Hey there! Great question—let’s break down how to tackle iterative multiple imputation in SAS, and how manageable it’ll be as a new user.

First off: it’s totally doable, and not as intimidating as you might think. You’re right that you’ll need a macro to handle the repeated steps, but SAS has built-in procedures that do the heavy lifting for the imputation logic, so you won’t have to code the model fitting or imputation from scratch.

Let’s map this directly to the steps you outlined for R:

Step 1: Fit a model to impute missing values #

SAS’s PROC MI is your go-to here. It supports all the common distributions and model types (linear, logistic, etc.) to generate imputed datasets. Here’s a basic example to get you started:

proc mi data=your_raw_data nimpute=1 out=first_imputation;
  var age income score; /* List variables with missing values */
  class gender; /* Declare categorical variables */
  model age income = score gender; /* Specify your imputation model */
run;

This generates a single imputed dataset, just like your first step.

Step 2: Iterate the fit-impute process 50 times #

This is where the macro comes in—we’ll wrap the PROC MI call in a loop to repeat the process 50 times, each time using the most recent imputed dataset to refit the model. Here’s a simplified macro that does this:

%macro iterative_impute;
  /* Start with the first imputation */
  proc mi data=your_raw_data nimpute=1 out=imputed_0;
    var age income score;
    class gender;
    model age income = score gender;
  run;

  /* Loop 49 more times to get 50 total datasets */
  %do i=1 %to 49;
    proc mi data=imputed_%eval(&i-1) nimpute=1 out=imputed_&i;
      var age income score;
      class gender;
      model age income = score gender;
    run;
  %end;

  /* Combine all 50 datasets and add an identifier for each iteration */
  data all_imputations;
    set imputed_0-imputed_49 indsname=ds_name;
    iteration = input(scan(ds_name, 2, '_'), best.); /* Extract iteration number from dataset name */
  run;

  /* Grab every 10th dataset (10th, 20th, ..., 50th) */
  data selected_imputations;
    set all_imputations;
    if mod(iteration + 1, 10) = 0; /* Adjust since we start counting at 0 */
  run;

  /* Merge results across selected imputations with PROC MIANALYZE */
  proc mianalyze data=selected_imputations;
    /* Replace with the effects from your final analysis model */
    modeleffects intercept score gender;
  run;
%mend iterative_impute;

/* Run the macro */
%iterative_impute;

How hard is this for a SAS newbie? #

• The macro syntax (%do loops, macro variables like &i) is a basic SAS skill—you can pick it up quickly with a few tutorials or examples.
• PROC MI handles all the complex imputation logic, so you just need to tweak parameters (like model variables, categorical classes) to match your data.
• Compared to R’s mice package (where you can generate multiple imputations in one line), SAS requires a bit more boilerplate, but the logic is identical.

A few quick tips to make this smoother:

• Test with a small number of iterations first (like 5) to make sure your macro works before running 50.
• If your data has monotonic missingness, add the MONOTONE option to PROC MI for more efficient imputation.
• Always check the output logs to confirm each iteration ran without errors.

内容的提问来源于stack exchange，提问作者Adrian