Dear all,
I have a question concerning the use of formative LVs as endogenous resp. mediating variables. I have followed the previous discussion on this topic which is scarce and discourages the use of formative constructs as endogenous LV. (viewtopic.php?t=660)
Wiley (2005) states that there's "no mechanism by which an antecedent variable can influence a formative index". But what if you don't use an index but several indicators?
I intent to model a Type II second order construct (1st order indicators are reflective, 2nd order indicators are formative) as a mediating latent variable, let's call it VALUE. There are 5 formative indicators (LV Scores) to measure VALUE, so I do not use an index.
I did two steps:
1. Estimated a model in which VALUE is an exogenous variable only (influencing one endogenous LV called INTENT); as a result I get the factor weights between VALUE and its five 1st order indicators plus a plausible and significant path coefficient between VALUE and INTENT.
2. Estimated a model in which VALUE is an exogenous and endogenous variable (=mediating). The factor weights between VALUE and its five 1st order indicators change slightly compared to Step 1, the resulting path coefficients between VALUE and its antecedent LVs are plausible and significant. The path coefficient between VALUE and INTENT remains the same. SmartPLS calculates an acceptable R Square value for VALUE. So obviously, something "happens" when I use VALUE as a mediating variable.
Can anybody give some advice if I can proceed that way or if it were completely false?
I'd really appreciate your thoughts on this matter.
Thank you.
Kati