please forgive me for bothering you: Before testing my multilevel hypotheses I first tried to run a "twolevel basic" analysis to check if my data gets read correctly - and somehow I keep failing already. I made sure that I do not have any blanks in my data file, all decimal points are dots instead of commas, all missings are coded as -99. The data file contains only numerical values. All command lines in my input file are shorter than 90 characters. But weirdly, in MPlus I get a different number (120) and size (between 1-6) of clusters than in SPSS (165 clusters with a size between 3-6). Just for your info, my clusters are persons and each person provided data on 3-6 days. The results seem to be the same for a dat.-data file and a csv.-data file. On top of that I get the following, very scary error message: "A MATRIX COULD NOT BE INVERTED DURING THE H1 MODEL ESTIMATION. THE ESTIMATED WITHIN COVARIANCE MATRIX IS NOT POSITIVE DEFINITE AS IT SHOULD BE. COMPUTATION COULD NOT BE COMPLETED. PROBLEM INVOLVING VARIABLE SCD_T. THE PROBLEM MAY BE RESOLVED BY SETTING ALGORITHM=EM AND MCONVERGENCE TO A LARGE VALUE. THE H1 MODEL ESTIMATION DID NOT CONVERGE. THE SAMPLE STATISTICS COULD NOT BE COMPUTED."
I would be immensely grateful if you could help me in figuring out what I´m doing wrong! And please forgive me if this is a very stupid question.
I am so very sorry, I just realized that the first part of my question was indeed stupid: The number and size of clusters differs because of the simple fact that I enabled listwise deletion in MPlus... I am very sorry! So I guess that my data file gets read correctly after all.
But sadly, there remains the problem detailed in the error message above. Background information: SCD_T is a within variable which varies within persons across six days, thus, in SPSS I could only look at the descriptive statistics and histogramms of the SCD_T variable of each day etc (SCD_T1,.., SCD_T6). But they look all quite unremarkable (in fact, even better than some other variables). The means and standard deviations do not vary much between the days (the mean is always about 30, the standard deviation about 12). The distributions look approximately normal, just slightly positively skewed.
Do you have any idea what is going on here and what I can do about this problem? I am very reluctant to just delete this variable - unfortunately it is my most important predictor...
I had a question regarding analysis of daily diary data using twolevel. I performed 2 analyses that I thought should produce the same result in daily diary data, but they don’t. In one analysis, I requested between-person correlations between my variables “e1” and “dtday” by adding the line “TYPE=TWOLEVEL BASIC” in the ANALYSIS section. In another analysis, I requested between- and within-person correlations among “e1” and “dtday” by using the following MODEL commands:
%WITHIN% e1 with dtday;
%BETWEEN% e1; dtday; e1 with dtday;
The problem is that the between-person correlation from the latter file is not equal to the between-person correlation from the former file (within-person correlations are also different across files, but to a lesser extent). Could you tell me why these values would differ? Thank you!
They shouldn't differ. Maybe you are comparing covariance and correlation. You should be getting a chi-square of zero to indicate that the model is the same as the h1 model. If it is not zero sharpen the convergence criterion. If you add output:samp you can do the comparison of results in one run. If you can't figure it out send your example and data to firstname.lastname@example.org