|
|
Can Mplus mimic a HLM growth curve tr... |
|
Message/Author |
|
Jim Akers posted on Monday, March 14, 2005 - 2:40 pm
|
|
|
Hi, In modeling language development of infants and toddlers, it makes sense to apply the approximation that a child’s vocabulary = 0 at age 12 months. This strategy can be applied to HLM 5 growth models (e.g., Huttenlocher et al, 1991), and naturally provides a strong anchor value for our tests of growth curves. Of course, trying the same thing in Mplus merely results in that no-variance-stupid! message. Is there a parallel to the HLM "workaround" strategy that could work for Mplus LGMs? Thank you, Jim |
|
|
You can use the VARIANCES = NOCHECK option of the DATA command causes Mplus to not check for zero variance. You can try that. |
|
Jim Akers posted on Tuesday, March 15, 2005 - 7:51 am
|
|
|
Dear Linda. Nope, using "variances=nocheck" did not do the trick. I included the model and error messages below. Let me note two things: 1. There is missing data, with up to 84% missing on one covariance. 2. However, without 0 scores included, the model converges with excellent fit statistics. I am sure no expert, but I have to admit that if latent var models ultimately depend on covariance matrices, the "preponderance of zeros," if I can borrow the term, seems to be an intractable issue for this kind of model. Thanks for your help, Jim MODEL: i s | tpr11_12@0 tpr11_24* tpr11_36* tpr11_48* tpr11_54@1; SAMPLE STATISTICS ESTIMATED SAMPLE STATISTICS THE MISSING DATA EM ALGORITHM FOR THE H1 MODEL HAS NOT CONVERGED DUE TO ESTIMATED COVARIANCE MATRIX BEING NON POSITIVE DEFINITE. PROBLEM INVOLVING VARIABLE: 5 NO CONVERGENCE IN THE MISSING DATA ESTIMATION OF THE SAMPLE STATISTICS. THIS MAY BE DUE TO SPARSE DATA LEADING TO A SINGULAR COVARIANCE MATRIX ESTIMATE. |
|
|
You can set the model up as in HLM with the data long rather than wide, performing growth modeling as a two-level analysis. So you have a person id variable, one outcome variable, one time variable, and one between variable w varying across individuals. The input will include: CLUSTER = id; WITHIN = time; BETWEEN = w; TYPE = TWOLEVEL: MODEL: %WITHIN% s | y on time; %BETWEEN% y s ON w; |
|
Jim Akers posted on Saturday, March 19, 2005 - 1:48 pm
|
|
|
Dear Linda, Wish I could get back to you as fast as you do to me. I must be making an obvious input error, as the same messages (below) keep coming up whether or not I filter out cases wtih constant age = 12 (and score=0). Would you please check my input further down: (also note: HLM 5 will produce expected results with the same data) Thank you, Jim ******************************************** THE ESTIMATED BETWEEN COVARIANCE MATRIX IS NOT POSITIVE DEFINITE AS IT SHOULD BE. COMPUTATION COULD NOT BE COMPLETED. PROBLEM INVOLVING VARIABLE SCORE. THE CORRELATION BETWEEN SITE AND SCORE IS -1.000 THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ERROR IN THE COMPUTATION. CHANGE YOUR MODEL AND/OR STARTING VALUES. THE ESTIMATED BETWEEN COVARIANCE MATRIX IS NOT POSITIVE DEFINITE AS IT SHOULD BE. COMPUTATION COULD NOT BE COMPLETED. PROBLEM INVOLVING VARIABLE SCORE. THE CORRELATION BETWEEN SITE AND SCORE IS -1.000 ************************************* DATA: FILE IS 'VERTICAL MPLUS MULTILEVEL LGM.dat'; VARIANCES = NOCHECK; VARIABLE: NAMES ARE ID AGE SCORE SITE; USEVARIABLES ID AGE SCORE SITE; !SITE IS CODED 0 AND 1 !USEOBSERVATIONS AGE GT 0 ; MISSING ARE ALL (9999); CLUSTER = ID; WITHIN = AGE; BETWEEN = SITE; ANALYSIS: TYPE = TWOLEVEL RANDOM; MODEL: %WITHIN% GROWTH | SCORE on AGE; %BETWEEN% SCORE GROWTH ON SITE; |
|
bmuthen posted on Saturday, March 19, 2005 - 4:27 pm
|
|
|
Please send your input, output, and data to support@statmodel.com. |
|
Back to top |
|
|