RRR from Stata's -mlogit- command really is an odds ratio. See this note for an explanation.
- https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/files/Stata-mlogit-RRR.pdf
HTH.
Depends greatly on if you have any covariates in your model. Otherwisez conversion should fairly straightforward.
Hope this helps,
Matt
Dear Bruce and Matthew,
Thanks for your kind and thoughtful answers. Bruce’s link is very clear and better than everything else I’ve found online! But matthew’s comment arised a new question.
Bruce’s models in the PDF documents do not usa covariates, as far as I understood. But mine does! So, matthew has a point. How do I get RRRs using a proc surveylogistic model with link=glogit in SAS?
THANKS!
Hello Vitor. To test the hypothesis that presence of covariates changes things, I generated another example comparing RRR from Stata to Exp(B) from SPSS in a model that has covariates. Here is the Stata code, followed by the SPSS code:
// Example from -help mlogit- in Stata.
clear *
webuse sysdsn1
// Fit multinomial logistic regression model
// Make site 3 the reference category to match what SPSS does.
mlogit insure age male nonwhite ib3.site
// Replay, reporting relative-risk ratios
mlogit, rrr
// Export results to Excel.
putexcel set "C:\Temp\Stata_mlogit_with_covariates.xlsx" //, replace
// Save file to disk so that I can import it with SPSS
save "C:\Temp\sysdsn1.dta"
// SPSS code below!
GET STATA FILE='C:\Temp\sysdsn1.dta'.
* Stata mlogit command: mlogit insure age male nonwhite i.site
.
NOMREG insure (BASE=FIRST ORDER=ASCENDING) BY site WITH age male nonwhite
/PRINT=PARAMETER SUMMARY LRT CPS STEP MFI.
* Compare Exp(B) values from SPSS to RRR values from Stata.
* CONCLUSION: They are the same.
As the final comment line says, the RRR values from Stata and the Exp(B) values from SPSS are the same. I do not have SAS, so cannot test what it does. But I would be shocked if its Exp(B) values did not match what I get with Stata and SPSS.
HTH.
Hi again, Vitor. I just noticed that you are talking about proc surveylogistic in SAS. My examples thus far have not used survey weights. In Stata, one does that by including the -svy- prefix command before -mlogit-. In SPSS, one would have to use the CSLOGISTIC command, I believe. From the help file:
"CSLOGISTIC performs logistic regression analysis on a binary or multinomial dependent variable using the generalized link function for samples that are drawn by complex sampling methods."
Source: https://www.ibm.com/support/knowledgecenter/en/SSLVMB_25.0.0/statistics_reference_project_ddita/spss/complex_samples/syn_cslogistic.html
Hello, Bruce. Thanks again for your help. Yes, I do use proc surveylogistic. Again – I’m pretty confused with this whole relative risk =\= risk ratio idea -, does using a survey procedure make SAS ORs different from the Stata RRR, like in bruce’s PDF file?
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running