I have a data set with about 900 participants and lots of missing data by design. When I start my analyses with constraint=items the process is stopped due to "The variance estimate of dimension 1 has become negative. Try increasing the number, or range, of the nodes.". Increasing the nodes to 250 leads to error message for many cases ("Error processing case #").
A colleague suggested to use constraint=cases insted and it worked well. But now I am a little bit concerned if something fundamental is wrong. An indicator might be the sum of the estimates. I wanted to know how difficulty the test all in all is and summed up the item difficulty estimates to do so. The sum is about -70 which seems unbelievable to me. Or is this procedure wrong? I am using constraint=items normally so I can see the difficulty directly.
What do you think I could do to check for the reasons of these outcomes?
