Proc MI: impute only some missing values and select range of imputed values

riccardo88 — Wed, 06 Jan 2021 16:57:20 GMT

Hallo everybody,

I am working with a database that looks like this:

data have;                    
input id trt $ var1 $ var2 $;
datalines; 
1 placebo 0.2 3.0
2 placebo 1.1 <1.0
3 drug <0.2 2.0
4 placebo 2.3 .
5 drug <0.2 2.2
6 drug . 1.2
;

As you can see I have missing values and values that are reported as "below the limit of detection" from the laboratory.

My aim is to use multiple imputation to impute ONLY the values under the limit of detection and not the real missing values.

To do this I differentiated the two "kinds" of missing values like this:

data have; set have;
		if var1='' then var1_n=.a;
	else if var1 ne '' then var1_n = input(var1, 8.);
	if var1_n=. then var1_n=.b;

		if var2='' then var2_n=.a;
	else if var2 ne '' then var2_n = input(var2, 8.);
	if var2_n=. then var2_n=.b;

	drop var1 var2;
run;

Now A are real missing values and B are values below detection limit.

My questions are:

1) Is it possible to use proc mi specifing that I want to impute only the missing values specified with B and not all missing values?

2) Is there an option in proc mi to specify that the imputed values should be between 0 and the limit of detection (for example for var1 between 0 and 0.2)?

Thank you very much for your help!

SAS Version: 9.4

Re: Proc MI: impute only some missing values and select range of imputed values

ballardw — Wed, 06 Jan 2021 17:21:48 GMT

1) Is it possible to use proc mi specifing that I want to impute only the missing values specified with B and not all missing values?

Yes. The though process would be a bit different than state: Where the value is not .A. You can use a data step option : data=have (where=(var1_n ne .A)).

Since you have two (or possibly more variables) that would involve a separate imputation step for each variable individually since the subset of records is different for each Variable.

2) Is there an option in proc mi to specify that the imputed values should be between 0 and the limit of detection (for example for var1 between 0 and 0.2)?

The Proc MI statement options MINIMUM= and MAXIMUM= set the limits on imputed values.

topic Proc MI: impute only some missing values and select range of imputed values in Statistical Procedures

Proc MI: impute only some missing values and select range of imputed values

Re: Proc MI: impute only some missing values and select range of imputed values