Indeed, as you've probably noticed, the RLH computation is different when chips are involved, and much lower than standard discrete choice.
The RLH norms will depend on how much respondents tend to push the chips all onto one concept versus dividing them more evenly.
Maybe some users out there who have constant sum datasets can give you some norms from their observations. I don't deal with chip allocation much at all, as my projects are all discrete choice.
