Help understanding how Mixed Models handle missing data

Hi All --

So I've taken several classes and workshops on Mixed Models, and in all of them got the message that mixed models were useful in cases when independent variable data might be missing, as missing one value for an IV wouldn't result in the observation being lost for the analysis. However I've recently run a mixed model in SPSS, where it seems that is exactly what is happening -- even though I have 4500 observations with data for the dependent measure, when I save the predicted values from the model, the model is only predicting values for something like 3000 observations, and this appears to be happening because cases are missing data for even a single one of the 16 or so IV's I have. Did I just radically misunderstand how Mixed Models are handling missing data? What can I do to bring some of those 1500 observations back into the analysis?

Many thanks in advance for your help!


Less is more. Stay pure. Stay poor.
I may be wrong, but could the model be using the full data for calculation of coefficients, but it is then unable to "apply" those coefficients to rows with missing data for prediction, since it does not know what value to assign the missing variable in it's design matrix?

So say I am predicting you body mass index (BMI). I have your weight and sex, but missing your height. The model may use the available data in the analyses, but if you ask it to predict your BMI it can't since you didn't provide it a value for your height!
Thanks hlsmith. It is hard to tell whether the output for the model supports it being built on the full data -- the degrees of freedom for the IV's suggest no, but I know that DF's in mixed models are often complex and aren't necessarily a perfect map on to sample size...