Violations of transitive preference can be accounted for by both the noncompensatory lexicographic semiorder heuristic and the compensatory additive difference model. However, the two have not been directly compared. Here, we fully develop a simplified additive difference (SAD) model, which includes a graphical analysis of precisely which parameter values are consistent with adherence to, or violation of, transitive preference, as specified by weak stochastic transitivity (WST) and triangle inequalities (TI). The model is compatible with compensatory, within-dimension evaluation. We also develop a stochastic difference threshold model that also predicts intransitive preferences and encompasses a stochastic lexicographic semiorder model. We apply frequentist methods to compare the goodness of fit of both of these models to Tversky's (1969) data and four replications and Bayes factor methods to determine the strength of evidence for each model. We find that the two methods of analysis converge and that, for two thirds of the participants for whom predictions can be made, one of these models predicting violations of WST has a good and the best fit and has strong Bayesian support relative to an encompassing model. Furthermore, for about 20% of all participants, the SAD model (consistent with violations of WST or TI) is significantly better-fitting and has stronger Bayesian support than the stochastic difference threshold model. Finally, Bayes factor analysis finds strong evidence against transitive models for most participants for whom the SAD model consistent with violation of WST or TI is strongly supported. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

中文翻译：

---

违反传递性偏好：补偿性帐户和非补偿性帐户的比较。


传递偏好的违反可以通过非补偿性词典半序启发式和补偿性加性差异模型来解释。不过，两者尚未直接进行比较。在这里，我们完全开发了一个简化的加性差异（SAD）模型，其中包括对哪些参数值与遵守或违反传递偏好一致的精确图形分析，如弱随机传递性（WST）和三角不等式所指定的（ TI）。该模型与补偿性维度内评估兼容。我们还开发了一个随机差异阈值模型，该模型也可以预测不及物偏好并包含随机词典半序模型。我们应用频率论方法来比较这两个模型与 Tversky（1969）数据的拟合优度，并使用四种重复和贝叶斯因子方法来确定每个模型的证据强度。我们发现这两种分析方法趋于一致，并且对于三分之二的可以进行预测的参与者来说，其中一个预测违反 WST 的模型具有良好且最佳的拟合性，并且相对于包含模型具有强大的贝叶斯支持。此外，对于大约 20% 的参与者来说，SAD 模型（与违反 WST 或 TI 一致）明显比随机差异阈值模型拟合得更好，并且具有更强的贝叶斯支持。最后，贝叶斯因子分析发现了针对大多数参与者的传递模型的有力证据，对于这些参与者来说，与违反 WST 或 TI 一致的 SAD 模型得到了强烈支持。 （PsycInfo 数据库记录 (c) 2024 APA，保留所有权利）。