Limited memory capacity, retrieval constraints and anchoring are central to expectation formation processes. We develop a model of adaptive expectations where individuals are able to store only a finite number of past experiences of a stochastic state variable. Retrieval of these experiences is probabilistic and subject to error. We apply the model to scheduling choices of commuters and demonstrate that memory constraints lead to sub-optimal choices. We analytically and numerically show how memory-based adaptive expectations may substantially increase commuters' willingness-to-pay for reductions in travel time variability, relative to the rational expectations outcome.