topic Re: how to handle multi choices question in SAS Procedures

how to handle multi choices question

Radwan — Mon, 18 Mar 2019 09:36:50 GMT

Hello everyone here

please i have an issue which is

i have data from questionnaire, but there are some questions as multi choices means the answer wil be over two values, so please how can i deal with such problem.

especially when i need to know the frequency for the choices.

thanks in advance.

Re: how to handle multi choices question

Kurt_Bremser — Mon, 18 Mar 2019 09:52:02 GMT

Please supply example data in usable form (data step with datalines), so we have something to test code against.

And an example for the expected output form that example data.

Re: how to handle multi choices question

RW9 — Mon, 18 Mar 2019 09:53:24 GMT

A common issue, occurs when people insist on keeping more than one data item in a single variable. Expand your data out so that each observation contains one data element per data item, then run your procedures against it. You can always alter for output later:

So:

1,5,2,6,7 would become:

Re: how to handle multi choices question

Radwan — Mon, 18 Mar 2019 10:18:26 GMT

when i run proc means the variables with multi responses do not appear

Re: how to handle multi choices question

Radwan — Mon, 18 Mar 2019 10:20:13 GMT

YOU mean i have to separate the question into sub-questions that equal to the number of choices in the question?

Re: how to handle multi choices question

RW9 — Mon, 18 Mar 2019 11:03:48 GMT

I don't know if you utilise CDISC models, however that should instruct you on how to model questionnaire data. Each row of data presents:

Questionnaire Question Response

In this way you can proc freq all question separately, apply multi level formats to them so they get categorised correctly etc. This is the way you should be modelling your data anyways, however it is done that way for a reason!

For example, i have questions 1, 2, 3

My data looks like:
Question Response

1 1

2 0

3 1

To group 1 and 3 together I can use a multi-level format:

proc format;
  value tmp (multilabel)
    1="1"
    2="2"
    3="3"
    1,3="1,3";
run;

data have;
  question=1; result=1; output;
  question=2; result=0; output;
  question=3; result=1; output;
run;

proc means data=have;
  class question / mlf;
  var result;
  output out=want n=n;
  format question tmp.;
run;

Re: how to handle multi choices question

ballardw — Mon, 18 Mar 2019 14:49:14 GMT

@Radwan wrote:

Hello everyone here

please i have an issue which is

i have data from questionnaire, but there are some questions as multi choices means the answer wil be over two values, so please how can i deal with such problem.

especially when i need to know the frequency for the choices.

thanks in advance.

My only somewhat tongue-in-cheek answer would be: Refer to the analysis plan that was drawn up before the data collection occurred.

Any data collection should support analysis. Which means that a plan should be in place as to how each component is to be used the analysis after collection. If you don't know how you are going to use a data item then I would ask: Why collect it?

Some things to consider with multiple choice answers are such things as:

Does the order of response make a difference: example is selecting response 5 before response 2, as in your first and third rows, to be treated at all differently then 2 before 5, as in the 4th row?

Are single responses such as the second row different than when a choice appears in combination with others?

Or specific combinations of particular interest occurring together?

Is the number of responses of importance? Possibly the main point of interest?

Or how many any specific responses occurred regardless of order?

Addition bits are how this variable is to be used against/with other variables? Is Q22 going to be used to "predict" variable Q23 (or other)

This is not exhaustive but gives some idea of the things you can get involved with. Many of these will require splitting out into additional variables such as Q22_1 = 1 when the value 1 was selected and 0 when not. Or creating another variable to hold the count of responses, of if specific combinations of interest occurred such as Q22_123 =1 when all (or possibly any) of responses 1, 2 and 3 are selected and 0 otherwise, or the number of 1, 2 and 3 when present and 0 otherwise.

What question(s) are the responses to Q22 supposed to answer? That is what the analysis plan should cover as well as specific measures or tests to be used.