Combining two Likert-scale questions into one variable

[Caroline Jaszuk]

Hello everyone,
I have model with 4 main effects (my independent variables) and one dependent variable. I conducted a survey where I have 2 questions pers ech variable, all measured on the 1-7 Likert scale My questions are:

1. How do I combine the two questions per each variable so that they indicathe each variable together. I read that I could do this with factor analysis - can someone please explain how I should do it? Should I check any assumptions in the process?
2. Should I calculate the Cronbach's Alpha?
3. How should I further carry out my analysis? I was going to do a Multiple Linear Regression Model.

I would very much appreciate your helpful!

[Daniel Schaefer]

Welcome to the forum Caroline. Questions like this don't usually get answered. These questions are vary vague, and a few would require a class or textbook to properly address. Please do read through the FAQ at the top of the page, especially the advice on asking questions. That said,

How do I combine the two questions per each variable so that they indicathe each variable together.

It depends on the variables. If the direction of both scales is the same (you don't need to reverse code) the simplest way is to add them together. Of course, you should make yourself aware of the problems with that kind of thing first.

I read that I could do this with factor analysis - can someone please explain how I should do it?

Yes, either exploratory or confirmatory factor analysis should work well here. If you want to learn more, I'd start with this book:

DeVellis, Robert F. 2017. "Scale Development: Theory and Applications." 4th ed. Thousand Oaks, CA: Sage Publications.

Should I check any assumptions in the process?

You should always try to check your assumptions.

How should I further carry out my analysis?

Sorry, but this is way too vague to answer. It depends on the specifics for your project and it wouldn't be possible to properly address this in a single post. You should start with some advice from an advisor, mentor, or teacher.

I was going to do a Multiple Linear Regression Model.

Sure, good idea. That is likely one of many things you should do for a statistical analysis. Why not include both scale items in the regression to start?

[Frode Andre]

It sounds like your independent and dependent variables are latent constructs, and you aim to generate each of these variables using two questions each. Is this true?
If so, you may want to look into the sem command (structural equation model), these models are easily built in the sembuilder. Your model looks similar to the model on the site below (the first graph, following the text

Enter your model graphically,

, but you have more independent variables.

https://www.stata.com/features/overv...tion-modeling/

Note that if you want to construct a latent construct – and evaluate it using measures as Cronbach’s alpha – you should at least have three questions. But you may try and estimate a full SEM-model using less than two questions per latent construct.