a Department of Public Health, School of Nursing and Midwifery , Maragheh University of Medical Sciences , Maragheh , Iran.
b Department of Epidemiology and Biostatistics, School of Public Health , Tehran University of Medical Sciences , Tehran , Iran.
Subst Use Misuse. 2019;54(4):601-611. doi: 10.1080/10826084.2018.1528462. Epub 2018 Dec 30.
Nonrandomized response (NRR) models are a new generation of surveys for sensitive issues. This study aims to evaluate the validity of estimates from the crosswise model (as one of the efficient models) through employing different response probabilities of nonsensitive questions.
This cross-sectional study was conducted during October and November 2015 among 1777 students of Tabriz University of Medical Sciences. Estimates of monthly alcohol consumption, and at least one instance of illicit drug use and extramarital sex over the last year were determined using direct questioning (DQ) and the Crosswise model (CM). In the last model, the probability of positive response to the nonsensitive questions was determined by using five different methods: uniform distribution (I), Benford's law (II), and estimations based on data from three other studies (III, IV, V).
Crosswise estimates of sensitive behaviors with different probabilities of a positive response to nonsensitive questions differed significantly. For example, estimates of history of using illegal opioids at least once in the last year among men varied significantly from 5.0% to 16.1% with different crosswise models based on the probability of being born in Spring using method I (0.250), III (0.287), IV (0.248), and V (0.310). The model based on Benford's law (II) was applied to estimate alcohol and cannabis consumption, and its estimates showed significant discrepancy with results of crosswise models I and V.
Estimates from crosswise model is highly sensitive to the response probability of nonsensitive questions. It seems that if this question is not selected carefully, the mentioned models will provide overestimates or underestimates, and the more-is-better hypothesis is not always valid. To achieve valid estimates, the exact probability of a positive response to the nonsensitive question must be known for the studied population.
非随机应答 (NRR) 模型是一种用于敏感问题的新一代调查方法。本研究旨在通过使用不同的非敏感问题的应答概率来评估横交模型（作为有效模型之一）的估计值的有效性。
这是一项在 2015 年 10 月至 11 月间进行的横断面研究，对象为大不里士医科大学的 1777 名学生。使用直接询问（DQ）和横交模型（CM）来确定过去一年中每月的酒精消费、至少一次非法药物使用和婚外性行为的估计值。在最后一个模型中，通过使用五种不同的方法来确定非敏感问题阳性应答的概率：均匀分布（I）、本福德定律（II）以及基于来自其他三项研究的数据的估计（III、IV、V）。
对于不同的非敏感问题阳性应答概率，横交模型对敏感行为的估计值有显著差异。例如，在男性中，过去一年至少使用过一次非法阿片类药物的历史，根据横交模型 I（概率为 0.250）、III（概率为 0.287）、IV（概率为 0.248）和 V（概率为 0.310），出生在春季的概率为 0.250），其估计值在 5.0%至 16.1%之间有显著差异。基于本福德定律（II）的模型用于估计酒精和大麻的消费，其估计值与横交模型 I 和 V 的结果有显著差异。
横交模型的估计值对非敏感问题的应答概率非常敏感。如果这个问题选择不当，那么上述模型将会提供过高或过低的估计值，而且“越多越好”的假设并不总是成立的。为了获得有效的估计值，必须了解所研究人群对非敏感问题的阳性应答的确切概率。
