Davern Michael, Jones Arthur, Lepkowski James, Davidson Gestur, Blewett Lynn A
School of Public Health, University of Minnesota, Minneapolis, MN 55414, USA.
Inquiry. 2007 Summer;44(2):211-24. doi: 10.5034/inquiryjrnl_44.2.211.
This study examines whether reasonable standard errors for multivariate models can be calculated using the public use file of the Current Population Survey's Annual Social and Economic Supplement (CPS ASEC). We restrict our analysis to the 2003 CPS ASEC and model three dependent variables at the individual level. income, poverty, and health insurance coverage. We compare standard error estimates performed on the CPS ASEC public use file with those obtained from the Census Bureau's restricted internal data that include all the relevant sampling information needed to compute standard errors adjusted for the complex survey sample design. Our analysis shows that the multivariate standard error estimates derived from the public use CPS ASEC following our specification perform relatively well compared to the estimates derived from the internal Census Bureau file. However, it is essential that users of CPS ASEC data do not simply choose any available method since three of the methods commonly used for adjusting for the complex sample design produce substantially different estimates.
本研究考察了是否可以使用当前人口调查年度社会经济补充调查（CPS ASEC）的公共使用文件来计算多元模型的合理标准误差。我们将分析限制在2003年的CPS ASEC，并在个体层面上对三个因变量进行建模，即收入、贫困和医疗保险覆盖情况。我们将对CPS ASEC公共使用文件进行的标准误差估计与从人口普查局的受限内部数据中获得的估计进行比较，后者包含计算针对复杂调查样本设计调整后的标准误差所需的所有相关抽样信息。我们的分析表明，按照我们的规范从CPS ASEC公共使用文件得出的多元标准误差估计与从人口普查局内部文件得出的估计相比表现相对较好。然而，CPS ASEC数据的使用者必须注意不能简单地选择任何可用方法，因为常用于调整复杂样本设计的三种方法会产生截然不同的估计结果。
