Hello Shanjida Chowdhury. Forgive me for responding to your question with a question. Why do you need to create one variable with 3 levels? What are you trying to do ultimately? I ask out of concern that you may be falling prey to the XY problem, as some people call it.

• https://xyproblem.info/

Thank you for clarifying.

Hi Shanjida Chowdhury

by forcing your five response options into three, you will lose valuable information. It is questionable whether a variable/factor consisting of items on what will then be a 3-point Likert-scale can still be treated as a meaningful variable with a mean and standard deviation in subsequent analyses.

I once thought about reducing a 4-point Likert-scale into a 3-point Likert-scale, but only because I tried to compare measurement invariance between two groups and the programme did not let me do this because one group did not choose the lowest response option "fully disagree" even once. In the end, I decided against it. It was just not worth sacrificing that much information.

As Bruce Weaver has pointed out, your argument seems to be upside down: You say you need to reduce the number of response options, but are looking for justification at the same time.

My advice would be the following: don't, unless you absolutely have to.

Best

Marcel

One approach would be to conduct taxometric analysis (https://www.amazon.com/Multivariate-Taxometric-Procedures-Distinguishing-Quantitative/dp/B015X4MUCS/ref=sr_1_1) to see if the five questions do differentiate people into three categories, rather than into dimensions. If not, then doing this would be inappropriate. Of course there may be other reasons why the three levels is needed (like the red, green, yellow lights you see on traffic lights and some warning systems). As Bruce Weaver says, tell us why.

Transforming a 5-point Likert or Likert-type scale into three categories should not be performed unless you have a valid rationale for it. That being stressed, using SPSS (Transform> Recode into Different Variables), you could record 1 and 2 into Low (its value = 1), keep 3 as Moderate (its value = 2), and record 4 and 5 into High (its value = 3). For various methods of running a similar procedure, you could refer to the Libguide by Kent State University Libraries, fully cited below.

KSU Libraries. (2021, October 4). SPSS tutorials: Recoding variables. Libguides at Kent State University. https://libguides.library.kent.edu/spss/recodevariables

Good luck,

Note to other commentators: She is not asking how to turn a 5 category variable into a 3 category one.

Hello, Daniel Wright thanks for your comment. Actually, I have to develop one index, and responses of 5 point scale show that the maximum answer lies in Agree, indecision, and disagree categories. As the frequency is low in the Strongly agree and strongly disagree category, that's why I thought to reduce to 3 categories with names low, moderate, and high.

Hello, Marcel Grieger thanks for your comment. Actually, I have to develop one index, and responses of 5 point scale show that the maximum answer lies in Agree, indecision, and disagree categories. As the frequency is low in the Strongly agree and strongly disagree category, that's why I thought to reduce to 3 categories with names low, moderate, and high.

Thanks Mohialdeen Alotumi

Hey Shanjida Chowdhury

EDIT: I am just confused now. Please disregard my comment in case I am missing the point here.

Not having too many people tick "strongly disagree" or "strongly agree" happens, see the example I mentioned. To me, that is not a valid reason to reduce a 5-point Likert-scale into a 3-point one, but I would like to hear other people's thoughts on this.

Best

Marcel

Dear Shanjida Chowdhury ,

from what I understand, you try to classify respondents of 5 questions in regard of their "gender awareness" (with 3 categories) and as Daniel Wright points out, not to reduce data in terms of response options endowed. I think Bruce Weaver has a viable point in asking why you want to do this. I assume you want to perform analyses including this variable as IV or DV. I think that the amount of categories should be the result of theoretical considerations (e.g., is it necessary to include a "moderate" category?) and agree with you Shanjida Chowdhury that you do not need to include all response options in your index or score variable (that´is what your gender awareness represents).

Here is, what I would do:

1. Assign values to your responses (e.g. 1 to 5)

2. Aggregate your respondents' answers (mean or sum)

3. Define cut-off values for your categories. You could either do this on theoretical grounds or you basicaly use the resulting distribution of sums or means to categorize into equally sized categories. For example the median split method is used to obtain two equally sized groups wth values either higher or lower than the median. The same principle applies to the more than two categories. The advantage of the latter approach would be that categories that contain an equal amount of participants are more likely to allow you to use parametric instead of non-parametric methods to analyze your data.

It should be restated however, that all these steps depend on your ultimate goal as indicate before.

Best of luck,

Sven

Shanjida Chowdhury , you say "Actually, I have to develop one index." You have not explained why you "have to."