Anonymous posted on Tuesday, February 17, 2015 - 9:07 am
Hello. I am trying to model the effect of an intervention on a latent construct, assessed before and after intervention. The data has two levels, pupils nested within schools, with randomisation done at the school level. All indicators/factors are continuous, for simplicity.
I can model this as a two level SEM to estimate the treatment effect, however I would like to model this as a multiple group model, group 1 being the control group and group 2 the intervention, so that I can assess invariance of all loadings/intercepts. However I note that means/intercepts of the factors are only available at the between level. Is it appropriate to estimate the treatment effect at this level? Would it be wrong to do so if the indicators were categorical? I am using the grouping keyword/option.
I would also like to assess treatment effects further using a within level variable eg. pupil gender, to see if these vary by genders. Is it appropriate to split the data into four groups eg male control/female control or should it be modelled along the lines of Webnote 16? Thank you in advance for any comments you have on this matter.
Q1. Yes, you should estimate the treatment effect in the latent variable means on Between, also for categorical items. Your randomization is on the Between level so this is appropriate also from that point of view.