

Logistic regression using clustering 

Message/Author 

mplususer posted on Friday, January 03, 2014  6:19 pm



(My apologies if this has already been asked.) I am using MPlus 4.0. I would like to conduct a logistic regression analysis, and my data are clustered at the family level. The following logistic regression (without accounting for clustering) runs with no problems: Title: Logistic regression for suicide attempts  MDD onset/recurrence Data: File is "file"; Variable: Names are IDYRFAM ID IDSEX MDDADOL MDDREC MDDGRP1 MDDGRP2 MDDGRP3 MDDGRP4 MDDGRP F3SUIC F3PABU F3SABU; Usevar are IDSEX MDDADOL MDDREC F3SUIC; Categorical are F3SUIC; Missing are all (99); Analysis: ESTIMATOR = ML; Model: F3SUIC on IDSEX MDDADOL MDDREC; Output: CINTERVAL; However, when I include CLUSTER = IDYRFAM, I am told I must use TYPE = COMPLEX. When I add TYPE = COMPLEX, I am told that ESTIMATOR = ML cannot be used with TYPE = COMPLEX. My understanding is that the command ESTIMATOR = ML is what determines that this analysis is a logistic regression. Is there anyway to run a logistic regression with clustered data in MPlus 4.0? If not, can one do so in a later version of MPlus? Thank you very much for your help. 


Try ESTIMATOR=MLR; 

mplususer posted on Saturday, January 04, 2014  7:38 am



That worked, thank you very much for your reply. I am now running the following: Title: Logistic regression for suicide attempts  MDD onset/recurrence Data: File is "file"; Variable: Names are IDYRFAM ID IDSEX MDDADOL MDDREC MDDGRP1 MDDGRP2 MDDGRP3 MDDGRP4 MDDGRP F3SUIC F3PABU F3SABU; Usevar are IDYRFAM IDSEX MDDADOL MDDREC F3SUIC; Categorical are F3SUIC; CLUSTER = IDYRFAM; Missing are all (99); Analysis: ESTIMATOR = MLR; TYPE = COMPLEX; Model: F3SUIC on IDSEX MDDADOL MDDREC; Output: CINTERVAL; I was hoping to ask one more question. It looks like later versions of MPlus give twotailed pvalues for the estimates. Is it possible to request pvalues in MPlus 4.0? Or is there a way for me to calculate the pvalues for the estimates with the information given? Thanks again for your help. 


No, there are no pvalues in Version 4. The ratio of the parameter estimate to the standard error is a zvalue in large samples. You can look it up in a ztable to getthe pvalue. 

mplususer posted on Sunday, January 05, 2014  6:49 am



Of course. Thanks again! 

Back to top 

