I'm trying to use STATA ice command for multiple imputation for my
multilevel data, students nested within school. As far as I know,
STATA does not provide special function for multilevel multiple
imputation. But many people still use it to impute missing values in
hierarchically structured data. Put aside any discussions of potential
problems that it may cause, I'm just wondering technically how you
could impute both individual- and group-level variables in STATA.
When I used 'ice' command along with 'eq ( )' to restrict predictors
in the imputation for school-level variables (that is, only
school-level predictors are said to be entered to impute school-level
variables), imputed values turned out to be different across students
within the same school.
I've found article by Han (2008, Developmental Psychology), where she
said that she assigned the same imputed values for school variables to
students from the same school to preserve multilevel data structure in
multiple imputation procedures. She said she used STATA. I am
wondering technically how you could do this.
And I also found someone else did it by dividing a master data set
into two subsets: individual level and school level, and then
performing multivariate imputations at each level (Jeong, 2009, EEPA).
In this case, you would get, let's say 5, 5 imputed individual level
datasets and 5 imputed school level datasets, right? Then, doesn't it
produce 25 combinations of datasets??
Also, if some of school-level variables should be obtained by
aggregating student-level data, are these aggregated variables
supposed to be included in school-level variables imputation model?
Is there any way to deal with this issue when using STATA 'ice'
command? How would you usually produce both imputed school- and
student-level variables using STATA?
And my second question is about structurally missing values, like
valid skips (missing father's education for mother-only families)..
Including these valid skips as missing values to be imputed seems not
correct. Do you have any special techniques for these structurally
missing values? Would you please let me know how you could handle
these valid skips when you do multiple imputation?
Your help would be greatly appreciated. Thanks
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
