Hi,
I need help with the following problem. It would be great if anyone can help me.
I want to show in my sample (303 datas) that two groups differ by their influence. For this i will use the multiple group analysis by a categorial variable. (moderating effect)
For comparison I have to create a structural equation model which fits for the whole sample.
But in this sample the dimension of the two groups isn´t equal. The one group has 231 datas the other one only 72 datas.
When i modify now the strucural equation model for the whole sample there will be a bigger influence by the bigger group and the smaller group will be under represented in the modell. How shell I handle this Problem?
Thanks for your comments
Fabian
multiple group analysis by categorial Variable
Alternative to multigroup analysis
Hi Panfabi,
From your description it sounds like you intend running a MANOVA. Bagozzi et al., (1991) provide guidance on modeling experimental designs in PLS. The multigroup analysis should overcome the heterogeneity of variance across the two groups, but you will need to watch out for the statistical power of the effects for the smaller group. The following forum provides more detail:
viewtopic.php?t=152&highlight=shipley
I have also been considering a related problem... In the event that the assumptions are met (i.e., Box's M, Levene's and Bartlett's tests all non-significant), is it possible to avoid running separate group models by including a dummy variable for each group. In such cases, would it be possible to include interaction affects between the dummy variables?
Can someone please clarify if this is incorrect!
Regards,
Byron.
From your description it sounds like you intend running a MANOVA. Bagozzi et al., (1991) provide guidance on modeling experimental designs in PLS. The multigroup analysis should overcome the heterogeneity of variance across the two groups, but you will need to watch out for the statistical power of the effects for the smaller group. The following forum provides more detail:
viewtopic.php?t=152&highlight=shipley
I have also been considering a related problem... In the event that the assumptions are met (i.e., Box's M, Levene's and Bartlett's tests all non-significant), is it possible to avoid running separate group models by including a dummy variable for each group. In such cases, would it be possible to include interaction affects between the dummy variables?
Can someone please clarify if this is incorrect!
Regards,
Byron.