Combine items (5 point likert) into one scale

[jgraupner]

I've used an externally validated, 17-item self-efficacy scale at pre/post an intervention. Answers range from 1 to 5 on a Likert scale. To analyze the different between pre/post, I've been told to combine all 17 items into one item and then do a t-test (or rank sums since ordinal?).

My question is what is meant here by "combine" and how do I go about doing this in SAS?

data looks like this:

Survey id             pre/post            item1           item2       item3     ..........        item 17

1                             pre                  4                    3             3                               4

1                             post                5                   5              4                               5 

2                             pre                  3                   4              5                               3

2                             post                4                    4             5                                4

***

20                           pre                  3                    4             5                               4

20                           post                5                     5             5                               5

[Reeza]

You need to ask the person who told you to 'combine them' what they meant. Depending on your data there may be a common composite score where you make sure all are aligned and sum them or average them or have to weight some in certain ways. Once you understand the methodology of combining them the code is quite simple. Figuring out what to code is always the hard part but this isn't really a question for us.

[novinosrin]

Very neat.

For all I know from my short experience in industry so far:

1. There is a disconnect between Industry and Academia

2. There is a disconnect between Business & Technology

3. There is a disconnect between Business  & process

Overall, it appears the show is run by sales, only sales and more sales, everything else is compliance. 

[Reeza]

---

@novinosrin wrote:

Very neat.

For all I know from my short experience in industry so far:

1. There is a disconnect between Industry and Academia

2. There is a disconnect between Business & Technology

3. There is a disconnect between Business  & process

Overall, it appears the show is run by sales, only sales and more sales, everything else is compliance. 

---

1. Yes

2. Depends on org

3. Depends on org

Unfortunately, in my experience it takes a really good data science/analytics manager to manage these relationships and have the correct skill set to cross all three of these areas. Not many people have the ability to cross all three.

[jgraupner]

I went back to the email and had gotten the explanation wrong. They said  "add the 17 items into 1 scale and compare the mean total scores between the pre and posttest." So it looks like I should sum up the items and compare means. Does this mean I create a new "sum" variable? Thanks.

[Reeza · Accepted solution]

Yes, you add up all 17 into one score and then you can use PROC TTEST with a paired t-test most likely unless you were planning to run more complicated model via GLM or something else.

[jgraupner]

Great, thanks. Do I use the SUM function? Do i have to use a class statement?

I'm trying to sum (item1, item2, item3...item17) but it doesn't appear to be the right thing to do.

[Reeza]

That's exactly the right thing to do. Show your code if you think it's wrong somehow.

[jgraupner]

If I want to do a paired t-test but my data (including new summed items variable) is organized like above (with pre=1, post=2), do i need to create new variables for a pre_sum_item and post_sum item?

so that I can do something like this?:

 proc ttest data=response sides=2 alpha=0.05 h0=0;
	title "Paired sample t-test example";
	paired pre_sum_item * post_sum_item;
   run;

 if so, how? Thanks very much for you time!

[Reeza]

Transposing data tutorials:
Long to Wide:
https://stats.idre.ucla.edu/sas/modules/how-to-reshape-data-long-to-wide-using-proc-transpose/

https://stats.idre.ucla.edu/sas/modules/reshaping-data-long-to-wide-using-the-data-step/