@Astounding wrote:
You make many valid and well thought out points As a lifelong SAS bigot, I notice a few considerations that should be, well, considered.
For some variables, the character equivalent has to be invented. For example, survey responses might use a scale of:
1 = strongly agree
10 = strongly disagree
If you have to make up translations for all 10 possible values, that's work. And it makes life difficult down the road if you want to consider a range of values such as 1 to 3, or if you want to take the average of the responses.
I like the survey response example. In one set of surveys I worked with we had about 70% of the questions with the same Yes, No, Don't Know, Refused answer set. One Format covered the responses for literally hundreds of questions (similar survey over 30+ years). So we didn't need to recreate look-up lists for all of those.
And the corresponding INFORMATS for reading the initial data sets allowed setting the Don't Know and Refused to special missing in the first place so the remaining 1/0 coding for Yes/No allowed going directly to analysis without having to create yet another recoded variable.
And don't forget that some procedures really want numeric values, Proc Corr anyone?
... View more
