Message/Author 

Sara Suzuki posted on Thursday, January 10, 2019  8:42 am



I am running a latent class analysis with complex survey data in Mplus. What does Mplus do to the complex survey information if I don't provide the Finite Population Correction Factor? I am asking because in Stata, without the Finite Population Correction Factor, clustering beyond the first stage is ignored. svyset IDSCHOOL [pweight=TOTWGT], strata(IDSTRAT)  IDCLASS Note: Stage 1 is sampled with replacement; further stages will be ignored for variance estimation. pweight: TOTWGT VCE: linearized Single unit: missing Strata 1: IDSTRAT SU 1: IDSCHOOL FPC 1: <zero> In Mplus, if I enter: WEIGHT = TOTWGT; CLUSTER = IDSCHOOL IDCLASS; STRATIFICATION = IDSTRAT; does it ignore IDCLASS? 


Correct. In fact, I don't think Mplus will allow you to specify two cluster variables unless you are using "type=complex twolevel mixture;" 

Sara Suzuki posted on Thursday, January 17, 2019  5:34 am



Hi Tihomir, Thank you for your response. Yes, I am using type=complex twolevel mixture. You indicated that if I do not give Mplus a finite population correction factor, it will ignore IDCLASS (classroom level clustering). I would like to make an assumption of infinite populations, therefore I do not wish to enter a finite population correction factor. Is there a way I can account for both school (IDSCHOOL) and classroom (IDCLASS) clustering without giving Mplus a finite population correction factor? 


When you are using type=complex twolevel mixture; both IDSCHOOL and IDCLASS are accounted for. IDCLASS is accounted for through the between level random effects. IDSCHOOL is accounted for through the robust ML estimation that accounts for the nonindependence of classrooms in the same school. 

Sara Suzuki posted on Saturday, January 19, 2019  1:42 pm



Thank you. In Mplus, is there a way to use IDSCHOOL and IDCLASS both for the robust ML estimation, in the way that Stata can account for multiple levels in the Taylor linearized variance estimator? 


Using finite population correction and Taylor linearized variance estimator Mplus uses only the first order correction. https://www.statmodel.com/download/SurveyJSM1.pdf page 2720 only the first term V1 is included but V2 is not included. As you can see there V2 contributes to the estimation only when fs is large, i.e., only when nearly all schools are sampled, which is not a realistic assumption. Once you assume infinite school population V2=0. 

Back to top 