Conflict-related sexual violence is an international security problem and is sometimes used as a weapon of war. It is also a complex and hard-to-observe phenomenon, constituting perhaps one of the most hidden forms of wartime violence. Latent variable models (LVM) offer a promising avenue to account for differences in observed measures. Three annual human rights sources report on the sexual violence practices of armed conflict actors around the world since 1989 and were coded into ordinal indicators of conflict-year prevalence. Because information diverges significantly across these measures, we currently have a poor scientific understanding with regard to trends and patterns of the problem. In this article, we use an LVM approach to leverage information across multiple indicators of wartime sexual violence to estimate its true extent, to express uncertainty in the form of a credible interval, and to account for temporal trends in the underlying data. We argue that a dynamic LVM parametrization constitutes the best fit in this context. It outperforms a static latent variable model, as well as analysis of observed indicators. Based on our findings, we argue that an LVM approach currently constitutes the best practice for this line of inquiry and conclude with suggestions for future research.

中文翻译：

---

衡量战时性暴力的潜在变量方法

与冲突有关的性暴力是一个国际安全问题，有时被用作战争武器。这也是一种复杂且难以观察的现象，可能是战时暴力最隐蔽的形式之一。潜在变量模型 (LVM) 提供了一种有前景的途径来解释观察到的测量值的差异。三个年度人权来源报告了自 1989 年以来世界各地武装冲突参与者的性暴力行为，并被编入了冲突年份发生率的顺序指标。由于这些措施的信息存在显着差异，我们目前对问题的趋势和模式缺乏科学理解。在本文中，我们使用 LVM 方法来利用跨战时性暴力的多个指标的信息来估计其真实程度，以可信区间的形式表达不确定性，并解释基础数据的时间趋势。我们认为动态 LVM 参数化构成了这种情况下的最佳拟合。它优于静态潜在变量模型，以及对观察指标的分析。根据我们的发现，我们认为 LVM 方法目前构成了这一调查线的最佳实践，并提出了对未来研究的建议。