In this paper, we generalize the model of Quiggin and Chambers (2004) to allow for ambiguity, and derive conditions, referred to as generalized invariance, under which a two argument representation of preferences may be obtained independent of the existence of a unique probability measure. The first of these two arguments inherits the properties of standard means, namely, that they are upper semi-continuous, translatable and positively linearly homogeneous. But instead of being additive, these generalized means are superadditive. Superadditivity allows for means that are computed (conservatively) with respect to a set of prior probability measures rather than a singleton probability measure. The second argument of the preference structure is a further generalization of the risk index derived in Quiggin and Chambers (2004). It is sublinear in deviations from the generalized mean discussed above.