Formative vs reflective vs MIMIC-Model and mod.effect

[Hans Tappert]

Hi everybody,

I have a question on how to specify a model because as I think there exist several possible methods in my case.

The structure of my model is as follows:
I have three sub-dimensions of the customer satisfaction, e.g. customer satisfaction with the offer letter or satisfaction with the process of delivery, which are linked to the overall customer satisfaction. The sub-dimensions as well as the overall-satisfaction are measured directly, which means that the customer was asked directly for his satisfaction with certain aspects.
I am now wondering if I should model the sub-dimensions as reflective LV`s with the satisfaction with e.g. the offer letter as the reflective indicator (therefore having only one indicator per construct) or use these sub-dimensions as formative indicators for the overall-satisfaction (therefore having three indicators for the construct "overall satisfaction"). However, using formative indicators would mean that I one the one hand treat "satisfaction" as a LV (overall satisfaction) but at the same time use "satisfaction" (the sub-dimension which are also measures of satisfaction) as indicators, does this really makle sense? Another possible method would be a MIMIC-model, where the overall-satisfaction represents the phantom-variable. I believe that the three sub-dimensions mentioned above are not all dimensions that are necessary to explain the overall satisfaction, but according to my in literature research really ALL formative indicators are required to describe a formative construct. Therefore this MIMIC-model makes more sense in my eyes.

Moreover, it was also asked for the satisfation with the best competitor, over-all-satisfaction as well as satisfaction on the same sub-dimensions. I would like to link this overall-satisfaction with the best competitor as a direct causal link to the satisfaction with the firm I am looking at (this means the better a customer experience with the competition, the worse the satisfaction with the firm I am looking at). Further, I would also like to use this overall-satisfaction with the competitor as a moderating effect that influences the relationship between customer satisfaction with the firm I am looking at and customer loyalty. Is it possible to use this LV "customer satisfaction with the best competitor" twice? I believe so because it is only used to calculate the interaction effect.

However, I am not sure whether to use formative or reflective indicators and how to model the mod. effect. Does somebody have a recommendation for me?
Thank you all very much for your advice.
Best,

Hans