Bias in Telephone Surveys That do not Sample Cell Phones Uses and Limits of Poststratification Adjustments
Bias in Telephone Surveys That do not Sample Cell Phones Uses and Limits of Poststratification Adjustments
复制标题
DOI：
10.1097/mlr.0b013e3182028ac7
复制
发表时间：
2011-04-01
期刊：
影响因子：
3
通讯作者：
Nelson, Justine
中科院分区：
文献类型：
--
作者：
Call, Kathleen Thiede;Davern, Michael;Nelson, Justine
目的：检验健康调查在忽略仅使用手机的家庭 (CPOH) 时的偏差程度，并探讨事后分层是否可以减少这种偏差。方法：我们使用 2008 年全国健康访谈调查 (NHIS) 的数据，该调查采用区域概率抽样和面对面访谈；因此，所有电话状态的人都包括在内。首先，我们检查了居住在 CPOH 的人们与非居住在 CPOH 的人们在几个重要的健康监测领域是否有所不同。我们将标准 NHIS 估计值与一组不包括居住在 CPHO 的人的“重新加权”估计值进行了比较。重新加权的 NHIS 病例是通过对 NHIS 控制总数进行一系列分层后调整来拟合的。除了对地区、种族或族裔以及年龄进行分层后调整之外，我们还研究了对住房所有权、受教育年龄和家庭结构的调整。结果：分层后减少了非老年人口所有健康相关估计值的偏差。然而，这些调整对西班牙裔和黑人的效果较差，对年轻人（18 至 30 岁）效果更差。对于无保险和没有常规护理来源的估计，偏差的减少幅度最大，而对于饮酒、吸烟以及由于成本而放弃或延迟护理的估计，偏差的减少效果更差。结论：对排除 CPOH 的数据进行分层后调整，对于健康保险等估计在总人口水平上效果很好，但对于获取和健康行为的估计则效果较差。然而，分层后调整不足以减少亚人群水平上与健康相关的估计的偏差，特别是对于那些有兴趣测量和监测种族、民族和年龄差异的人。
Objective: To examine how biased health surveys are when they omit cell phone-only households (CPOH) and to explore whether poststratification can reduce this bias.Methods: We used data from the 2008 National Health Interview Survey (NHIS), which uses area probability sampling and in-person interviews; as a result people of all phone statuses are included. First, we examined whether people living in CPOH are different from those not living in CPOH with respect to several important health surveillance domains. We compared standard NHIS estimates to a set of "reweighted" estimates that exclude people living in CPHO. The reweighted NHIS cases were fitted through a series of poststratification adjustments to NHIS control totals. In addition to poststratification adjustments for region, race or ethnicity, and age, we examined adjustments for home ownership, age by education, and household structure.Results: Poststratification reduces bias in all health-related estimates for the nonelderly population. However, these adjustments work less well for Hispanics and blacks and even worse for young adults (18 to 30y). Reduction in bias is greatest for estimates of uninsurance and having no usual source of care, and worse for estimates of drinking, smoking, and forgone or delayed care because of costs.Conclusions: Applying poststratification adjustments to data that exclude CPOH works well at the total population level for estimates such as health insurance, and less well for access and health behaviors. However, poststratification adjustments do not do enough to reduce bias in health-related estimates at the subpopulation level, particularly for those interested in measuring and monitoring racial, ethnic, and age disparities.