Dear All, I have a dataset, wherein the outcome variable is a four level nominal variable and the predictor variables are both categorical and continuous with varying levels of missing ness. I am using SAS proc MI to create 5 multiply imputed datasets, and then use PROC LOGISTIC with the link=glogit option to fit models to the generalized logits. For each imputed dataset I am able to obtain an estimate corresponding to each generalized logit(For the intercept and all predictor variables).
However when I use PROC MIANALYZE to combine the parameter estimates from the 5 imputed datasets, instead of getting a combined estimate for each logit I get just one single estimate i.e. In the individual imputed datasets I get parameters estimates for the intercept corresponding to the 3 logits (since the outcome variable is 4 level), however proc MIANALYZE gives me a single estimate just one estimate for the intercept. I am not sure if this is because PROC MIANALYZE does not support logistic regression for a nominal variable or whether I am doing something wrong. If any of you has come across this issue in SAS or can advise an alternative way of doing this, I'd appreciate it. Thanks, Jasmine -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060118/40399b73/attachment.htm From jmathias.stat <@t> gmail.com Wed Jan 18 20:44:27 2006 From: jmathias.stat <@t> gmail.com (Jasmine Mathias) Date: Wed Jan 18 21:48:19 2006 Subject: [Impute] How to combine parameters estimates from logistic regression with nominal outcome variable using SAS v 9.1.3? Message-ID: <974a452a0601181844s37321e0cja7a41806123bb...@mail.gmail.com> Dear All, I have a dataset, wherein the outcome variable is a four level nominal variable and the predictor variables are both categorical and continuous with varying levels of missing ness. I am using SAS proc MI to create 5 multiply imputed datasets, and then use PROC LOGISTIC with the link=glogit option to fit models to the generalized logits. For each imputed dataset I am able to obtain an estimate corresponding to each generalized logit(For the intercept and all predictor variables). However when I use PROC MIANALYZE to combine the parameter estimates from the 5 imputed datasets, instead of getting a combined estimate for each logit I get just one single estimate i.e. In the individual imputed datasets I get parameters estimates for the intercept corresponding to the 3 logits (since the outcome variable is 4 level), however proc MIANALYZE gives me a single estimate just one estimate for the intercept. I am not sure if this is because PROC MIANALYZE does not support logistic regression for a nominal variable or whether I am doing something wrong. If any of you has come across this issue in SAS or can advise an alternative way of doing this, I'd appreciate it. Thanks, Jasmine -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060118/d60d51a3/attachment.htm From acock <@t> comcast.net Wed Jan 18 22:42:16 2006 From: acock <@t> comcast.net (Alan Acock) Date: Wed Jan 18 22:42:20 2006 Subject: [Impute] How to combine parameters estimates from logisticregression with nominal outcome variable using SAS v 9.1.3? In-Reply-To: <974a452a0601181908i686058ccn922b139ce32b3...@mail.gmail.com> Message-ID: <002601c61cb2$b96ad2e0$6a01a...@alan1204> If you have access to Stata, there is a user written command called ice that can do this. It is an implementation of MICE for Splus. It uses a variety of estimation models in the imputation process, regression, logistic regression, multinomial logistic regression, etc. and works with a wide variety of regression related Stata commands to automatically pool the parameter estimates and standard errors (regress, logit, clogit, mlogit, count models, negative binomial models, etc.). It is extremely easy to implement. If you have Stata run the command findit ice. Alan Acock _____ From: impute-boun...@lists.utsouthwestern.edu [mailto:impute-boun...@lists.utsouthwestern.edu] On Behalf Of Jasmine Mathias Sent: Wednesday, January 18, 2006 7:08 PM To: imp...@lists.utsouthwestern.edu Subject: [Impute] How to combine parameters estimates from logisticregression with nominal outcome variable using SAS v 9.1.3? Dear All, I have a dataset, wherein the outcome variable is a four level nominal variable and the predictor variables are both categorical and continuous with varying levels of missing ness. I am using SAS proc MI to create 5 multiply imputed datasets, and then use PROC LOGISTIC with the link=glogit option to fit models to the generalized logits. For each imputed dataset I am able to obtain an estimate corresponding to each generalized logit(For the intercept and all predictor variables). However when I use PROC MIANALYZE to combine the parameter estimates from the 5 imputed datasets, instead of getting a combined estimate for each logit I get just one single estimate i.e. In the individual imputed datasets I get parameters estimates for the intercept corresponding to the 3 logits (since the outcome variable is 4 level), however proc MIANALYZE gives me a single estimate just one estimate for the intercept. I am not sure if this is because PROC MIANALYZE does not support logistic regression for a nominal variable or whether I am doing something wrong. If any of you has come across this issue in SAS or can advise an alternative way of doing this, I'd appreciate it. Thanks, Jasmine -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060118/e6142dad/attachment.htm From allison <@t> soc.upenn.edu Thu Jan 19 07:10:01 2006 From: allison <@t> soc.upenn.edu (Paul Allison) Date: Thu Jan 19 07:10:12 2006 Subject: [Impute] How to combine parameters estimates from logisticregression with nominal outcome variable using SAS v 9.1.3? In-Reply-To: <974a452a0601181844s37321e0cja7a41806123bb...@mail.gmail.com> Message-ID: <200601191310.k0jda1b5014...@facstaff.sas.upenn.edu> Jasmine: This problem is easily solved. In PROC MIANALYZE on the VAR statement (now MODELEFFECTS statement), you need to give a different name for each coefficient for each variable. These names can be found by looking at the variable names on the OUTEST data set produced by PROC LOGISTIC. For example, suppose the dependent variable has three levels with values of 1, 2 and 3. Suppose the independent variables are X and Z. Then the statement would be MODELEFFECTS intercept_1 x_1 z_1 intercept_2 x_2 z_2; ----------------------------------------------------------------- Paul D. Allison, Professor and Chair Department of Sociology University of Pennsylvania 3718 Locust Walk Philadelphia, PA 19104-6299 215-898-6712, 215-898-6717 215-573-2081 (fax) http://www.ssc.upenn.edu/~allison -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060119/62dbf233/attachment.htm
