Message/Author 

Rich Mohn posted on Wednesday, September 17, 2008  2:11 pm



I'm running my first MTMM in Mplus and decided to also do it in AMOS where I've done them before. I got very different results, i.e., not just little difference I've seen before, but even directional changes. Wondered if anyone sees anything glaring in the code below that I may be missing, or has any experience to share. All continuous variables, using ML estimator, and no missing data. Thanks much. sdab BY sapt1* s20log_1 dshi_lg1 sbq_lg1; odab BY tapt1* t20log_1 lha_agg bp_ph_ag; behav BY sapt1* s20log_1 tapt1 t20log_1; s_r BY bp_ph_ag* lha_agg sbq_lg1 dshi_lg1; sdab@1 odab@1 behav@1 s_r@1; sdab WITH odab behav@0 s_r@0; behav WITH s_r odab@0; s_r WITH odab@0; 


If you have the same model, estimator, and data, you will obtain the same results. If you want us to tell you the difference between the two analyses, send your AMOS and Mplus outputs and your license number to support@statmodel.com. 


Dear M & M, I am running my model of MTMM in Mplus, as below: variable: names are id person ex1ex6 ag1ag6 co1co6 es1es6 op1op6; usevariables are ex1ex6 ag1ag6 co1co6 es1es6 op1op6; ANALYSIS: ITERATIONS = 5000; model:ex by ex1ex6; ag by ag1ag6; co by co1co6; es by es1es6; op by op1op6; gen by ex1 ag1 co1 es1 op1; F by ex2 ag2 co2 es2 op2; Pa by ex3 ag3 co3 es3 op3; YR by ex4 ag4 co4 es4 op4; Prof by ex5 ag5 co5 es5 op5; Str by ex6 ag6 co6 es6 op6; OUTPUT: STANDARDIZED; And I got a reply of 'THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING PARAMETER 1.' I have tried a much simpler model: variable: names are id ex1ex6 ag1ag6 co1co6 es1es6 op1op6; usevariables are ex1ex6; ANALYSIS: ITERATIONS = 500; model:ex by ex1ex6; OUTPUT: STANDARDIZED; But I still get the same warning, with the standard errors of the estimates not being computed. The data is good, I mean, the correlations within the variables are good enough to fulfill a model. So I wonder in which step there would be a mistake I made. Thank you. 


Say that I'm running a MTMM using a correlated uniqueness approach and I have ~200 indicators with ~50 per method for four methods. Is there a shortcut to writing the code for this analysis or do I need to type out all of the approx. 196 "WITH" command lines to cover all of the combinations of pairs of indicators? 


You can write, for example, y1y196 WITH y1y196; Duplicate WITH statements are ignored. 


That's very intuitive. Thank you. 


When fitting a MTMM via CFA, is it necessary to first evaluate whether the measurement model fits the data well (using a separate CFA)? Is it problematic to fit a MTMM via CFA using a measurement model that violates many model assumptions and does not fit the data well? 


I would think it would be problematic. You might want to broach this subject on SEMNET or another general discussion forum. 


Thanks! 

Back to top 