Help with drawing a mixed linear model, please

[TruthSeeker]

I'm trying to draw a model of the hypotheis underlying a longitudinal study which will test a mixed linear model (latent growth curver) of the relationship between X and Y and how it is mediated/influenced by M

I have been looking but unable to find a clear explanation of the conventions for this type of model. Do I use circles or squares for my outcomes and my predictors? Do I have to show every possible inter-relationship or would it be suffice to not that not all are shown in the figure legend? How about error terms? Do I have to show all of them?

Thanks!

[Simon Bridge]

That's hardly surprising - the description of what you are trying to do includes some pretty steep assumptions about the context it is being read in.

It looks like you are trying to draw a chart to illustrate the hypothesis in question. The rest is far too generic to admit a proper responce.

In general - you model should be complete enough to describe the phenomina you want to study. The conventions and so on that you use need to be to the point and clearly spelled out.

Beyond that, do what you like.

[TruthSeeker]

Thanks Simon.

I have something I am happy with now and which I think is clearly specified. I've sent it to my co-investigators for feedback.

Sorry I didn't go into detail about the variables. I didn't want anyone to do my work. I was just hoping if conventions for these types of models existed someone could point me to them.

Thanks again for your input :)

[lenny]

How about error terms? Do I have to show all of them?
you might want to clearly display both the magnitude of observational error and the residual fitting error. and contrast the two.

which is bigger?

why are you using the particular linear model you are using?

are the results (the values of the model parameters you fit with the data) robust if you leave out at faction of the data and refit? repeatedly? that might give you an idea of how accurate your parameter estimates are, under the assumption that your model stucture is a useful one!

not to suggest that you put all that on one graph...

[TruthSeeker]

These are excellent suggestions! However, this is just a proposal at this point, so I don't have the data to carry them out. I have talked about carrying out analyses such as those you describe once I do get some data. And some funds :) Would be nice to have some funds.

Anyway, the model as proposed is derived from literature review and preliminary pilot data analysis (but this is from a small sample so unstable). The pilot data is really just to show the recruiters that the study is feasible rather than to really test the model.