WAVE-SEAM EFFECTS IN THE SIPP
WAVE-SEAM EFFECTS IN THE SIPP
复制标题
SIPP 中的波缝效应
DOI：
--
复制
发表时间：
2002
期刊：
影响因子：
--
通讯作者：
N. Young
中科院分区：
文献类型：
--
作者：
N. Young
接缝是两块独立材料邻接的点。在封接工艺中，产品的质量通常由接缝的质量决定。纵向调查也有接缝，与制成品一样，这些接缝是决定调查整体质量的关键因素。纵向调查中的接缝出现在将覆盖两个不同访谈期（或波）的数据相结合的地方。由于SIPP每次收集四个月的信息，每隔四个月收集一次，因此它有许多"波纹"。"在SIPP中，不同波中连续两个月的反应变化要比同一波采访中连续两个月的反应变化大得多。响应对波浪变化的依赖性称为波缝效应。下面是1985年6月和1985年6月在F e B r u a r y访问的家庭的具体例子，这使波缝效应更加清楚。在法国，调查人员将收集10月、11月、12月和1月的个人收入信息。6月份，面试官将收集2月、3月、4月和5月的个人收入信息。报告的个人收入在1月和2月之间变化的可能性是其他任何连续月份之间变化的两倍多。这是因为1月份的数据是在一次访谈中收集的，2月份的数据是在下一次访谈中收集的。当波缝效应较大时，基于SIPP的统计推断是否具有误导性会引起质疑。这些问题并不是SIPP独有的，因为调查中的重复测量往往会产生不同的答案。SIPP的问题不一定是存在更多的固有响应误差。相反，S I P P的问题是具有所需的保留测量。本研究从一个新的角度探讨了纵向调查中的数据质量问题。在SIPP中横截面数据质量的研究中（King et. al，1987：达林普尔and Carlson，1986），SIPP已被证明与其他横截面调查一致。在SIPP变化率的研究中（King et. Al，1987）已经表明平均变化率表现出大的波缝效应。在这项研究的SIPP的另一个问题是解决。考虑到SIPP中的反应行为受到访谈时间的影响，重要的是要问：在SIPP中，基于跨期关联的推断有多有效？在回答这个问题时，我们指出，在SIPP的动力分析，而不是免疫波缝效应，是不是它的削弱。在第一部分中，我们讨论了我们的直接调查结果从描述性研究的波缝效应。第一部分直接表明，虽然绝对变化率受到波缝效应的强烈影响，但变化率的相关性并不受影响。在第二部分中，我们讨论了SIPP中波缝效应的一些最合理的来源及其对分析的影响。特别是，我们讨论了恒定波响应，这是一种可能存在于许多回顾性调查中的响应偏差（测量误差），但只有在SIPP这样的回顾性和纵向调查中才会发现。
A seam is the point at which two separate pieces of material abut. In manufactur ing processes, the quality of the product is often dictated by the quality of seams. Longitudinal surveys also have seams, and as with manufactured products, these seams are critical elements in determining the overall quality of the survey. The seams in a longitudinal survey occur in p laces where data that cover two d i f fe ren t interview periods (or waves) are combined. The SIPP because it gathers information for four months at a time, at four-month intervals, has many "waveseams." In the SIPP, responses change much more between two consecutive months in different waves than between two consecutive months in the same interview wave. The dependency of responses on the change in waves is called the wave-seam effect. The wave-seam effect is made clearer by the following specific example of households interviewed in F e b r u a r y and June of 1985. In F e b r u a r y , in t e rv iewers would col lec t persona l earn ings information for the months of October, November, December, and January. In June, interviewers would collect personal earnings information for the months February, March, April, and May. Reported personal earnings would be more than twice as likely to change between January and February than they would be between any other pair of consecutive months. This is because data for January is collected during one interview, and data for February is collected during the next interview. Where wave-seam effects are large, questions are raised about whether statistical inferences based on the SIPP are misleading. These questions are not unique to the SIPP, as repeated measurements in surveys often generate different responses. The problem of the SIPP is not necessarily that there is more intrinsic response error. Rather the problem of the S I P P is t h a t t h e r e a re m o r e r e p e a t e d measurements. This study addresses data-quality in longitudinal surveys from a new perspective. In studies of crosssectional data quality in the SIPP (King et. al, 1987: Dalrymple and Carlson, 1986) the SIPP has been shown to concur with other cross-sectional surveys. In studies of rates of change in the SIPP (King et. al, 1987) mean rates of change have been shown to exhibit large wave-seam effects. In this study of the SIPP another question is addressed. Given that response behavior in the SIPP is influenced by timing of the interviews, it is important to ask: How valid are inferences based on intertemporal associations in the SIPP? In answering this question we show that dynamic analysis in the SIPP, while not immune to the wave-seam effect, is not crippled by it. This paper is divided into two parts. In the first part we discuss our direct findings from a descriptive study of the wave-seam effect. The first section demonstrates directly that while absolute rates of change are strongly influenced by the wave-seam effect, that the correlations of the rates of change are not. In the second part of we discuss some of the most plausible sources of the wave-seam effect in the SIPP and their imp l i ca t ions for ana lys is . In particular, we discuss the constant wave response, a form of response bias (measurement error) that may be present in many retrospective surveys, but which will be revealed only in surveys like the SIPP that are both retrospective and longitudinal.