Message/Author 

Joseph posted on Sunday, December 15, 2013  7:44 am



Hello Dr.Muthen, I am analyzing a model with 4 moderator variables (continuous). I want to study gender differences in the model as well. Can I prepare two data files one for male and another for female and analyze the model separately for male and female? or should I prepare 3 data files  one for male , one for female and another for combined(male+female) and analyze the model and the compare 3 results or is it mandatory to use TYPE = MIXTURE RANDOM with KNOWNCLASS option 


You should have one data file with gender as one of the variables. You can use the USEOBSERVATIONS option to select males or females for separate analyses. 

Joseph posted on Sunday, December 15, 2013  4:53 pm



Thank You, Dr Muthen. So if I understand you correctly, it is okay to have separate analysis and it is not mandatory to use TYPE = MIXTURE RANDOM with KNOWNCLASS option 


If you don't want a multiple group analysis, it is fine to do them separately. I don't think I get your question. 

Joseph posted on Sunday, December 15, 2013  10:23 pm



Thank you, Dr.Muthen.. Because of the complexity involved (with four moderator variables), I want to analyze the data for female and male separately rather than doing a multi group analysis. So the model will be independently tested for female and male and model fit will also be reported separately. My worry is whether I should prove that there is no measurement invariance before I could do the analysis for female and male separately. 


You don't need to establish measurement invariance if you are not going to compare the groups. 

Joseph posted on Monday, December 16, 2013  8:52 pm



Thanks Dr.Muthen. I don't statistically compare both the groups. But I hope its okay to compare the final model results. For instance my model has 4 moderators  when the data was separately analyzed I found that group 1(male) has 3 moderators that are significant and group 2 (female) has 2 moderators that are significant.(significance was tested by ztest third column in the output). I believe its okay to cite this as a difference between these groups. Kindly confirm.Thanks. 


If you have not established that the constructs represent the same thing in each group, I don't believe any comparison, statistical or not, is valid. 

Joseph posted on Wednesday, December 18, 2013  6:25 am



Thanks.. So is it fine if I do the multi group CFA to establish measurement invariance and then analyze the groups separately to establish the moderator relations (TYPE = RANDOM using XWITH option) and then compare the model results . Or should I mandatorily do TYPE = MIXTURE RANDOM with KNOWNCLASS option 


If you want to compare the groups, I would establish measurement invariance using a multiple group analysis. Once established, I would continue the analysis in a multiple group setting with measurement invariance equalities in place. 

Joseph posted on Wednesday, January 01, 2014  8:35 am



Thank You,Dr.Muthen. If I understand you correctly I should constrain factor loading, loading intercept and residual variance to be equal across groups as I continue the analysis after CFA( and also after establishing MI) to examine predictor relations ( eg: M on A B C). And in Mplus mixture modeling, Intercepts and residual variances are constrained equal across classes by default so there isn't anything extra that needs to be done there.I need to take care of factor loading  by mentioning it only in %OVERALL% it will be constrained equal across classes. Also as intercepts are constrained to be equal across groups I must fix the means to zero (eg: [A@0] Is my understanding correct? 


If you are using KNOWNCLASS, all parameters are held equal across class except factor means which are zero in the last class and free in the other classes. You would need to relax these equalities and then constrain them in steps to test for measurement invariance. See the Topic 1 course handout under multiple group analysis to see how this is done. 

Joseph posted on Friday, January 10, 2014  10:49 am



Thanks a lot for your help. May I ask ,one question on moderation( though unrelated to the topic here) To check a model , where an independent variable's relation on dependent variable is moderated by two variables. Which among the below options would be a correct choice : (BY statements are skipped) Option 1 : Model : AxB  A XWITH B; AxC  A XWITH C; Y on A B AxB AxC; Option 2 : Model : !(Choice of B or C first will be a concern in this) AxB  A XWITH B; AxBxC  AxB XWITH C; Y on A B AxB AxBxC; Option 3 Model: AxB  A XWITH B; AxC  A XWITH C; AxBxC  AxB XWITH AxC; Y ON A B AxB AxC AxBxC; Or are all the options correct depending on situations.. 

Joseph posted on Friday, January 10, 2014  11:00 am



Sorry.. Forgot to include C in ON statement : To check a model , where an independent variable's relation on dependent variable is moderated by two variables. Which among the below options would be a correct choice : (BY statements are skipped) Option 1 : Model : AxB  A XWITH B; AxC  A XWITH C; Y on A B C AxB AxC; Option 2 : Model : !(Choice of B or C first will be a concern in this) AxB  A XWITH B; AxBxC  AxB XWITH C; Y on A B C AxB AxBxC; Option 3 Model: AxB  A XWITH B; AxC  A XWITH C; AxBxC  AxB XWITH AxC; Y ON A B C AxB AxC AxBxC; Or are all the options correct depending on situations.. 


I would keep it simple and go with option 1, unless there was theory or evidence to be more complex. 

Joseph posted on Friday, January 10, 2014  4:46 pm



Thank a lot , Dr.Muthen... 

Back to top 