I have two different advertisements and two dependent variables, which are the same - 'how positive or negative is your attitude toward the advertisement?'. One ad is a 'femvertisement' (feminist advertisement) and I plan on doing an ordinal regression with 3 models.
Model 1 independent variables: interest, knowledge, etc. to determine how individuals process the ads
Model 2: processing variables and feminist attitudes
Model 3: processing variables, feminist attitudes, and socio-demographics
I'm wondering whether, from the two regressions, I will be able to infer how different processing levels (as inferred from the interest, knowledge, etc.) will effect attitudes toward the advertisements. Is there an analysis I can do to compare the effects or can I explain this with the two regressions?
Thanks in advance.
Hello,
it's not clear for your dependent variables, are they a dichotomy (0: neagtive, 1: positive) ? also how you measure feminist attitudes ?
If you work with what we call "Latent variables", I suggest you to make a SEM (Structural Equations Models) ; otherwise: a logistic regression does the trick.
Hi Hannah ~
Ok - first you need to figure out what your variables are, how they are measured, and then how you want to model them.
Your Outcome variable
Attitude - how is this measured?
A composite score (one number) - or several items measuring different aspects of attitude?
If this is a latent variable (measured using several items - sub-scales of attitude) - then a Structural Equation Model would be best.
Then think of your Predictors:
Model 1 independent variables: interest, knowledge, etc. to determine how individuals process the ads
Interest / Knowledge - are these composite scores? Or do several items make up these variables?
Model 2: processing variables and feminist attitudes
How is 'processing' measured? - again, if this is indicated with multiple items, SEM is best.
Model 3: processing variables, feminist attitudes, and socio-demographics
The same applies to this model ...
If you have composite scores, and no latent variables, then numerous regression models would do the trick.
It is hard to advise when
we don't know what constructs you've measured ("etc" isn't very elaborate ;)
and we don't know how you've measured them.
but I hope this has helped you think it through a bit more :)
You need to use structure equation modeling insted of two regression model.It gives you proper answer.
Using the GraphPad Prism software you can easily compare linear or nonlinear regression models and their coefficients.
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design