I am using MLwiN to analyze a data in which subjects have multiple measurements. The dependent variable is a continuous trait. When I put one specific independent variable into the model, it showed it had significant effect on the dependent variable, however, the explained variance (between subject) went down about 4%. For all the other independent variables, it is fine, this is, as long as the independent variable is significant, the explianed variance will go up. I have checked the correlation between this specific variable and the other independent varaibles and it is quite low. Has anyone here run into the sample problem? Any suggestion will be greatly apprecaited.