SAS: how feasible is it to do an iterative multiple imputation

1
votes

I am new to SAS, and I would like how easy/difficult it would be to try to do an iterative multiple imputation in SAS. In R, this is relatively easy.

The algorithm is as follows:

  1. impute missing data using known distribution
  2. fit model to complete data in 1
  3. use model fit in 2 to impute missing data
  4. repeat model fitting and imputation steps 50 times (e.g. 50 data sets total)
  5. take every 10th dataset and pool the results

Based on my limited experience in SAS, I'm guessing I would have to write a MACRO. I am specifically interested in using proc nlmixed to fit my model. I am not using R because SAS's nlmixed is more flexible and gives more robust results.

1
sas
Its easy depending on how well you code. But look into the PROCS specifically designed for it, instead of rolling your own. stats.idre.ucla.edu/sas/seminars/multiple-imputation-in-sas/… – Reeza
I've looked at that page, but it doesn't seem like they have specific PROCS for imputing data based on a model that I fit previously (e.g. continuously alternating between imputation and model fitting). The back-and-forth between imputation and model fitting is my main concern. – Adrian

1 Answers

1
votes
  1. proc mi NIMPUTE=n
  2. proc sort; by _Imputation_
  3. proc NLMIXED; by _Imputation_
  4. proc mianalyze;