Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: Multiple imputation for longitudinal data

 From Eduardo Nunez <[email protected]> To [email protected] Subject st: Multiple imputation for longitudinal data Date Thu, 2 Dec 2010 18:11:40 -0500

```Dear Statalisters,

I have Stata 11.1 (MP - Parallel Edition).

I am interested in performing multiple imputation on a longitudinal
data (on several variables with a percent of missing between 1-15%),
were subjects are the cluster units with few observations in time.
See below the data structure:

xtdes, pattern(1000)

pid:  1, 2, ..., 1438                                   n =       1432
visit:  1, 2, ..., 12                                     T =         12
Delta(visit) = 1 unit
Span(visit)  = 12 periods
(pid*visit uniquely identifies each observation)

Distribution of T_i:   min      5%     25%       50%       75%     95%     max
1       1       1         2         3       6      12

Freq.  Percent    Cum. |  Pattern
---------------------------+--------------
650     45.39   45.39 |  1...........
359     25.07   70.46 |  11..........
202     14.11   84.57 |  111.........
91      6.35   90.92 |  1111........
52      3.63   94.55 |  11111.......
44      3.07   97.63 |  111111......
11      0.77   98.39 |  1111111.....
9      0.63   99.02 |  11111111....
6      0.42   99.44 |  111111111...
4      0.28   99.72 |  1111111111..
3      0.21   99.93 |  11111111111.
1      0.07  100.00 |  111111111111
---------------------------+--------------
1432    100.00         |  XXXXXXXXXXXX

The article included in Stata FAQ ("How can I account for clustering
when creating imputations with mi impute?") suggested using a
"multivariate
normal model to impute all clusters simultaneously" or strategy 3,
although mentioned that is best suited to balanced repeated-measures
data.

Clearly, my data is not balanced. Moreover, the percent of data
missing increased as patient follow-up gets far from baseline.

Is there any other method suited for this type of longitudinal data?
If not, how stringent is the limitation of not being balanced.

Please, any help is welcome!

Eduardo
*
*   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/
```