Rescorla (2001) used the compound test procedure to compare associative changes to cues located at different points on a performance scale. He found that associative changes to cues conditioned in compound are not necessarily equal, as predicted by common error term theories like Rescorla and Wagner (1972), but instead are larger for the poorer predictor of a trial outcome. Hence, Rescorla proposed a modification to the Rescorla-Wagner model whereby associative change is calculated as the product of 2 error terms: a common error term, as in the original model, and a unique error term for each cue present, which accounts for his findings that the poorer predictor of a trial outcome undergoes more associative change. In a recent study, Spicer, Mitchell, Wills, and Jones (2020) reported findings that appear to be inconsistent with Rescorla's proposal. These authors compared associative changes to cues that differed in associative strength as well as the certainty with which they predicted a trial outcome: One cue had greater strength than did the other, but its prediction of the trial outcome was less certain. Spicer et al. found that the cue that evoked a larger prediction error (the more certain cue) underwent less (not more) associative change and, thereby, concluded that associative change in people is not primarily determined by prediction error. Instead, they argued that cues that predict certain outcomes are somewhat protected from further associative change (theory protection), resulting in greater change to cues that predict uncertain outcomes. In this article, we offer an alternative explanation for the Spicer et al. findings using an approach described by Holmes, Chan, and Westbrook (2019). We show that if the learning-to-performance mapping function is a double sigmoid across the full range of associative strength, the Rescorla-Wagner model accommodates Rescorla's compound test results, as well as those reported by Spicer et al. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

中文翻译：

---

保护 Rescorla-Wagner (1972) 理论：对 Spicer 等人的回复。(2020)。

Rescorla (2001) 使用复合测试程序来比较与位于表现量表上不同点的线索的关联变化。他发现，如 Rescorla 和 Wagner (1972) 等常见错误术语理论所预测的那样，复合条件下线索的关联变化不一定相等，但对于较差的试验结果预测因子，关联变化更大。因此，Rescorla 提出了对 Rescorla-Wagner 模型的修改，其中关联变化被计算为 2 个误差项的乘积：一个常见的误差项，如原始模型中，以及存在的每个线索的唯一误差项，它解释了他的研究结果表明，对试验结果的较差预测会经历更多的关联变化。在最近的一项研究中，Spicer、Mitchell、Wills 和 Jones（2020 年）报告的发现似乎与 Rescorla'不一致 的提议。这些作者将联想变化与联想强度不同的线索以及他们预测试验结果的确定性进行了比较：一个线索比另一个线索具有更大的强度，但它对试验结果的预测不太确定。斯派塞等人。发现引起较大预测误差的线索（越确定的线索）经历较少（不是更多）的联想变化，从而得出结论，人们的联想变化主要不是由预测误差决定的。相反，他们认为预测某些结果的线索在某种程度上受到保护，不会受到进一步的联想变化（理论保护），从而导致预测不确定结果的线索发生更大的变化。在本文中，我们为 Spicer 等人提供了另一种解释。使用 Holmes、Chan 描述的方法的发现，和威斯布鲁克（2019 年）。我们表明，如果学习到绩效的映射函数是整个联想强度范围内的双 sigmoid，则 Rescorla-Wagner 模型可容纳 Rescorla 的复合测试结果，以及 Spicer 等人报告的结果。（PsycInfo 数据库记录 (c) 2021 APA，保留所有权利）。