Dear All,
I came to the understanding from replies to a previous question (thank you Hengkov!) and from further communications (with Dr. Wynne Chin) that the repeated indicators for second-order factors should be connected in the same direction/mode as the first-order factors. So, if the 1st-order LV's indicators are reflective (Mode A), you need to connect the repeated indicators to the second-order LV as reflective also, EVEN if the first-order LV is connected as formative to the the second-order LV.
Now, the question is: what if you have a 1st-order LV that is formative (and connected to the 2nd-order LV as formative) while another 1st-order LV is reflective, and connected to the same 2nd-order LV as reflective? In that case, what direction should we use for the 2nd-order indicators? I am sorry if this looks like a puzzle, but I am actually facing this situation in my thesis model!
Thank you in advance for your advice.
Regards,
Moataz Soliman
Direction of repeated indicators when first-order LVs mixed!
Hi,
It is true that you can have a mixture of indicators types in the measurement model. But when it comes to hierarchial constructs, I think you should first start by checking difference for why one first order latent variable is reflective and the other is formative. I think they should be all the same if they are forming one concept (the global construct) and therefore they should match.
I am not an expert my self, but that's what comes to my mind based on my humble understanding of PLS application!
Good luck in finding your answer!
It is true that you can have a mixture of indicators types in the measurement model. But when it comes to hierarchial constructs, I think you should first start by checking difference for why one first order latent variable is reflective and the other is formative. I think they should be all the same if they are forming one concept (the global construct) and therefore they should match.
I am not an expert my self, but that's what comes to my mind based on my humble understanding of PLS application!
Good luck in finding your answer!
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hallo Moataz,
Hmm, your case is special issue.;-)
I know one solution for this.
Repeated not appropriate for this problem (because your construct first-order is MIXED => formative and reflective, so can't repeated).
For alternative method and "new idea" you can used Two-Step Approach.
Step 1 => you run algorithm PLS for estimate only first order construct and save latent variable score.
Step 2 => Input LVs for indicators second-order.
I'm test this procedure using WarpPLS program and success, and I think SmartPLS too.
Regards,
Hengky
Hmm, your case is special issue.;-)
I know one solution for this.
Repeated not appropriate for this problem (because your construct first-order is MIXED => formative and reflective, so can't repeated).
For alternative method and "new idea" you can used Two-Step Approach.
Step 1 => you run algorithm PLS for estimate only first order construct and save latent variable score.
Step 2 => Input LVs for indicators second-order.
I'm test this procedure using WarpPLS program and success, and I think SmartPLS too.
Regards,
Hengky
Hello Shatha and Hengky,
I thank you both for your replies, and I will comment briefly on each!
For Shatha, thank you for your insights. Please let me add, though, that it is possible to have "mixed" models -formative and reflective. Jarvis et al. (2003, p. 204) mention that "indicators. Mixed models could result either because some of the first-order dimensions are formative indicators of the second-order construct and some are reflective indicators of the second-order construct or because some of the first-order dimensions themselves have formative indicators and some have reflective indicators."
Jarvis, C. B., MacKenzie, S. B., and Podsakoff, P. M. “A Critical Review of Construct Indicators and Measurement Model Misspecification in Marketing and Consumer Research,” Journal of Consumer Research (30), 2003, 199-218.
For Hengky, thanks a lot for your advice. I actually did use the two-step approach, especially that the multidimensional factor I am referring to is an endogenous one (so if I use repeated indicators, the R2 will be artificially too high!) However, when I am now using the LV score as indicators, you still can let them point to the factor (Mode B) or away from it (Mode A), and when you have mixed 1st-order factors, it's tricky to decide which mode to use :-)
Regards,
Moataz
I thank you both for your replies, and I will comment briefly on each!
For Shatha, thank you for your insights. Please let me add, though, that it is possible to have "mixed" models -formative and reflective. Jarvis et al. (2003, p. 204) mention that "indicators. Mixed models could result either because some of the first-order dimensions are formative indicators of the second-order construct and some are reflective indicators of the second-order construct or because some of the first-order dimensions themselves have formative indicators and some have reflective indicators."
Jarvis, C. B., MacKenzie, S. B., and Podsakoff, P. M. “A Critical Review of Construct Indicators and Measurement Model Misspecification in Marketing and Consumer Research,” Journal of Consumer Research (30), 2003, 199-218.
For Hengky, thanks a lot for your advice. I actually did use the two-step approach, especially that the multidimensional factor I am referring to is an endogenous one (so if I use repeated indicators, the R2 will be artificially too high!) However, when I am now using the LV score as indicators, you still can let them point to the factor (Mode B) or away from it (Mode A), and when you have mixed 1st-order factors, it's tricky to decide which mode to use :-)
Regards,
Moataz
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi, Moataz
I think used mode B (because 1st-order mixed, formative-reflective) and 2st-order construct have direction formative => construct dimension to 2st-order (Type II and Type IV).
For evaluation construct dimension to 2st-order look significance weight (so, mode B).
For mode A => if all 1st-order construct is reflective and 2st-order direction Type I and III. :-)
This procedure same with Two-Stage interaction (formative-reflective)
Regards,
Hengky
I think used mode B (because 1st-order mixed, formative-reflective) and 2st-order construct have direction formative => construct dimension to 2st-order (Type II and Type IV).
For evaluation construct dimension to 2st-order look significance weight (so, mode B).
For mode A => if all 1st-order construct is reflective and 2st-order direction Type I and III. :-)
This procedure same with Two-Stage interaction (formative-reflective)
Regards,
Hengky
Last edited by Hengkov on Mon Apr 30, 2012 11:27 pm, edited 1 time in total.