Combining 3 categorical variables

[Simina Fernea]

To create the DV for my research model, Group polarization, I have to combine the answers of three questions together. The questions and their coding are as follows (stata comments):

1. the first question included in this variable is Q15"Would you say your LinkedIn feed primarily consists of content that aligns with your own viewpoints or includes diverse perspectives and contrasting opinions?" with its coding being: mostly aligns=1, balanced mix=2, mostly diverse=3 and unsure=4; we will treat the "unsure" answers as mid-point values


I have recoded this variable as follows, so that a higher number means more polarization:


gen reversed_q15 = .

replace reversed_q15 = 4 if type_of_content == 1 //Mostly aligns becomes 4 (less diverse)

replace reversed_q15 = 3 if type_of_content == 2 //Balanced mix becomes 3 (still balanced mix)

replace reversed_q15 = 2 if type_of_content == 3 //Mostly diverse becomes 2 (more diversed)

replace reversed_q15 = 2.5 if type_of_content == 4 // unsure treated as midpoint



2. the second question included in this variable is Q16"How often do you come across content on LinkedIn that challenges your viewpoints and prompts you to reconsider or rethink your opinions?" with its coding being: very often=1, often=2, occasionally=3, rarely=4, never=5.


The name of this variable is challenging_content


3. *create new variable for q17 that captures the responses for it *

*Q17"When you come across topics or opinions on LinkedIn that you strongly disagree with, how do you typically respond or act? Please rate the frequency of your actions"

*the actions included were the following:Engage in Constructive Dialogue,Scroll Past or Ignore,Express Dissent or Disagreement:,Unfollow or Hide Content:,Report or Flag Content


*each action had to be rated by the participant on a likert scale ranging as follows: never=1, sometimes=2 about half the time=3, most of the time=4 and always=5

*if i consider the polarization levels of each action, i would say that

*Engage in Constructive Dialogue = 1

*Scroll Past or Ignore =2

*Express Dissent or Disagreement: = 3

*Unfollow or Hide Content: =4

*Report or Flag Content =5

*all actions = to 3, 4 and 5 show a larger polarization


I have created the following:


One variable that sums the rows of these 5 actions for each participant:


* for each row, add the value for the 5 responses of Q17


egen sum_q17 = rowtotal(Constructive_Dialogue Scroll_Past Disagreement Unfollow Report)



The issue that i have is that the first question is has a range of 4 values, the second has 5, and the third has also five. How do i combine these three questions together so that i show the polarization level of each survey participant??
Should I best recode the first question so that it has 5 answers? how do i do that?

I would really appreciate some inputs here as i am very stuck with my master's thesis because of this...

[George Ford]

Why not try factor analysis (polychoric) to get a continuous variable for use as the DV?

[Clyde Schechter]

Bearing in mind that I am not a psychologist, nor a psychometrician, nevertheless having some experience with surveys of this nature, I find the idea of assigning a score of 2.5 to the "unsure" response of the first question, well, questionable. You are, in effect, saying that a person who is unsure about the mix in his or her feed is closer to diverse than to mostly aligned. I don't find that all persuasive. I would treat the unsure response as a type of non-response. In fact, I'll guess that the designers of the original survey assigned it number 4 with the intent that that would be their numerical code for item non-response. I suggest you consult your thesis advisor about this.