Help me with Statistics: Multiple Regression

OtistheGuard

Hi everybody. I'm trying to wrap my head around how to do a multiple regression analysis in order to analyze pathways between groups of variables. I'm attempting to prove Azjen's Theory of Planned Behavior. Basically, in it survey results about Attitudes, Behavioral Control, and Social Norms can be used to predict Intention. And Intention in turn can be used to predict Behavior alongside Behavioral Control.

Example:

.

Anyways, each group (Intentions for instance) consists of 5 questions that have good correlations. When I use SPSS I can't figure out how to determine how much of an effect there is from each determinant - group to the entire Intentions group. I can only figure out how to analyze multiple independent variables and a single dependent variable.

I've tried looking online, but I haven't found much yet. One example had already combined the groups, but they didn't show the descriptives so I couldn't see if they used the mean. My partner and I think it's a quick solution for someone who knows. We've thought about using the means of each group, but that doesn't feel right to me.

Any ideas?

Aegis

Combine the determinants of Intention into an index and while doing that note the Alpha If Item Deleted indicator of each determinant to see how critical each individual variable is to the formation of the Index (ie- if the Alpha improves significantly if you were to delete the item, then the individual determinant is not as critical to the overall Index construction and thus likely measuring something different); then run the regressions of the index and the individual determinants against behaviour, looking at the standardized Beta coefficients to see if you can't draw any comparisons to particular indicator variables that are contributing more than others?

Folken Fanel

If your independent variables are correlated then it'll screw up your model. This comes in the form of individual predictors being significant when they're the only variables in the model, but when you put all of them in the model then none are significant.

What I mean is... sometimes you'll have a model

y = b_0 + b_1x_1 + b_2x_2

where x_1 and x_2 are both insignificant.

So you'll fit the models

y = a_0 + a_1x_1
y = d_0 + d_1x_2

and x_1 and x_2 are significant in their individual models.

Is this what you're talking about?

OtistheGuard

They are significant in their individual models. Basically, all of the items concerning Intentions are well correlated to each other, all of the items of Behavior are well correlated, etc.

Folken Fanel

Well it makes sense that they would be correlated right? Why would someone take a survey and answer in ways that contradict each other? Anyway... there's not too much you can do about collinearity problems like this outside of getting more data.

To be honest though... there's a very specific way to do an analysis like this and it just happens to be one of the things I haven't worked with during my time as a statistics grad student. I suggest you get some statistical consulting done.

Also not to play mod, but there's definitely a rule against asking this forum to do your homework for you. Seriously, seek a professional.

Daenris

You may want to look into structural equation modeling. It should be able to give you results indicating how well the model matches the data, with weights for the strength of each item's contribution to the others.

OtistheGuard

You can post asking for advice, and help, and looking for assistance doing whatever it is you're having trouble with, but threads that are obvious attempts at getting someone to write you a paper are not acceptable.

Not a problem, I'm just asking for general advice. I'm not asking for you to run the regression for me.

Anyways we're gonna be talking to a teacher tomorrow. I was hoping it was a quick fix though.