采样假设背景下的归纳推理的典型性效应
Typical Effects of Inductive Reasoning in the Context of Sampling Assumptions

人们进行类别归纳的程度取决于它们关于观察到的实例是如何从世界上采样的信念。采样假设对归纳推理具有重要的意义，而典型性效应是归纳推理中的一个关键现象。我们采用包含高典型性和低典型性的前提证据，前提证据分别在不同的采样假设情况下呈现：有意的(强采样)或随机的(弱采样)。根据贝叶斯模型理论我们预测，强抽样比弱抽样导致更有限的泛化，有更弱的推理强度。研究结果表明，在两种条件下存在较强的典型性效应，典型性效应依赖于样本证据的假设。我们的研究为归纳推理领域的研究提供了新的视角，人们关于证据如何产生的信念对归纳推理有着很大的影响。
When the subcategories of a category are often mentioned, they can also be called high typical, such as robins among birds, apples among fruits. Those that are less likely to come to mind or are less likely to be recalled are low typical, for example, penguins among birds, jujubes among fruits. The typicality effect of inductive reasoning means that when people infer the possibility of a conclusion in a high typical premise or a low typical premise, the probability of the conclusion in the high typical premise is greater than that in the low typical premise. Inductive reasoning also has its limitations. The data on which we make inferences based on experience are often incomplete and, therefore, the conclusions are inconclusive. However, more and more people believe that similarity alone is not enough to explain attribute induction. Sampling assumption means people’s reasoning depends on how they believe an inductive argument is formed. One must make a key “sampling” assumption about how the available data was generated. Sampling assumption plays an important role in inductive reasoning, and typicality effect is a key phenomenon in inductive reasoning. The purpose of this study is to observe the characteristics of the typicality effect of inductive reasoning under sampling assumption, and to provide further theoretical basis for the study of sampling assumption and typicality. Based on the previous experimental paradigm, the typicality premise and sampling assumption are independent variables. There are two levels of typicality: high typicality and low typicality, and two levels of sampling assumption: deliberate (strong sampling) or random (weak sampling). Inference strength is the dependent variable. Each set of arguments consists of a premise and a conclusion (i.e. premise: “duck”, conclusion: “bird”). The participants were asked to rate the likelihood of the premise’s conclusion being true on a 4-point Likert scale based on the information obtained from the instructions and premise. The results showed that there is a strong typicality effect under both the strong condition and weak condition, and it indicated that the typicality effect depends on the sampling assumption of evidence. We predicted that strong sampling leads to more limited generalization and has weaker inference strength