IRT Result with Binary Variables
Message/Author
 Yue Yin posted on Tuesday, April 03, 2018 - 1:43 pm
Hi, Dr. Muthén,

We ran an IRT analysis with the binary variables. After we ran it, the output illustrated the issues: WARNING: THE BIVARIATE TABLE OF ELNFR015 AND ELNFR009 HAS AN EMPTY CELL.

Below are the partial information about the covariance convergence:
Number of observations 3246

Number of dependent variables 52

the covariance convergence for the four items are .356, .195, .391, .554.

And all of the covariance convergence are low, and we had to change the minimum covariance convergence to .06 because we do not have that many.
And below are the percentage for the four items(like the item difficulty):
ELNFR015
Category 1 0.150 183.794
Category 2 0.850 1040.437
ELNFR009
Category 1 0.386 704.410
Category 2 0.614 1121.340
ELNFR115
Category 1 0.103 138.481
Category 2 0.897 1200.990
ELNFR102
Category 1 0.439 781.341
Category 2 0.561 999.914

I checked several discussions published before, that the empty cell is due to the correlation of the two variables of 1. But here I am wondering what kind of reason might result in this? Is because the low convergence, and percentage of the item(difficulty)are too high or too low?

Thank you so much!
 Bengt O. Muthen posted on Tuesday, April 03, 2018 - 2:41 pm
When you say "covariance convergence" perhaps you mean "covariance coverage". If that's the case, coverage of only .356, .195, .391, .554 is too low and missing data handling would rely too much on distributional assumptions. The empty cell message may be due to the high degree of missing.
 Yue Yin posted on Tuesday, April 03, 2018 - 2:50 pm
But the problem is that the covariance converage for other variables are also low just like the four values I listed before, how come other variables do not appear the empty cell message? Is
it due to the percentage?
 Bengt O. Muthen posted on Tuesday, April 03, 2018 - 3:29 pm
It also depends on the strength of relationship between the variables, not only on percentages and missingness.