Making Measurement Model Definition Easier in Multivariate SEM Platform

[Narayanan]

To define the measurement model in the current model specification area, the indicators that belong to a factor must be organized so that they are contiguous in a group. This is rarely the case in many of the datasets we deal with. In a recent attempt to define 13 factors out of 82 indicators where the items were not grouped was not an easy task to hunt and peck these variables. I would like to suggest an option to:

1. Sort the variables in alphabetic order within the specification window
2. Color code the used variables or make them inactive to choose
3. Sort according to a factor prefix letter or something

This suggestion is connected to my previous idea of importing factor definition from an EFA which will make all this so much easier.

[LauraCS]

Another great suggestion, thank you Narayanan! The good news on this one is that we have filters on the "From" and "To" lists that you can use to type a prefix or such and will help a lot in defining the latent variables. Please try it and let us know if it still feels complicated. And do keep the feedback coming =)

[Narayanan]

I did try the different choices. It still feels a little complicated.   We do not name the variables according to factor groupings. So, the choice becomes time consuming.  Is there a way to sort the variable by factor groupings?

[LauraCS]

Hi Narayanan,

Might you be able to list a small subset of variable names that illustrate the complexity so I can understand the issue a bit better? The filter has a number of cool features that can be found by clicking on the little black arrow next to the filter box.

You can even use regular expressions, which enables filtering in elegant ways. However, I understand that if variables aren't named according to factor groupings, this can be complicated. An example of a subset of variables names would help me here to think of a good solution for you.

In this example below, because the sample data table Job Satisfaction.jmp has suffixes for the "Conflict" variables (_C), I can use that in the filter, which then facilitates the selection of these items to create the latent variable:

Best,

~Laura