Setting some values of a continuous variable to missing randomly

[mmovahed]

I like to replace some values of a continuous variable to missing by random. I wonder what is the code in SAS for this.

Also Like to know if I can do multiple imputation for variable with whole missing values based on other variables in the data set. If so I wonder what is SAS codes for this as well.

Thanks

[PaigeMiller]

---

@mmovahed wrote:

I like to replace some values of a continuous variable to missing by random. I wonder what is the code in SAS for this.

---

Seems like an odd thing to do.

data want;
    set have;
    random=rand('uniform');
    if random<0.2 then x=.;
run;

Also Like to know if I can do multiple imputation for variable with whole missing values based on other variables in the data set. If so I wonder what is SAS codes for this as well.

Look at the examples in the documentation for PROC MI.

[mmovahed]

Thanks. It was useful.

[hashman]

@mmovahed:

The answer depends on whether you want to replace a variable with a missing value:

1. in an exact number of observations
2. approximately in some percentage of observations

A program for #1 will be quite different (and more complex - there're different algorithms for doing it) than for #2 at which @PaigeMiller has hinted.

And also it depends on whether every time you run the code you want to:

1. get exactly the same results
2. get different results
3. pick the observations on run #n all different from run #m
4. the missing values to be distributed uniformly or in some other manner

and possibly a number of other factors.

Kind regards

Paul D.

[ballardw]

---

@mmovahed wrote:

Also Like to know if I can do multiple imputation for variable with whole missing values based on other variables in the data set. If so I wonder what is SAS codes for this as well.

---

You may need to describe exactly what you mean by "whole missing values ". That sort of sounds like you have no actual values for a variable. Without at least a few values it is going to be real hard to determine any sort of rule that uses other variables to impute.

If you have a categorical variable that should have 4 levels but one of the levels doesn't appear, for example, then what rule would be used to determine the missing should be the category with no representation?

[mmovahed]

Hi

Thanks for your explanation. You are right. I think we need at least few available values for the variable with missing.

M