Management research increasingly recognizes omitted variables as a primary source of endogeneity that can induce bias in empirical estimation. Methodological scholarship on the topic overwhelmingly advocates for empirical researchers to employ two-stage instrumental variable modeling, a recommendation we approach with trepidation given the challenges associated with this analytic procedure. Over the course of two studies, we leverage a statistical technique called the impact threshold of a confounding variable (ITCV) to better conceptualize what types of omitted variables might actually bias causal inference and whether they have appeared to do so in published management research. In Study 1, we apply the ITCV to published studies and find that a majority of the causal inference is unlikely biased from omitted variables. In Study 2, we respecify an influential simulation on endogeneity and determine that only the most pervasive omitted variables appear to substantively impact causal inference. Our simulation also reveals that only the strongest instruments (perhaps unrealistically strong) attenuate bias in meaningful ways. Taken together, we offer guidelines for how scholars can conceptualize omitted variables in their research, provide a practical approach that balances the tradeoffs associated with instrumental variable models, and comprehensively describe how to implement the ITCV technique.

管理学研究越来越认识到遗漏变量是内生性的主要来源，它可以在经验估计中引起偏差。有关该主题的方法学研究绝大多数都建议经验研究者采用两阶段工具变量建模，鉴于与该分析程序相关的挑战，我们建议对此采取谨慎的态度。在两项研究过程中，我们利用一种称为混杂变量影响阈值（ITCV）的统计技术，更好地概念化了哪些类型的遗漏变量可能实际上导致因果推理偏倚，以及它们是否已在已发表的管理研究中出现。在研究1中，我们将ITCV应用于已发表的研究，发现大多数因果推论不太可能因遗漏变量而产生偏差。在研究2中，我们对内生性重新进行了有影响的模拟，并确定只有最普遍的被忽略变量似乎才对因果推理产生实质性影响。我们的模拟还表明，只有最强大的工具（也许不切实际地强大）才能以有意义的方式衰减偏差。总而言之，我们为学者如何在研究中概念化遗漏变量提供了指南，提供了一种实用的方法来平衡与工具变量模型相关的取舍，并全面描述了如何实施ITCV技术。