How to test variables across two conditions?

Question

I created a model that conceptually looks like this:

• 7 independent variables
• 1 dependent variable
• 1 moderator all variables are measured on a five-point Likert scale

For the total sample, the 7 IVs and the 1 DV were asked across two conditions, to test if there would be any differences between the two situations.

Now, if I were to run 1 regression for the model described above for the first condition and subsequently run a similar regression for the second condition. Can I compare these betas across the two conditions in the same way as related topics:

• Can we compare betas of two different regression analyses ?
• Testing equality of coefficients from two different regressions

Answer 1

If I understood you correctly, you are running a multiple regression analysis with 1 criterion variable (what you call the dependent variable) and 7 predictors (what you call the independent variables) and 1 moderator. So you perform the analysis in two different conditions.

You can absolutely compare the (standardised) betas, that is why they are there. However, you need to consider if it is still theoretically meaningful to you to compare them considering the change in the conditions.

Answer 2

if you talk about regression analysis -> I suppose your AIM is to find out the predictors of your DV - for 2 conditions... But in order just to compare means between 2 independent unpaired samples (relative 2 conditions in your case), t-test is enough... see details of stat method choice here...

Concerning regression: I'd better do OLS regression not, but Partial Least Squares (as alternative to ANOVA) for your explorative univariate multivariable analysis - as it is applied based on the correlation - e.g. python here... -- to see what IVs influence most on DV under certain Conditions (taken either as random effect or covariates - could be tested both models for choosing worth while)... Or, perhaps, you have another aim of your data analysis?

PLS could be applied for different purposes & with different selection methods... though any kind of LDA (as supervised as well as PLS) or factor analysis with rotation could also be used for dimensionality reduction

in any case everything depends on the NATURE of your variables - IVs & DV (categorical or numerical) - e.g. PLS-DA is used for categorical output

So, your question can be classified as "it depends" & besides I would really recommend to reduce the dimensionality before taking care about betas in different conditions where different predictors can become leading, at all