Proc tabulate with percent that do not add up to 100%?

[gabon]

Hello,

I have 21 variables named HS04A --> HS04U with possible values 1,2,8 and 9. I need to find out for every variable how many times the value 1 occurs and make a percentage out of it (divided by all observations). Basically the variables are diseases and value 1 means disease is present for that record (a person). I need to find out for every disease in how many percent out of all the people it occurs. Then I need to output a table where the 21 variables are rows and the one column is the percentage. I include also image of the layout of desired output. Is it possible to make something like that with proc tabulate? Or should I rather find out how to do it with proc sql? Thanks in advance for any advice

[Tom]

Sounds like you have values from a survey with the question coded as 1=YES 2=NO with 8 and 9 some type of missing categories.

If you recode it into a a nice BINARY variable then the MEAN is the PERCENTAGE.

[gabon · Accepted solution]

Hey Tom, thanks a lot for the idea, tried it and it worked. I did some primitive recoding, not sure if it's effective, instead of 1 I put 100 so the percentage is in format for example 3,1 instead of 0,031 and then executed proc means.

data work.subset6a;

  set library.database;

  keep pid HS04A--HS06U WGT;

  if HS04A^=1 then HS04A=0; else if HS04a=1 then HS04A=100;

  if HS04B^=1 then HS04B=0; else if HS04b=1 then HS04b=100;

  if HS04C^=1 then HS04C=0; else if HS04c=1 then HS04c=100;

....etc.

run;

proc means data=work.subset6a mean maxdec=1;

  var HS04A--HS04U;

  freq WGT;

run;

And the output looked like this:

[ballardw]

To expand on Tom's response,with a binary 0/1 coding you can use:

var*mean*f=percent8.1 (for example to get a percent) or if you don't want % signs then use a custom picture format.

The added advantages of this coding (besides use in logistic regression)

Sum= will give you the total Yes answers

N= gives you the number of responses

[gabon]

Thanks, at first I tried to recode the data with hundreds and zeros to get my mean correct, but after your advice that formatting can help, I found proc template for proc means and changed format for mean to percent and it worked, so I can have my data recoded to ones and zeros.