topic Re: Average of multiple variables and also versions of them in SAS Programming

Average of multiple variables and also versions of them

shasank — Wed, 14 Apr 2021 19:16:56 GMT

Hi SAS community,

I have the below dataset and need to calculate mean of all the variables possible with these 5 variables.

For ex: Mean(Ques fres tas mas des) and Mean(Ques fres tas mas) and (Ques fres tas )...

Is there a way to get the mean of all variations possible with these 5 variables.

data a;
input Ques fres tas mas des ;
cards;
2 1 4 2 3
5 1 5 5 7
3 3 4 6 8
4 2 4 4 8
9 5 1 2 3
8 2 4 5 3
4 3 3 2 7
;
run;

Let me know if this is possible .

Thank you for your time and effort.

Re: Average of multiple variables and also versions of them

ballardw — Wed, 14 Apr 2021 20:06:47 GMT

Look at the SAS Functions Call AllComb in the documentation.

You do realize this requires adding 26 variables to your data set? Hint: Comb function counts combinations of R elements from N choices. How are you going to keep track of which value is which set of variables?

Before going into details what is the utility of this?

Re: Average of multiple variables and also versions of them

shasank — Wed, 14 Apr 2021 20:10:56 GMT

Thank you for your reply. Yes, I do realize that. The purpose of this is to
take a mean of all the combinations possible with these 5 variable and the
next step is to find the find the car combo with mean closest to a
different variable in another dataset.

Re: Average of multiple variables and also versions of them

PGStats — Wed, 14 Apr 2021 21:04:57 GMT

Here is a simple way to get the means from all possible subsets (except the empty subset):

data a;
input Ques fres tas mas des;
cards;
2 1 4 2 3
5 1 5 5 7
3 3 4 6 8
4 2 4 4 8
9 5 1 2 3
8 2 4 5 3
4 3 3 2 7
;

/* Get the number of all possible subsets except the empty subset */
data _null_;
set a;
array x _numeric_;
nbSets = 2 ** dim(x) - 1; 
call symputx("nbSets", nbSets);
stop;
run;

data want;
set a;
array x _numeric_;
array y {&nbSets.};
do s = 1 to &nbSets.;
    ss = s; sum = 0; nb = 0;
    do i = 1 by 1 until(ss = 0);
        if mod(ss, 2) = 1 then do;
            sum = sum + x{i};
            nb = nb + 1;
            end;
        ss = int(ss/2);
        end;
    y{s} = sum / nb;
    end;
drop s ss i sum nb;
run;

I wouldn't try to scale this method beyond n=12.

Re: Average of multiple variables and also versions of them

ballardw — Wed, 14 Apr 2021 21:46:40 GMT

@shasank wrote:
Thank you for your reply. Yes, I do realize that. The purpose of this is to
take a mean of all the combinations possible with these 5 variable and the
next step is to find the find the car combo with mean closest to a
different variable in another dataset.

So what happens when you have the same mean for more than one of these combinations?

If any of the variables are ever missing then the one of the choose 2 versions will always have the same mean as one of the single variable. More missing values per record means even more same mean values are likely.

Re: Average of multiple variables and also versions of them

shasank — Thu, 15 Apr 2021 00:24:53 GMT

That's a great point. Thank you for the critical thinking. My actual data is in a format if 11.11111 and doesn't have any missing values. The reason for this request is to find what combo of variables did an equipment use to get to a final number. The machine always uses a set of variable and I am trying to find a trend of which is the most used combo.

Re: Average of multiple variables and also versions of them

shasank — Thu, 15 Apr 2021 00:35:44 GMT

Thank you so so much for the solution. I am trying to modify this code to get the combination calculated as variable name. Is there a way to modify this code to get the combo variable used to get to the mean???

Re: Average of multiple variables and also versions of them

PGStats — Thu, 15 Apr 2021 00:54:07 GMT

Get the combos with:

data combos;
set a;
array x _numeric_;
length combo $160;
do set = 1 to &nbSets.;
    ss = set; combo = " ";
    do i = 1 by 1 until(ss = 0);
        if mod(ss, 2) = 1 then do;
            combo = catx("-", combo, vname(x{i}));
            end;
        ss = int(ss/2);
        end;
    output;
    end;
stop;
keep set combo;
run;

Re: Average of multiple variables and also versions of them

shasank — Thu, 15 Apr 2021 01:43:10 GMT

Sorry for not being clear.
y{s} = sum / nb;
Instead of y1,y2,y3 is there a possibility to have the combo as a variable name. This would help know which all were combined to get to the value in the column.