Psychological research has demonstrated that people's preferences are not constant and that decision makers often depart from predictions made by so called `rational` models. Most decision making theories, however, either make deterministic predictions to address departures from rational principles, or handle variability of behavior with stochastic models that cannot account for such violations. This research examines a new, stochastic model of decision making, the proportional difference model (PD). PD is a simple, two-parameter model, based on the notion that decision makers trade attribute values in a proportional manner and that this trading is a variable process. A number of experiments will be conducted to examine the predictions of the PD model. The first study will test PD by comparing its descriptive ability to that of three other models found in the literature; preliminary tests suggest PD is a better model. The second study will focus on PD's predictions of context effects, such as effects of maximum attribute values in the choice environment and the known reflection effect. The third study will investigate decision making with vague information. The results of this research will advance the state of knowledge of descriptive decision making.