Stories, Statistics, and Memory

For most decisions, people rely on a myriad of relevant information encountered over the course of days, months or years. Such information come in various forms, including abstract summaries of multiple data points - statistics - and contextualized anecdotes about individual instances - stories. We propose that people form beliefs based on information they recall on the spot, that they do not always retrieve the full wealth of their accumulated information, and that the information type - story versus statistic - is a central determinant of selective memory. In controlled experiments we show that the effect of information on beliefs decays rapidly and exhibits a pronounced story-statistic gap: the average impact of stories on beliefs fades by 33% over the course of a day, but by 73% for statistics. This pattern is driven by the role of context in memory: prompting contextual associations with statistics slows their temporal decay. Guided by a model of similarity and interference in memory, we experimentally examine the explanatory power of a broad swath of different dimensions of interference. Consistent with the model, similarity relationships - rather than, e.g., memory load per se - are the key driving force behind the story-statistic gap.