Help on stats for research model needed

#1
I am in need of a quick statistical advice for a study I planned, and I hope you can help me a bit with it. I'd really appreciate your expertise. The research is on nudging (subtle psychological manipulations). My model is attached to the post.


The model looks kind of funny, so let me explain:
Very quickly: Independent variable: Nudge transparency with 3 levels: none, medium,high.
Dependent variables: Agent Awareness (the degree, to which people realize they have been manipulted), measured on 1-5 scale from 1=none, 5= very high, and Nudge effectiveness (percentage)
Mediators (or moderators) between awareness and effectiveness: Intentions & attitudes (Likert scales) and psy. reactance (Likert scale)

The effect of transparency on effectiveness is easy to test - 1-way ANOVA should do? Same goes for transparency on awareness.
However, I am having mixed feelings about what to do further. I want to see what happens to the transparency-effectiveness relationship, once I introduce awareness to the model. If transparency and awareness are strongly correlated (as I expect), I should perhaps include awareness as a covariate and look for the difference in coefficients to find the unique effect of transparency on effectiveness? I think I should also test for mediation (Barron & Kenny 4 step regression model) and moderation (multiple regression)?

Further I need to see whether awareness triggers effects from reactance and attitudes&intentions on effectiveness. Here's where I get totally lost. I am struggling between multiple correlations, stepwise regressions, struc.equation modelling, etc. Is there a fast way to test my model instead of running mediaiton and moderation tests on all possible relationships in it? The fact is, it is an exploratory study, and I cannot take guidance from anywhere.

Thank you in advance for your help, it is much appreciated.
 
#6
Hmm, would simple regression model work with ordinal IV?
Hmm not sure if I understand, but your endogenuous variable would be effectivness which is influenced by awaraness, attidutes and reactness, and their interactions. Now if i understand the arrows the effectivness does not reverse cause these covariates so with that respect E(epsilon|X) should not be broken. If you have correctly specified models, have no omitted variables, sampling issues i see no need for the IV concept. Or am I missing something?
 
#7
Dear Gokirop,

The problem is there is no information available yet (thus it is an exploratory study) in relation to the nature of any of the relationships in the model. i.e. I don't know whether to expect mediation, moderation, or mediated moderation b/n Transparency(ordinal)-Awareness(ordinal)-effectiveness(interval). Then I have to examine the Awareness(ordinal)-intentions(ordinal)-reactance(ordinal)-effectiveness(interval) and test what happens there - whether intentions & reactance mediate or moderate the awareness-effectiveness relationship. Thus I think I have to take awareness as IV on the second stage. Does that make sense?