I agree with @ Cordula Hinkes. Pepsi vs Coke, and 7up vs Sprite or even Mirinda vs Fanta. Keeping in view that Coke, Sprite and Fanta and Pepsi, 7up and Fanta not only belong to two different brands, Coke and Pepsi, respectively, but they don't normally compete with each other. For example, Coke competes Pepsi, but not 7up and Mirinda, simply because of huge differences in their flavour as @ Cordula Hinkes clrearly mentioned.

Given this scenario, we need to rely on

generic attributes for each alternative using a labelled discrete choice experiment, and I envisage respondents' (or consumers) preferences will still dominate their mostly preferred flavour. In such a case, preference heterogeneity will be compromised, and the modelling will only be confined to the MNL model, not RPL and LCM models, I fear !

In some situations, even for the sake of flavour, say Coke at McDonalds and Pepsi at KFC, or vice-versa, consumers are forced or psychologically in a predetermined situation ! . . So, what will be the case if I want to have KFC and Coke ! I don't get Coke at KFC! The similar situation exists for other flavours.

Hope this helps.

1. Even, these different flavours are not substitutes of each other. So, Poe et al test for Marginal Rate of Substitution (MRS) will surely exhibit highly significant differences I envisage.

2. Consumers, say Coke lovers, are sometimes even forced to choose non-substitutes like Sprite, 7up, Mirinda or Fanta !

3. So, only possibility in this specific situation is:

(i) Use labelled discrete choice experiment using varying alternatives but with two substitutes and two non-substitutes, say,

Choice set 1

Alternatives: Coke, Pepsi, Sprite, Mirinda

Choice set 2

Alternatives: Sprite, 7up, Pepsi, Fanta

Choice set 3

Alternatives: Fanta, Mirinda, 7up, Coke

Like this, just go on . . .

(ii) Given the above situation, you need generic attribute for Coke and Pepsi, Sprite and 7up, and Fanta and Mirinda, respectively.

I fully confident that if you apply the methodology I suggested above you can capture preference heterogeneity using RPL and LCM models.

Hope this will help.

One more thing. And, also ad an opt-out (or none of these flavours) alternatives simply because consumers are sometimes forced to drink Pepsi if Coke is not available. And, literature review suggests that there's always an opt-out situation in reality.

Good luck and best wishes