Retrospective life history designs are among the few practical approaches for collecting longitudinal network information from large populations, particularly in the context of relationships like sexual partnerships that cannot be measured via digital traces or documentary evidence. While all such designs afford the ability to “peer into the past” vis-à-vis the point of data collection, little is known about the impact of the specific design parameters on the time horizon over which such information is useful. In this article, we investigate the effect of two different survey designs on retrospective network imputation: (1) intervalN, where subjects are asked to provide information on all partners within the past N time units; and (2) lastK, where subjects are asked to provide information about their K most recent partners. We simulate a “ground truth” sexual partnership network using a published model of Krivitsky (2012), and we then sample this data using the two retrospective designs under various choices of N and K . We examine the accumulation of missingness as a function of time prior to interview, and we investigate the impact of this missingness on model-based imputation of the state of the network at prior time points via conditional ERGM prediction. We quantitatively show that—even setting aside problems of alter identification and informant accuracy—choice of survey design and parameters used can drastically change the amount of missingness in the dataset. These differences in missingness have a large impact on the quality of retrospective parameter estimation and network imputation, including important effects on properties related to disease transmission.

中文翻译：

---

生活史数据的回顾性网络插补：设计的影响

回顾性生活史设计是为数不多的从大量人口中收集纵向网络信息的实用方法之一，特别是在无法通过数字痕迹或文件证据衡量的性伙伴关系等关系中。虽然所有这些设计都提供了针对数据收集点“回顾过去”的能力，但对于特定设计参数对此类信息有用的时间范围的影响知之甚少。在本文中，我们调查了两种不同的调查设计对追溯网络插补的影响：（1）intervalN，其中要求受试者提供过去 N 时间单位内所有合作伙伴的信息；(2) lastK，其中要求受试者提供有关他们最近的 K 个合作伙伴的信息。我们使用 Krivitsky (2012) 的已发布模型模拟“真实”性伙伴关系网络，然后我们使用 N 和 K 的各种选择下的两个回顾性设计对这些数据进行采样。我们在采访之前检查缺失的累积作为时间的函数，并且我们通过条件 ERGM 预测研究这种缺失对先前时间点网络状态的基于模型的插补的影响。我们定量地表明，即使不考虑改变身份和信息提供者准确性的问题，选择的调查设计和使用的参数也可以极大地改变数据集中的缺失量。这些缺失的差异对回顾性参数估计和网络插补的质量有很大影响，