Yes, when aggregating preferences in one dimension using the arithmetic mean, people with a more divergent preference have relatively more impact.
> (I would argue we should always normalize the cast vote so that most wanted candidate gets 100 and the least wanted 0, ...
Yes, all other things equal, this appears like a reasonable thing to do.*
> but then we take away people's ability to express only mild interest.) And through that, they can in certain cases manipulate the outcome.
Right. You've given a clear example of the complexity of balancing desirable properties in voting systems.
* The problem of course is that there are many reasonable properties that most reasonable people would want, and yet, they don't coexist well.
I don't agree with the claim "it doesn't lose any voter information"; unless voter preferences are aligned, any preference aggregation method must lose information. (More broadly, this happens with any statistic.)
You probably know this, so I think you meant something slight different? Can you elaborate?
My interpretation, not sure if correct, is that basically you're expressing each vote as a point in d-dimensional Euclid space (where d is number of choices), and the result of the vote is the center of mass of all these points. Stated like this it perhaps makes more sense than just linear range voting.
Yes, you can argue that the more extremely somebody puts the ranges, they have an advantage. (I would argue we should always normalize the cast vote so that most wanted candidate gets 100 and the least wanted 0, but then we take away people's ability to express only mild interest.) And through that, they can in certain cases manipulate the outcome.
I feel like we cannot really meaningfully say that there is a difference between "real" preferences and the "manipulated" ones. For example, this article (from my cursory reading) has a model how media (or others) convince somebody to vote for B when his "real" preference is A. But how can we claim that he truly didn't change his "real" preference to vote for B, based on the media discussion?
Ultimately, I find it somewhat patronizing to claim that, if the real preferences are (internally in humans) expressed on numerical scale 0-100, and we ask for the number from that numerical scale 0-100, this number can somehow not correspond to real preference. I think there is then a contradiction in the definition of "preference", because then the preference has to be something more complicated, like some condition based on other people's choice, not a simple internal value that is independent of what other people want.