Aysha Abd-Rabo -

I'm not sure if independent and dependent are confused here. If you have multiple dependent variables with the same independent variable(s) then you could look at multivariate regression. Multiple regression refers to multiple independent variables. Not clear on what you mean by "dimensions."

If independent and dependent were confused here, there may be some value in looking at individual simple regressions, but the main concern may be what is needed to completely explain the dependent variable, Y, or form the best predicted-y. (For an example of explanation versus prediction, see the example in the appendix of Galit Shmueli's open version of her article on this: https://www.researchgate.net/publication/48178170_To_Explain_or_to_Predict.)

The independent (x) variables in a multiple regression work together, not what you might really consider "independent" of one another, notably impacted by collinearity but also by other considerations.

The fact that the best predicted-y is made up of the best set of variables, not too few, not too many, and just the right ones used the best way, also may impact whether we see heteroscedasticity as we should, associated with larger predicted-y-values.

Also, be careful of intercept terms. If y should be zero when the x or xs is/are (all) zero, then there should be no intercept term, or you are adding noise.

Best wishes - Jim

PS - Regarding the comment on intercept terms, see pages 109-110 of Brewer, K.R.W.(2002), Combined Survey Sampling Inference: Weighing Basu's Elephants, Arnold: London and Oxford University Press.  Mid-page 111 explains heteroscedasticity.

Hello Aysha Abd-Rabo. You said that "the independent variable contains several dimensions". I don't understand. If you can, please tell us what the independent variable is, and what the dimensions are. And what are the two DVs? Thanks for clarifying.

James R Knaub Thanks for your quick response to my question, and I really appreciate the help you gave me, especially the article.

I mean by word "dimensions.":

Bruce Weaver Thanks for your reply to my question, the independent variable and the dependent variables are as shown in the picture

Thanks Aysha Abd-Rabo. Unfortunately, I am having a hard time reading the words on that image. Specifically, what are the bits I circled in red? Is the independent variable a latent variable with the dimensions being measured (or manifest) variables, by any chance? Thanks for clarifying.

Following up on my previous reply, here is a rough illustration of the type of latent variable model that I think you might be describing. Apologies if I misunderstood.

Bruce Weaver the independent variable( emotional maturity)

the dependent variables (Self confidence and motivation)

Yes, you drew an illustration of what I mean.

Okay. For the benefit of other readers, here is an updated (still rough) diagram for your model.

What software do you typically use? And what software do you have access to?

Bruce Weaver spss

If my diagram of the model is correct, you need structural equations modeling (SEM) software, and IBM SPSS Statistics has no such commands. SPSS users tend to use AMOS. It is a separate module, and may not be available through your university. (It has never been included with SPSS at any university I have been at, for example.)

• https://www.ibm.com/products/structural-equation-modeling-sem

If you cannot get access to AMOS, you can try the lavaan package for R. I see that one can run lavaan command syntax via Jamovi or JASP too, but I have no idea if that makes things easier or not.

• https://semlj.github.io/syntax.html
• https://jasp-stats.org/2018/07/03/how-to-perform-structural-equation-modeling-in-jasp/

In any case, I think you should try to get some help from a local statistician, as I am not sure we have the complete picture yet. E.g., how are your two DVs measured? Are they scales of some kind? Or are they also latent variables?

Good luck with your work.

Just looking at Bruce Weaver 's final model (it was not clear to me the original description), am I reading this correctly that this is one latent variable and eight manifest variables all being influenced by the single latent variable? Since I do not see curved arrows, is it being assumed the manifest variables are uncorrelated after conditioning on the latent variable? If I am reading this correctly there are not any assumed relationships between latent variables, as there is only one. I don't know what the SPSS function is, but presumably you can tell an EFA function to extract a single factor and estimate the weights, can't you? But I may have misread the diagram. Is the word "dimensions" mean the relationships among these are being constrained in some way? Or is this meaning, first estimate the latent variable then use the latent variable for predicting the two other variables?

Daniel Wright, please note that I described my diagram as rough, and bear in mind that is just my guess as to what Aysha Abd-Rabo wants to do. Furthermore, I'm still not sure we have the entire picture, which is why I recommended in my previous post that Aysha Abd-Rabo seek help from a local statistician. ;-)

Bruce Weaver I know a little bit about AMOS, I'll read more about your suggestions, lavaan package for R........

thank you very much for your attention and I appreciate your time.

I took a few moments to generate another diagram (using the SEM Builder in Stata) that includes the curved arrow that (I think) Daniel was referring to. But note that I am still assuming that the two DVs are just measured variables. But we have not been told how self-confidence and motivation were measured. So for all we know, they be latent variables too. HTH.

Bruce Weaver , hopefully the questioner will say what their model is. If it is how the set of variables (EM1 .. EM6) related to the set (SelfC and Motivation) there are methods (canonical correlation at least was in the free version of SPSS when I was in grad school, and GIFI is in it which has a version). So hopefully we'll get clarification.

Study tools:

Three questionnaires were designed, one for each variable, and each questionnaire contains sub-domains

What do you know about the subdomains and how do you know It? Are they theoretical or empirically based e.g., by parallel analyses to determine how many factors for each questionnaire followed by factor analysis? (Whether or not you began with a theory, I'd still recommend the latter empirical approach.)

Paul Max Kohn An exploratory factor analysis was conducted for the three questionnaires. The first questionnaire resulted in 5 sub-domains, the second for four, and the third for two subdomains.

First, I hope that you decided on numbers of factors in each questionnaire via parallel analysis. Assuming that, you have a superabundance of choices. You could ignore the factor analytic results and do simple regressions for each dependent variable. You could do multiple regressions on each dependent variable based on the factor-based subscales of the independent variable. You could even use Structural Equation Modelling either to predict the subdimensions of your dependent variables from those of the independent variable or to predict the dependent variables from the independent variable along with the measurement models for all three initial variables.

I think you're in the best position to decide what makes most sense in terms of your research objectives.

Paul Max Kohn Yes, thanks for your detailed answer.

I will think again about the research objectives, especially since the research topic is important and these variables were not dealt with together before.