What type of regression model do I need to use?

Question

I am looking at the relationship between the two concepts of psychological strategy (PS) usage and athlete engagement (AE), looking to see if PS predicts AE as a whole and if certain subscales of PS predict subscales of AE.

I thought my data was ordinal but having second thoughts now as it may be different due to how I have inputted the data. Each component of both questionnaires is on a Likert scale of 1 to 5. In SPSS, I have inputted each item separately (e.g., item 1 and inputted their answer, e.g., 4) and then calculated specific subscales in a separate variable column. There are 9 subscales for PS and 4 for AE plus AE as whole.

Answer 1

You will face a number of technical difficulties. First, the AE subscales are correlated with each other, so looking at the results from individual regressions in isolation is going to overestimate the effect of the PS predictors. Second, as you already surmised, the data are not really interval data, so calculating the subscales is questionable (even though "everybody" does it).

My guess is that you should sit down first and formulate a few hypotheses. Which measures and subscales of AE do you think should be related to the PS measures and why? Which of the PS measures and subscales do you think they relate to? DO THIS WITHOUT PEEKING AT THE DATA!

After that, you can look at correlations. Are the PS and AE measures and subscales correlated in the way your hypotheses say they should be? To start with, ordinary Pearson correlation is probably good enough. You might eventually want to switch to something that more closely models your data, such as Kruskal's gamma.

Finally, you might want to consider ordinal logistic regression, possibly multivariate ordinal logistic regression.