A valid way to combine different scales into one?

[jimin kim]

Hello,

I want to know if there is a valid way to combine three different scales into one continuous variable.
- Loneliness: 1 question on a 5-point scale
- Isolation: 3 questions on a 4-point scale
- Perceived social support: 12 questions on a 7-point scale

I know it's dangerous to put different scales together, but according to previous studies, these three are sub-factors of "social isolation", and there are scales that integrates these three in one scale. Unfortunately, the integrated scale was not used in the data I use, so I am trying to find the way i can integrate separate scales.

The first thing I tried was the latent profile analysis. As a result of LPA, I got the class variable which is ordinal, but when I tried the meologit command to verify the influencing factors for it, the fixed model was okay but the random coefficient model did not proceed. I guess it is because the number of goups and samples per group is not enough to run multi-level logistic regression with random coefficient.

To combine scales into a continuous variable(because I got to know my data size is not enough for multilevel logistic), I'm considering using z-score and factor analysis but i am not convinced if it is okay.

Please give me any advices.

Thank you!

[George Ford]

factor analysis would be one way to do it. need to use polychoric correlation matrix as the variables are ordinal.

[Nick Cox]

These questions arise quite often here. It's simple but perhaps not quite trivial to underline that researchers vary quite a lot in what they would do or not do. For example, it's my impression that recourse to PCA or factor analysis seems rare among economists, but more common in some other social or behavioural sciences.

https://www.statalist.org/forums/for...onbach-s-alpha is a concurrent thread which raises similar issues.

Search for Cronbach alpha or indeed polychoric will find some (but not all) similar threads.

[George Ford]

One issue that loneliness is sometimes taken to be a function of isolation and social support. And, I suspect that all those topics/responses may not condense to a single factor.