I have used three questionnaires in a study that all measure musical training. Each of these three questionnaires (MT1/MT2/MT3) consist of various Likert scales that are averaged to calculate the score of the questionnaire. Now I am interested how musical training is correlated with another variable S (sight-reading ability) in three different conditions. In general my linear mixed model would look like the following, if I had only one variable for musical training:

S ~ cond * MusicalTraining + (1|participant)

The scores of the questionnaires are not surprisingly correlated:

          [,1]      [,2]      [,3]
[1,] 1.0000000 0.3615148 0.7092172
[2,] 0.3615148 1.0000000 0.5723699
[3,] 0.7092172 0.5723699 1.0000000


Moreover, when calculating the reliability across all items from the three questionnaires (MT1/MT2/MT3) assessing musical training they show a Cronbach-Alpha of alpha = .86 95%CI [.8, .91].


It seems to me to be too complicated to use all three variables for musical training in the regression and I am also not to sure how to interpret the results of such a model, e.g. S ~ cond * MT1 * MT2 * MT3 + (1|participant).

I tried to run a PCA and use the first principal component in the regression. However, I think a PCA is difficult to interpret if then used in a regression.

Is there any other way to deal with this issue? Could I just average the different measurements of musical training for each each participant and use them as a single variable?

  • $\begingroup$ How is MT measured? Is this one numerical variable? If not, can you describe it? $\endgroup$ Commented May 22, 2023 at 8:49
  • $\begingroup$ @user2974951 There are three questionnaires that each assess musical training. They all consist of multiple Likert scales that are averaged to obtain the questionnaire score. $\endgroup$
    – Pearson
    Commented May 22, 2023 at 9:01
  • $\begingroup$ If you run a reliability analysis in which you enter all items from all the questionnaires simultaneously, what reliability estimate you get? $\endgroup$
    – Sointu
    Commented May 22, 2023 at 9:16
  • 1
    $\begingroup$ I meant calculating reliability across all items, yes $\endgroup$
    – Sointu
    Commented May 22, 2023 at 11:18
  • 1
    $\begingroup$ With that alpha & if there are no strong theoretical or other reasons to say these questionnaires would measure different things, I would definitely average all items (from all 3 questionnaires) into a composite measure and use that as the predictor. $\endgroup$
    – Sointu
    Commented May 23, 2023 at 12:54

1 Answer 1


If you can reasonably assume that the three questionnaires measure the same dimension/latent variable representing musical training (MT), you could aggregate the three scores and use a single composite variable representing musical training. Alternatively, you could use the three scores as separate indicators of an MT factor (latent variable) in a structural equation model.

Before doing that though it may be a good idea to run item-level factor analyses with all MT items to see how many factors you would get at the item level (i.e., is there a strong single factor/dimension of MT or multiple factors at the item level?)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.