www.cmm.bristol.ac.uk/forum

I am wondering if the MLwiN can support using plausible values as dependent variable in multilevel analysis?
In international large-scale assessment data (e.g., TIMSS, PIRLS, PISA), scores in mathematics for example are calculated (scaled) by using jackknifing design 'JK2, JRR, etc.' and therefore when calculating a student's performance (achievement) the programme should consider all plausible values in the analysis. For example, in TIMSS 5 plausible values are used and therefore calculating students' achievement in a subject (i.e., Mathematics) a programme need to consider the 5 plausible values in calculating the means and its standard error. The question is: Does MLwiN consider this or has an option to select all scores? For example, some multilevel programmes like HLM have this optional and a user can select to calculate the statistics using one or 5 plausible values. I assume the MLwiN have the same, but honestly still did not find that. Anyone can provide a help on this? Thanks a lot!

MLwiN does not currently have direct support for this, however you may be able to trick it by storing your different plausible values in the format output by the Realcom-Impute software, and then using the Model->Imputation Menus. This would then fit the current model for each of the datasets and use Reuben's Rules to combine the results, which I believe would be the same as what you are after.

The Realcom-Impute software would help you with the problem.