hi, a bit of an intro question, can one look at the variance-covariance matrix to see which variables correlate, then using only those which correlate, can you then hypothesize a model to test based on which variables correlate. I ask as I have a set of variables that I think will correlate but I'm not sure, so can I 'throw' out those which don't when building my hypothesized model. Thanks. jmck.
This is an analysis with an already collected data set that was designed to answer a different question than the one I am proposing. I have my hypotheses regarding which variables I think will correlate and subsequent model to test based on that, but what if some do not correlate at the outset obviously my original model was off, but can I refine my original model based on the results? Or am I using the data to drive my analyses vs. theory. Isn't the former the process involved in model modification using modification indeces? Thanks! JMCK
If you have a theory, then the variables you need in the data set are defined by that theory. If the data do not support the theory, then you have answered your research question. You can use modification indices to see where the misfit of the model lies but adding parameters should be based on theory not simply to improve model fit improvement.
Xiaolu Zhou posted on Wednesday, November 23, 2011 - 1:54 pm
I have two questions:
1. my data is binary, but one of the item showed 4 categories under "SUMMARY OF CATEGORICAL DATA PROPORTIONS", the other items are all have 2 categories. This 4 categories ends up creating extra threshold statistics. I went back and checked the original data, no problem with coding. Why this happened?
2.Is there a way way to link C to both A and B, but set the predicted association of C and B at 0? I tried C ON B@0. But it did not work.