ABSTRACT

The most frustrating outcome of an SEM analysis is nonconvergence. Nonconvergence typically happens when the sample size is small ($N<100$) or very small ($N<50$). To minimize the frequency of nonconvergence, this paper proposes a solution called bounded estimation. The idea is to use data-driven lower and upper bounds for a subset of the model parameters during estimation. In this paper, we provide a rationale to compute these bounds, and we study the effect of different approaches to employ these bounds on the frequency of nonconvergence. A simulation study shows that bounded estimation dramatically decreases the frequency of nonconvergence in both correctly and misspecified models, without any (negative) effects on the quality of the point estimates for the unbounded parameters.

摘要

SEM 分析最令人沮丧的结果是不收敛。不收敛通常发生在样本量很小（$ñ<100$) 或非常小 ($ñ<50$）。为了最小化不收敛的频率，本文提出了一种称为有界估计的解决方案。这个想法是在估计期间对模型参数的子集使用数据驱动的下限和上限。在本文中，我们提供了计算这些界限的基本原理，并研究了采用这些界限的不同方法对不收敛频率的影响。一项模拟研究表明，有界估计显着降低了正确和错误指定模型中的不收敛频率，而对无界参数的点估计质量没有任何（负面）影响。