Ebrahimi Mehr Mehdi, Rad Jamal Amani
Institute for Cognitive and Brain Sciences, Shahid Beheshti University, Tehran, Iran.
Choice Modelling Centre & Institute for Transport Studies, University of Leeds, Leeds LS2 9JT, UK.
Behav Res Methods. 2025 Aug 26;57(10):269. doi: 10.3758/s13428-025-02784-2.
This study critically examines the cognitive and theoretical foundations of the alpha parameter within the Lévy flight model (LFM), an extension of the diffusion decision model (DDM) that incorporates heavy-tailed noise distributions. The alpha parameter, which modulates the tail of these distributions, is assessed for its test-retest reliability - an essential criterion for its classification as a cognitive style measure. Utilizing data from three previous studies, we observed that alpha demonstrates consistent reliability across tasks and time points, supporting its role as a trait-like characteristic. Our observation regarding the interrelations between LFM parameters showed that although most parameters exhibited weak correlations, reflecting their representation of distinct aspects of data, moderate correlations were observed between alpha and both threshold and non-decision time. Furthermore, investigating practice effects, we observed consistent reductions in non-decision time, threshold, and often alpha across sessions, accompanied by a corresponding increase in drift rate in demanding tasks. Notably, alpha showed a strong relationship with the mean reaction time of error responses, indicating its critical role in explaining fast error responses. Additionally, our examination of the predicted decision-time distribution found that lower alpha values correspond to shorter response times in the first quartile of both correct and error responses, highlighting its impact on capturing the dynamics of fast decision-making. Employing the BayesFlow framework for parameter estimation, we evaluated its precision across varying trial counts. These findings offer insights for future research on LFM and similar models.
本研究批判性地审视了 Lévy 飞行模型（LFM）中 α 参数的认知和理论基础，LFM 是扩散决策模型（DDM）的扩展，纳入了重尾噪声分布。调节这些分布尾部的 α 参数，就其重测信度进行了评估——这是将其归类为认知风格测量的一项重要标准。利用来自之前三项研究的数据，我们观察到 α 在不同任务和时间点上表现出一致的信度，支持了其作为一种特质样特征的作用。我们对 LFM 参数之间相互关系的观察表明，尽管大多数参数表现出弱相关性，反映了它们对数据不同方面的表征，但在 α 与阈值和非决策时间之间观察到了中度相关性。此外，在研究练习效应时，我们观察到在各实验环节中，非决策时间、阈值以及通常还有 α 都持续减少，同时在要求较高的任务中漂移率相应增加。值得注意的是，α 与错误反应的平均反应时间显示出很强的关系，表明其在解释快速错误反应方面的关键作用。此外，我们对预测决策时间分布的考察发现，较低的 α 值对应于正确和错误反应的第一个四分位数中较短的反应时间，突出了其对捕捉快速决策动态的影响。采用 BayesFlow 框架进行参数估计，我们评估了其在不同试验次数下的精度。这些发现为未来关于 LFM 和类似模型的研究提供了见解。
