How to impute a random distribution for missing values, within a specified range?

[slh277]

Hello, on SAS 9.4, I am trying to impute values falling below the limit of detection (LOD) for a laboratory variable (measles antibody titers). All values below 5.00 are <LOD. Rather than impute these as LOD/2 = 2.5, is there some proc I can use to impute a random distribution for this specific variable, between a specified range: 0 to 5? 

I did try setting all values "<5.00" to missing (".") in a new variable, measles_imp, and running the below code - & found an error:

proc mi data=temp out=measlesimp minimum = 0 maximum = 5;

   mcmc;

var measles_imp; 

run;

ERROR: Fewer than two analysis variables.

I only want to do this for the one variable, so not sure what to do here. Proc hpimpute does not seem to have an option to specify the range (0-5). 

[PaigeMiller]

You can't impute using that method if there is only one variable.

What is wrong with using 2.5 as the imputed value? Or a random number generator?

[slh277]

Thank you for your response!

Plotting the residuals of both log-transformed and non-transformed data, there is a "boundary line" due to the n=36 values set to 2.5. A biostatistician recommended that I consider doing random values within 0-5 rather than setting values at 2.5 to get a better data structure/distribution (see attached of log-transformed data).

I could use a random number generator, but was hoping to do this more automatically in SAS - Sorry I may be missing something, can SAS randomly generate those values and then automatically fill them into the missing value of the variable?

[PaigeMiller · Accepted solution]

A data step can replace missings with random numbers. Example:

data want;
    set have;
    if missing(measles_imp) then measles_imp=rand('uniform',0,5);
run;

[slh277]

Thank you - this is perfect!