Problem with variance absolute

[MaraMlwin]

Hello everyone,

I'm running a binary response 3 level-model, MQL1 extended to PQL2. I am checking the effect of some predictors in order to expplain student achievement. One of the predictors, has a strange for me "behaviour". The absolute of the variances overlap the absolute variance of model 0. I asked my advisor and told me to calculate the variance in each level by taking into account the ceof. minus SE (of each level variance) and then substract them to find the asolute and then the explained variance. But it stills overlap model's 0 total.

Is there anything else to do in order to fix this?

I'm new with mlwin procedures. I have read related modules about binary response, but i can't find any related advice about that. I don't know even how this problem called. Is it called "overfitted variance"? If not, then how? Because i would like to learn further about that and how to solve it.

Kind regards,
Mara

[ChrisCharlton]

Are you able to provide any further information about this, such as the actual model specification and results, along with a worked example of your calculations?