Hyun Shin posted on Wednesday, March 12, 2014 - 9:59 pm
Greetings! I ran a CFA model with 42 indicators for 2 latent variables (n=140). The indicators are categorical (i.e., 0, 1). Below is my MPlus input command:
Variable: Names are r1-r21 p1-p21; Categorical are r1-r21 p1-p21; MODEL: f1 BY r1-r21; f2 BY p1-p21; f1@1f2@1;
I got a warning message as follows: WARNING: THE BIVARIATE TABLE OF R11 AND R5 HAS AN EMPTY CELL. WARNING: THE BIVARIATE TABLE OF R13 AND R11 HAS AN EMPTY CELL. WARNING: THE BIVARIATE TABLE OF R14 AND R5 HAS AN EMPTY CELL. WARNING: THE BIVARIATE TABLE OF R19 AND R14 HAS AN EMPTY CELL. WARNING: THE BIVARIATE TABLE OF R19 AND R18 HAS AN EMPTY CELL.
I double-checked my data and no cells are empty. I am wondering if this problem occurs due to the fact that all my variables are categorical (coded as 0 or 1) which may cause some problems in estimating covariance matrix. Can you please help?
Use the CROSSTABS option of the OUTPUT command to view the bivariate tables of your categorical items. This is what is referred to. When a bivariate table has an empty cell, this implies a correlation of plus or minus one. In this case, both variables cannot be used in the analysis because they are not statistically distinguishable.
Hyun Shin posted on Friday, March 14, 2014 - 11:54 pm
Thank you very much for your response. I checked the CROSSTABS and there was an empty cell, for example, for two items, R5 and R11. Specifically, there was no student who answered correctly for R5 AND didnít answer correctly for R11. But when I checked the correlation between R5 and R11, it was just 0.38, far from ďa correlation of plus or minus oneĒ as mentioned by you. I also checked other bivariate relationships and found similar results. So I guess multicollinearity is not an issue in my case. Is there any other sources of problem?
I also have another question. For the same codes mentioned above, I've got the following sign:
"WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE /RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES. CHECK THE TECH4 OUTPUT FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE F2."
Could you please let me know how to address this problem? Thank you very much!