Dear PLS community,
I'm doing some analysis on Knowledge Management using SmartPLS. The data was collected by both mail and telephone survey. However, it turns out that responses of some constructs from mail is significantly different from responses from telephone by Kruskal-Wallis test.
How can I control this collection method factor?
One possible way may be doing separate PLS analysis for sample from mail and sample from phone. But the problem is that our sample size is too small to do this.
Therefore, can I just add the collection method factor (1= mail, 0=phone) as a latent variable to the original model? But since there are no theory showing which constructs will be affected by this method factor, shall I just assume all other constructs are affected (i.e. paths from the collection method factor to all other latent variables), or only those constructs significantly different according to K-W tests are affected (paths from the collection method factor to only these constructs)?
Thanks a lot.