Type IV hierarchical construct

[ying_cheng]

Dear all

I hope the answers to the questions could help many of us. Thank you for your time.

I attempt to set up a 2nd order formative construct (with no measurement items), with two 1st order formative constructs with their own measurement items. This Type IV hierarchical construct (without measurement items, I emphasize again) is hypothesized to lead to another latent variable (with measurement items).

Literature and the discussions here suggested four solutions to set up the 2nd order construct, which are:
1) repeated indicator (Chin, et al., 2003; Reinartz, Krafft and Hoyer, 2004)
2) "run a Varimax rotation and use the scores of the both components as formative indicators of the 2nd order LV", see here: viewtopic.php?t=660&postdays=0&postorder=asc&start=0
3) "run the principal components analysis and extract just one component, save its score to use in SmartPLS. also see the former link
4) "use the average of the indicators be a better solution (if the indicators are measured in different scales, you should standardized them before the average). also the former link

My questions include,
1. first of all, I remembered that the 1) approach is applicable ONLY for reflective indicators (1st order). However unfortunately I cannot find the literature again, hope someone can help: Do I remember right? and in which literature this issue has been discussed? If I remember right, then the methods adopted by Reinartz, Krafft and Hoyer, 2004 is inappropriate, right?

2. the approach 2), never seem it before. Can someone gave a more specific explanation? Or literature?

3. I don't understand how the approach 3) works. again, can someone gave a more specific explanation? Or example, or literature?

4. Anyone used the 4) approach before? Again, literature or example, or explanation?

Overall I am quite confused about what to do now. I hope that any of you can help, furthermore:
5. Given my model, how can I proceed to set up the hierarchical construct?
6. how can I test the multicollinearity among the two 1st order formative construct? Any literature?
7. Should I take the two steps approach also? see Yi and Davis (2003). Does SmartPLS allows the co-appearance of the hierarchical construct, and the following casual relationships in a same graph? Any literature?

Any suggestions or leads would be highly appreciated. Many thanks

Ying