Errors in autobiographical memory and their effects in time-to-event analysis

用于分析事件时间数据的最常见的数据分析过程之一是风险模型。在社会科学和流行病学中的大多数应用使用来自横断面调查的数据。这类调查通常依赖于受访者基于自传式记忆的报告。因此，这些数据可能会由于内存失效而产生测量误差。在实际研究中经常发生的忽略这种误差是一种危险的做法，因为参数估计和模型拟合受到影响。虽然有大量的文献研究测量误差对线性和非线性回归模型的影响，但记忆误差对危害模型的影响通常并不广为人知。本项目将在流行病学和社会学背景下，利用访谈和行政记录调查自传式回忆日期的测量误差的类型和数量。将开发用于解释记忆失败的模型，使用以前未应用于回顾性数据的技术。将研究记忆失效对风险模型估计和拟合的影响。此外，利用基于经验的记忆故障统计模型，将研究不同纠错方法的性能。

One of the most common data analysis procedures for analyzing time-to-event data are hazard models. Most applications in the social sciences and in epidemiology use data from crosssectional surveys. Such surveys usually rely on respondents' reports based on their autobiographical memory. Therefore, these data may suffer from measurement errors due to memory failure. Ignoring such error, as often happens in practical research, is a dangerous practice, since parameter estimates and model fit are affected. While there is an extensive literature on the effects of measurement error in linear and nonlinear regression models, the impact of memory errors on hazard models in general is not widely known.This project will investigate type and amount of measurement error in autobiographical recall of dates in an epidemiological and sociological context using interview and administrative records. Models for the explanation of memory failures will be developed, using a technique which was not applied before on retrospective data. The effect of memory failures on estimates and fit of hazard models will be investigated. Further, using the empirically based, statistical model for memory failures, the performance of different error correction methods will be studied.