I need someone who can help me with structural equation modelling in AMOS.

...

Anyone?

looks like I'll have to become an expert myself then (mince)

Hi Sneaks, Is it that you need advice on how to use AMOS, how to do SEM, or both? I use SEM but in Mplus software. However, the statistician in my department runs public statistical training courses that may be useful for you (I help teach in them). In the past we ran one on EFA/CFA, and the CFA component was done in AMOS. CFA is the pre-cursor to SEM, the only difference being that the first has only correlational paths (i.e., associations between constructs), and the second integrates structural paths (i.e., one way, predictive). He *also* runs a course on how to do SEM in Mplus. If you are interested in these, have a look at this: http://www.offbeat.group.shef.ac.uk/FIO/trainingcourses.htm 8-)

Well I'm now not sure if I even need AMOS. I'm so confused!

I basically want to do a moderating regression - fine. But my moderator is categorical, with 3 categories.

So I was going to regress it on to the outcome separately for each group and then compare the coefficients. but then I find that if the measurement errors for the predictor are different for each group (which they are), then I shouldn't do this and I should use SEM.

Now I'm stuck :-(

No, you don't need to use SEM. SEM is good for when you have a complex model that includes multiple mediators and multiple outcomes, and when you want to use latent rather than observed variables. I think you can do what you want to in SPSS.

If your categorical moderator is ordinal, you can assume it is continuous for the purpose of the analysis. If your categorical moderator is discrete (e.g., completely different groups such as political preferences, locations) which it sounds like it is, you need to dummy code your categories.

Do a search for categorical moderation in Google - there's plenty of useful stuff on there. There are even books dedicated to the subject.

:-( I found one book, but can't see all the pages.

If I dummy code it though, how will I understand the moderation? Usually I would create an interaction term e.g. IV x moderator. and if the interaction was significant then I would say there's a moderation .

Would you create two interaction terms (for both dummy variables - assuming the categorical moderator has 3 levels)?

Yes. You can do this with GLM Anova in SPSS. If you have three categories, two of the interactions need to go in the model, and the reference category is left out. I'm afraid I don't have time to go into detail (I'm handing in my Thesis on Tuesday and have mad-crazy amounts of work to do!) but follow the links to the courses I posted; on here are details of the statistician. He may be able to help (for a fee). The Regression course we run also covers how to examined moderated regression with categorical moderators. Sorry I can't be of more help right now.

thanks Charls!