https://communities.sas.com/t5/Statistical-Procedures/Log-binomial-regression-with-complex-survey-data-and-potential/m-p/748086#M36392 <P>I'm using complex survey data (PRAMS) to look at "risk" of a binary outcome using log binomial regression.&nbsp;</P> <P>&nbsp;</P> <P>I'm not too familiar with LBR, but I've read that the only real difference is the link function, so I considered setting up a PROC SURVEYLOGISTIC code block &amp; specifying LINK = LOG. These models are planned to be stratified (DOMAIN) by race/ethnicity.</P> <P>The only issue is that I'm concerned about potential multicollinearity. I used&nbsp;CORRB to check the standard, weighted logistic model and didn't find any high correlation between the main exposure variable and the other control variables, but is this accurate of a method to be able to ignore that potential issue?&nbsp;</P> <P>&nbsp;</P> <P>I guess I'm trying to figure out how (or if I can) use log binomial regression with complex survey data, while also trying to account for potential multicollinearity.&nbsp;</P> <P>&nbsp;</P> <P>The SAS resources all usually point back to using PROC GENMOD to estimate log binomial models, but you can't use that with complex survey data. And regression-type models make it a bit difficult to look at collinearity.</P> <P>&nbsp;</P> <P>Which battle should I choose?&nbsp;&nbsp;</P> Tue, 15 Jun 2021 10:45:20 GMT SAS93 2021-06-15T10:45:20Z https://communities.sas.com/t5/Statistical-Procedures/Log-binomial-regression-with-complex-survey-data-and-potential/m-p/748086#M36392 <P>I'm using complex survey data (PRAMS) to look at "risk" of a binary outcome using log binomial regression.&nbsp;</P> <P>&nbsp;</P> <P>I'm not too familiar with LBR, but I've read that the only real difference is the link function, so I considered setting up a PROC SURVEYLOGISTIC code block &amp; specifying LINK = LOG. These models are planned to be stratified (DOMAIN) by race/ethnicity.</P> <P>The only issue is that I'm concerned about potential multicollinearity. I used&nbsp;CORRB to check the standard, weighted logistic model and didn't find any high correlation between the main exposure variable and the other control variables, but is this accurate of a method to be able to ignore that potential issue?&nbsp;</P> <P>&nbsp;</P> <P>I guess I'm trying to figure out how (or if I can) use log binomial regression with complex survey data, while also trying to account for potential multicollinearity.&nbsp;</P> <P>&nbsp;</P> <P>The SAS resources all usually point back to using PROC GENMOD to estimate log binomial models, but you can't use that with complex survey data. And regression-type models make it a bit difficult to look at collinearity.</P> <P>&nbsp;</P> <P>Which battle should I choose?&nbsp;&nbsp;</P> Tue, 15 Jun 2021 10:45:20 GMT https://communities.sas.com/t5/Statistical-Procedures/Log-binomial-regression-with-complex-survey-data-and-potential/m-p/748086#M36392 SAS93 2021-06-15T10:45:20Z https://communities.sas.com/t5/Statistical-Procedures/Log-binomial-regression-with-complex-survey-data-and-potential/m-p/748107#M36394 <P>Well, collinearity doesn't depend on the dependent variable.&nbsp; Consequently, if you want to look at collinearity diagnostics you can use PROC REG.&nbsp; If you have categorical variables, you'll likely have to run GLMMOD first to set up all the dummy variables.</P> <P>&nbsp;</P> <P>If you have IML, you could manipulate your design matrix from SURVEYLOGISTIC to accomplish the eigenvalue based tests for collinearity.&nbsp; Check <a href="https://communities.sas.com/t5/user/viewprofilepage/user-id/13684">@Rick_SAS</a>&nbsp;'s blog for this topic.</P> <P>&nbsp;</P> <P>SteveDenham.</P> Tue, 15 Jun 2021 12:40:47 GMT https://communities.sas.com/t5/Statistical-Procedures/Log-binomial-regression-with-complex-survey-data-and-potential/m-p/748107#M36394 SteveDenham 2021-06-15T12:40:47Z