This research studies a model for voter preference change based on two normative principles. First, justification: the voters must be able to justify their preference changes by making qualities of the alternatives relevant or irrelevant to their decisions. Second, consistency: voters' consecutive preference shifts must stem from consistent deliberations, based on awareness, regarding which attributes to deem relevant or irrelevant for future choices. Justification entails that at any time preference changes if and only if some attributes become relevant or irrelevant. Consistency ensures that if an attribute is made relevant and then irrelevant again, this change must be justified by alterations in the relevance of other attributes meanwhile. These two principles are shown to be necessary and sufficient to rationalize preference changes by the maximization of an ordering on the preferences themselves, named a meta-preference. The alternatives are defined by their attributes and the successive preferences are known, as well as a stable ordering the alternatives attributes, so that it is possible to determine the attributes that affect the voter's preference at each period -- defined as the relevant attributes.It is shown that this model is identifiable, empirically testable and can generate the polarization of political preferences among ex-ante identical voters.