Posterior Separable Cost of Information
给出了信息获取成本等于不确定性度量（如熵）期望减少的可检验条件，即后验可分离性，并展示了如何从可观测的选择行为中非参数地识别和估计信息成本。
We provide testable conditions under which the cost of acquiring information is given by the expected reduction of a measure of uncertainty (e.g., entropy). The assumption, under the name of posterior separability, is nearly universal in the literature of rational inattention; yet, a testable characterization has been lacking. In applications to experimental data, we indicate situations in which posterior separability is—and is not—a compelling assumption for the cost of information; we propose a generalization to address some of its shortcomings. We also show how to identify and estimate nonparametrically the cost of information from observable choice behavior.