topic Re: How do create an accurate scoring method? in SAS Data Management

How do create an accurate scoring method?

Missmichelle — Mon, 21 Oct 2019 20:50:24 GMT

Hi!

I am not sure how to phrase my question. So I have a survey data set that I am analyzing. I have a variable that holds the count of how many times a survey participant indicated they "needed" a service. My goal is to create a "needs score" of some sort to categorize the sample in a way so I can have a group who is high, medium, and low need. I also want to cross this needs score with demographics such as race gender age etc, eventually (perhaps some chi-square analysis as well?). How should I go about this?

I would appreciate any suggestions you have!

Thank you,

Michelle

Re: How do create an accurate scoring method?

TomKari — Mon, 21 Oct 2019 21:14:43 GMT

Hi, Michelle

First question, are you planning to do this completely within a tool like Enterprise Guide, or are you comfortable with SAS language programming?

Next, we really need to get a bit better idea of what you're dealing with. Please show us what your data looks like, and what your desired result looks like. Use text, not a picture, so we can cut and paste it.

Tom

Re: How do create an accurate scoring method?

ballardw — Mon, 21 Oct 2019 21:20:03 GMT

Have you done any distribution analysis (simple Proc Freq to start with) on that raw "need" variable?

Of if your survey data comes from a sample that had a complex sample, Proc Survey Freq?

I would expect that may give you some idea of where to place break points for categories. If you have a very large range of "need" responses you may want to look at the percentiles as well.

Also, you likely would not need to add any variables for the other analysis, create a custom format or two and use the format in the analysis.

The format might look like this:

Proc format library=work;
value needscore
0 - 1 = 'Low'
2 - 5 = 'Medium'
6 - high = 'High'
;
run;

You don't mention anything about actual range of values or number of respondents. I would suggest if you want to have something moderately sensitive when split across race or age groups that you set ranges likely to keep at least 5 if not 10 respondents in each "need" interval.