Abstract. Introspection and abundant field and lab evidence indicate that people often infer population characteristics from the data available. When people lack records and are not given feedback, however, they might underweight some of such data, perhaps due to inattention or memory failures. In this paper we use lab experiments and a parsimonious analytical framework to explore inference when limited attention and memory lead to (extreme) data underweighting, the relevance of such phenomenon, and whether people exhibit different patterns of underweighting. In our experiment, subjects arguably face a simple problem of inference. Even so, we find statistically significant evidence of underweighting. Yet subjects are heterogeneous: some infer as if they underweighted very little while others seem to underweight a lot of the signals; these patterns correlate with the subject's CRT score, an individual measure of reflectiveness. Further, our classification analysis also suggests that some subjects exhibit primacy effects in their patterns of underweighting, while others are better characterized by recency effects. When subjects get some experience and their payoffs are non-ambiguous, further, we find that underweighting is reduced, but only if the inference problem is simple enough.

Keywords: Beliefs, Biases, Inference, Inattention, Limited recall, Underweighting