Good evening, David!

 If I understood you correctly, you should perform a multivariate analysis of variance. If you use SPSS, go to Analyze  - general linear model - multivariate. Depending on the type of your values put them either to fixed factors (more probable than random factors) if they are nominal, or put them into covariates, if they are ratios. Put your dependent variables in the according field. Got to 'Model' and define you model (direct and moderating effects). Go to 'Options' and tick 'observed power', 'parameter estimates' and 'homogeneity of variance' (the latter is for models with at least one nominal predictor).

Tick Box's M test (I cannot remember on which tab, but somewhere there). Run the MANOVA. You will find both F values with the according level of significance, as well as the B-value in the parameter estimates power. B-value shows in how much your dependent variable is influenced by the independent variable. Partial ETA square shows how much percent in the variance of the dependent variable is explained by the independent one.

You can find more tutorials which explain Levene's test of homogeneity of variance and Box's M test of sphericity.

I hope I understood you needs and the available data correct.

Sincerely,

Eigene

Hi Yevgen

Thank you for your answer. My values are all ordinal data derived from Likert scales. I have also been recommended to review SEM is an option. Of course this would be more complex, although I would use AMOS for the analysis. I will likely review both options.

Thanks once again

David

I would favor the recommendation to use SEM, because you can directly model the effects that interest you. 

David, thank you for your input.

I am unfamiliar at present with exactly how to carry out SEM, but I have seen a number of good online tutorials. With regards the output I am looking to achieve a number of research objectives. Firstly to show that the output variables vary as a function of the input variables. Secondly, which input variables have particularly large effects on the output variables overall. Finally which particular output variables affect which specific output variables. There is no apriori theory to guide me as to which relationships exist.

Any thoughts on how best to set this up?

Regards

David

Your model will be more complex than most, because you are specifying a second-order factor analysis, due to the sub-scales. So, there isn't much alternative but to do some reading on Structural Equation Modeling,

I did a quick skim over the table of contents for a couple of books, and I see that Byrne's Structural Equation Modeling With AMOS has a chapter (#5) that introduces the general concept of second-order factors, so that might be a good place to start.

David

Thank you once again.

I am not sure why my sub-scales are any different to separate but semi-correlated variables. I am not pulling them apart in the model. They are simply an aggregate score of the items they represent. I obviously have a lot of reading to do. Starting with your book recommendation.

Kindest regards

David

I would go for AMOS if I had more complex relationships (mediation, moderation, etc). You can save your items as supervariables (dimension reduction procedure in SPSS. Inexperienced several problems with the second order constructs in AMOS.

You can also use SmartPLS. The version 2 is available for free. It is a bit tricky with the second  order constructs, but you have less problems with legitimation procedure. For AMOS you need to perform several tests for discriminant and convergent validity, internal consistency, common method boss, model fit, and multicollinearity. Yes, and the data should be normally distributed. For SmartPLS you need less tests. So, I would go for ANOVA in simple relationships with a previous  aggregation of items to supervariables. Otherwise SmartPLS. AMOS would be the third option.

Best,

Eugene

In such cases SEM is a good approach to handle the complexity.