Is asking students to generate predictions an effective technique to induce meaningful cognitive conflict and to facilitate conceptual change?

科学教学是具有挑战性的，因为它需要改变学生持久和普遍的误解。最近的研究表明，产生预测是一种很有前途的策略，有助于儿童修正错误观念。当儿童事先做出明确的预测时，这种促进作用与对违反期望的结果的增强的生理惊讶反应有关。意外反应的程度进一步表明，这取决于参与者对其预测的信心以及预测误差的大小。第二个项目阶段的首要目标是更精确地捕捉对惊喜产生预测的效果以及随后对误解的修正。为此，我们将建立每个孩子的先验信念的正式模型，并将他们的生理惊讶反应和信念修正与最优贝叶斯学习者进行比较。贝叶斯推理框架对于这个目的特别有用，因为它提供了置信度、预测误差和信念修正之间相互作用的正式模型。它精确地告诉我们，根据概率论，违反预期的结果应该如何用于更新信念和模型（即，理性归纳推理）。因此，我们可以测试a)通过明确预测使用先前模型的儿童是否比事先不使用先前模型的儿童更好地（即更接近统计最佳）使用相互矛盾的证据，以及b)信心和预测误差在这种好处的发生中发挥了什么作用。

Teaching science is challenging because it entails changing persistent and pervasive misconceptions in students. Recent research suggests that generating predictions is a promising strategy that facilitates children’s revision of misconceptions. This facilitation has been found to be related to an enhanced physiological surprise response to expectancy-violating outcomes when children generated an explicit prediction beforehand. The magnitude of the surprise response has further been shown to depend on participants’ confidence in their prediction as well as on the size of the prediction error. The overarching goal of the second project phase is to more precisely capture the effect of generating predictions on surprise and the subsequent revision of misconceptions. To this end, we will build a formal model of each child’s prior beliefs and compare their physiological surprise response and belief revision to an optimal Bayesian learner. The Bayesian inference framework is particularly useful for this purpose because it offers a formal model of the interplay between confidence, prediction error, and belief revision. It tells us precisely how, according to probability theory, expectancy-violating outcomes should be used to update beliefs and models (i.e., rational inductive inference). We can thus test a) whether children who engage their prior models by making an explicit prediction make better (i.e., closer to statistically optimal) use of conflicting evidence than children who do not engage their prior models beforehand, and b) what roles confidence and prediction error play for this benefit to occur.