Most prior research has focused on characterizing averages in cognition, brain characteristics, or behavior, and attempting to predict differences in these averages among individuals. However, this overwhelming focus on mean levels may leave us with an incomplete picture of what drives individual differences in behavioral phenotypes by ignoring the variability of behavior around an individual's mean. In particular, enhanced white matter (WM) structural microstructure has been hypothesized to support consistent behavioral performance by decreasing Gaussian noise in signal transfer. Conversely, lower indices of WM microstructure are associated with greater within-subject variance in the ability to deploy performance-related resources, especially in clinical populations. We tested a mechanistic account of the "neural noise" hypothesis in a large adult lifespan cohort (Cambridge Centre for Ageing and Neuroscience) with over 2500 adults (ages 18-102; 1508 female; 1173 male; 2681 behavioral sessions; 708 MRI scans) using WM fractional anisotropy to predict mean levels and variability in reaction time performance on a simple behavioral task using a dynamic structural equation model. By modeling robust and reliable individual differences in within-person variability, we found support for a neural noise hypothesis (Kail, 1997), with lower fractional anisotropy predicted individual differences in separable components of behavioral performance estimated using dynamic structural equation model, including slower mean responses and increased variability. These effects remained when including age, suggesting consistent effects of WM microstructure across the adult lifespan unique from concurrent effects of aging. Crucially, we show that variability can be reliably separated from mean performance using advanced modeling tools, enabling tests of distinct hypotheses for each component of performance.

SIGNIFICANCE STATEMENT Human cognitive performance is defined not just by the long-run average, but trial-to-trial variability around that average. However, investigations of cognitive abilities and changes during aging have largely ignored this variability component of behavior. We provide evidence that white matter (WM) microstructure predicts individual differences in mean performance and variability in a sample spanning the adult lifespan (18-102). Unlike prior studies of cognitive performance and variability, we modeled variability directly and distinct from mean performance using a dynamic structural equation model, which allows us to decouple variability from mean performance and other complex features of performance (e.g., autoregression). The effects of WM were robust above the effect of age, highlighting the role of WM in promoting fast and consistent performance.

大多数先前的研究都集中在表征认知、大脑特征或行为的平均值，并试图预测个体之间这些平均值的差异。然而，这种对平均水平的过度关注可能会导致我们无法完整地了解是什么驱动了行为表型的个体差异，而忽略了个体平均水平附近行为的变异性。特别是，假设增强的白质（WM）结构微观结构可以通过减少信号传输中的高斯噪声来支持一致的行为表现。相反，WM 微观结构指数较低与受试者内部署绩效相关资源的能力差异较大相关，尤其是在临床人群中。我们在一个大型成人寿命队列（剑桥衰老和神经科学中心）中测试了“神经噪音”假说的机械解释，该队列有超过 2500 名成年人（18-102 岁；1508 名女性；1173 名男性；2681 次行为治疗；708 次 MRI 扫描）使用 WM 分数各向异性来预测使用动态结构方程模型的简单行为任务的反应时间表现的平均水平和变异性。通过对人内变异性中稳健且可靠的个体差异进行建模，我们发现了对神经噪声假设的支持（凯尔，1997 ），较低的分数各向异性预测了使用动态结构方程模型估计的行为表现的可分离成分的个体差异，包括较慢的平均响应和增加的变异性。当考虑到年龄时，这些影响仍然存在，表明 WM 微观结构在整个成人寿命中具有一致的影响，这与衰老的并发影响不同。 至关重要的是，我们表明使用先进的建模工具可以将变异性与平均绩效可靠地分开，从而能够对绩效的每个组成部分进行不同的假设测试。

意义声明人类认知表现不仅取决于长期平均值，还取决于围绕该平均值的试验间变异性。然而，对认知能力和衰老过程中变化的研究在很大程度上忽略了行为的这种变异性成分。我们提供的证据表明，白质（WM）微观结构可以预测成人寿命样本中平均表现和变异性的个体差异（18-102）。与之前对认知表现和变异性的研究不同，我们使用动态结构方程模型直接对变异性进行建模，并且与平均表现不同，这使我们能够将变异性与平均表现和其他复杂的表现特征（例如自回归）分离。 WM 的影响远远超过年龄的影响，凸显了 WM 在促进快速、一致的表现方面的作用。