Christensen Alexander P, Golino Hudson, Abad Francisco J, Garrido Luis Eduardo
Department of Psychology and Human Development, Vanderbilt University, Nashville, TN, 37203, USA.
University of Virginia, Charlottesville, VA, USA.
Behav Res Methods. 2025 Mar 14;57(4):114. doi: 10.3758/s13428-025-02640-3.
Psychometric assessment is the foundation of psychological research, where the accuracy of outcomes and their interpretations depend on measurement. Due to the widespread application of factor models, factor loadings are fundamental to modern psychometric assessment. Recent advances in network psychometrics introduced network loadings which aim to provide network models with a metric similar to factor loadings to assess measurement quality when the data are generated from a factor model. Our study revisits and refines the original network loadings to account for properties of (regularized) partial correlation networks, such as the reduction of partial correlation size as the number of variables increase, that were not considered previously. Using a simulation study, the revised network loadings demonstrated greater congruence with the simulated factor loadings across conditions relative to the original formulation. The simulation also evaluated how well correlations between factors can be captured by scores estimated with network loadings. The results show that not only can these network scores adequately estimate the simulated correlations between factors, they can do so without the need for rotation, a standard requirement for factor loadings. The consequence is that researchers do not need to choose a rotation with the revised network loadings, reducing the analytic degrees of freedom and eliminating this common source of variability in factor analysis. We discuss the interpretation of network loadings when data are believed to be generated from a network model and how they may fit into a network theory of measurement.
心理测量评估是心理学研究的基础，其结果的准确性及其解释取决于测量。由于因子模型的广泛应用，因子载荷是现代心理测量评估的基础。网络心理测量学的最新进展引入了网络载荷，旨在为网络模型提供一种类似于因子载荷的度量，以便在数据由因子模型生成时评估测量质量。我们的研究重新审视并完善了原始的网络载荷，以考虑（正则化）偏相关网络的特性，例如随着变量数量增加偏相关大小的减小，而这些特性以前并未被考虑。通过模拟研究，相对于原始公式，修订后的网络载荷在各种条件下与模拟的因子载荷表现出更高的一致性。该模拟还评估了用网络载荷估计的分数能够多好地捕捉因子之间的相关性。结果表明，这些网络分数不仅能够充分估计模拟的因子之间的相关性，而且无需旋转（这是因子载荷的标准要求）就能做到这一点。其结果是，研究人员在使用修订后的网络载荷时无需选择旋转，从而减少了分析自由度并消除了因子分析中这种常见的变异性来源。我们讨论了在数据被认为由网络模型生成时网络载荷的解释，以及它们如何可能融入测量的网络理论。
