This study is the first to investigate subject-level variability in sociolinguistic evaluative judgements by 30 adult L2 German learners and explore whether the observed variability is characterizable as a function of individual differences in proficiency, exposure, and motivation. Because group-level estimates did not paint an accurate picture of the individual, we propose methods capable of integrating population-level estimates with person- and ensemble-centered approaches so as to reconcile generalizability and individuality. Using random effects from Bayesian mixed-effects models, we found that global subject-level variability in evaluative judgements was not predicted by individual differences. By building homogeneous ensembles (i.e., subgroups of individuals with similar evaluative judgements), however, it was possible to assess whether ensembles were characteristic of certain levels of individual differences. This ensemble-centered approach presents an innovative way to address the group-to-individual generalizability issue in cross-sectional data and transcend individual variability in order to make tentative generalizations of individual cases to wider populations.

本研究首次调查了 30 名成年德语第二语言学习者社会语言评价判断的学科水平变异性，并探讨观察到的变异性是否可以表征为熟练程度、接触程度和动机方面个体差异的函数。由于群体层面的估计并不能准确描绘个体，因此我们提出了能够将群体层面的估计与以个人和整体为中心的方法相结合的方法，以协调普遍性和个体性。利用贝叶斯混合效应模型的随机效应，我们发现评价判断中的全局受试者水平变异性并不是通过个体差异来预测的。然而，通过建立同质整体（即具有相似评价判断的个体亚组），可以评估整体是否具有一定程度的个体差异的特征。这种以整体为中心的方法提供了一种创新的方法来解决横截面数据中群体到个体的普遍性问题，并超越个体差异，以便将个体案例初步推广到更广泛的人群。