Warning Message for EFA PreviousNext
Mplus Discussion > Exploratory Factor Analysis >
Message/Author
 Thomas A. Schmitt posted on Friday, May 16, 2008 - 3:42 pm
Hello:

Any idea why I would be getting this warning message below and what it means?

Best,

Tom
WARNING: THE SAMPLE CORRELATION OF SZOID1 AND PARND6 IS -0.998
DUE TO ONE OR MORE ZERO CELLS IN THEIR BIVARIATE TABLE.
INFORMATION FROM THESE VARIABLES CAN BE USED TO CREATE ONE NEW VARIABLE.
 Linda K. Muthen posted on Friday, May 16, 2008 - 5:10 pm
This means that the bivariate frequency table for SZOID1 AND PARND6 has an empty cell which implies a correlation of one. Both of these variables cannot be used in the analysis. You can combine them into one variable with three categories, for example, observations with 0 0 can be assigned 0 on the new variable, those with 0 1 can be assigned 1 and those with 1 1 can be assigned 2.
 Thomas A. Schmitt posted on Thursday, May 29, 2008 - 1:01 pm
I've been trying to figure out as to how it comes up with this correlation. Because as far as I can tell the correlation is not -0.998 between these two variables.
 Linda K. Muthen posted on Thursday, May 29, 2008 - 1:25 pm
The correlation referred to is the tetrachoric correlation. I think you are referring to the Pearson product-moment correlation.
 Thomas A. Schmitt posted on Thursday, May 29, 2008 - 2:26 pm
Thank you! I thought I was losing my mind . . . which I guess I was. :-)
 Thomas A. Schmitt posted on Wednesday, June 04, 2008 - 8:16 pm
Hello Linda:

I've been continuing to work on this issue and better understand it. I was also given the following advice:

"But how's about you try the model again, with the logit (with the ML or MLR estimator) rather than the probit link function (with the least squares family) for the regressions of the items on the factors? Correlations (moments) aren't used for this, so maybe the same problem wouldn't arise."

My understanding is that the logit and probit link is used with ML estimation in Mplus. Can this also be done with the least squares family?

Best,

Tom
 Linda K. Muthen posted on Wednesday, June 04, 2008 - 8:46 pm
Maximum likelihood has both a logit and probit link. Weighted least squares has only probit.
 Thomas A. Schmitt posted on Thursday, June 05, 2008 - 6:44 pm
Thank you Linda! Would you have an idea why this simulation would not work. The error and program are below. The program is taken exactly from page 343 of the Mplus manual (Example 11.5). I have tried several modifications to the program to get it to work to no avail. I have the required computer parameters and all the latest updates for Mplus.

Best,

Tom

Mplus VERSION 5.1
MUTHEN & MUTHEN
06/05/2008 3:28 PM

INPUT INSTRUCTIONS

montecarlo:
names are y1-y6;
nobservations=500;
nreps = 10;
seed=45335;
model population:
f1 BY y1-y3*.7;
f2 BY y4-y6*.8;
f1-f2@1;
f1 WITH f2*.5;
y1-y3*.51;
y4-y6*.36;
ANALYSIS: TYPE = EFA 1 2;
OUTPUT: TECH9;

INPUT READING TERMINATED NORMALLY

*** FATAL ERROR
THERE IS NOT ENOUGH MEMORY SPACE TO RUN Mplus ON THE CURRENT
INPUT FILE. YOU CAN TRY TO FREE UP SOME MEMORY BY CLOSING OTHER
APPLICATIONS THAT ARE CURRENTLY RUNNING. NOTE THAT THE MODEL MAY
REQUIRE MORE MEMORY THAN ALLOWED BY THE OPERATING SYSTEM.
REFER TO SYSTEM REQUIREMENTS AT www.statmodel.com FOR MORE
INFORMATION ABOUT THIS LIMIT.
 Linda K. Muthen posted on Thursday, June 05, 2008 - 8:28 pm
TYPE=EFA is not available for Monte Carlo. It looks like the message to this effect has somehow been removed.
 Thomas A. Schmitt posted on Thursday, June 05, 2008 - 8:54 pm
That is unfortunate. Why was it removed? We had planned to do an EFA simulation study with the Mplus mechanism. Do you have any plans to put it back in?
 Linda K. Muthen posted on Thursday, June 05, 2008 - 9:05 pm
We did not remove EFA. We removed the error message. There are probelems with EFA and Monte Carlo because for some models the factors may not keep the same order across replications which renders the results averaged over replications meaniningless. A better approach is specifying EFA in the MODEL command a feature which was added in Version 5.1. The ordering issue has been addressed for this case although improvements will be made in the next update. See the Version 5.1 Language and Examples Addendum on the website with the user's guide.
 Thomas A. Schmitt posted on Sunday, June 08, 2008 - 10:50 pm
Thank you Linda!

Best,

Tom
 Ola Rostant posted on Wednesday, October 19, 2011 - 5:01 pm
Dear Linda,
I'm having the same problem as Thomas, the difference is that I have 45 binary variables and the error message says WARNING: THE BIVARIATE TABLE OF V25 AND V1 HAS AN EMPTY CELL.

WARNING: THE BIVARIATE TABLE OF V25 AND V8 HAS AN EMPTY CELL.

WARNING: THE BIVARIATE TABLE OF V25 AND V11 HAS AN EMPTY CELL.

Do you have some advice on how to approach combining these variables?

Thanks,
 Linda K. Muthen posted on Wednesday, October 19, 2011 - 5:25 pm
When a bivariate table has an empty cell, this implies a correlation of one. Both variables should not be used in the analysis.
 Maria Efthymiou posted on Sunday, December 04, 2011 - 10:32 am
Hello

I have a nominal variable in my dataset (independent variable). As there are 3 categories in the nominal variable, I have created 2 dummy variables(taking 0,1 values each) instead. I get this warning:

WARNING: THE SAMPLE CORRELATION OF SERV_DUM AND PROD_DUM IS -0.989
DUE TO ONE OR MORE ZERO CELLS IN THEIR BIVARIATE TABLE.
INFORMATION FROM THESE VARIABLES CAN BE USED TO CREATE ONE NEW VARIABLE.

This is because I have no values which are both 1. Is there a way to get away with this? By creating 2 dummy variables for a 3 category variable implies that I will not have cases with both a value of 1. Thanks in advance
 Maria Efthymiou posted on Sunday, December 04, 2011 - 2:37 pm
I have found a mistake I was doing so question answered.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: