MODE DIFFERENCES BETWEEN FACE-TO-FACE AND WEB SURVEYS: AN EXPERIMENTAL INVESTIGATION OF DATA QUALITY AND SOCIAL DESIRABILITY EFFECTS
MODE DIFFERENCES BETWEEN FACE-TO-FACE AND WEB SURVEYS: AN EXPERIMENTAL INVESTIGATION OF DATA QUALITY AND SOCIAL DESIRABILITY EFFECTS
复制标题
DOI：
10.1093/ijpor/edn054
复制
发表时间：
2009-03-01
影响因子：
1.8
通讯作者：
Heerwegh, Dirk
中科院分区：
文献类型：
--
作者：
Heerwegh, Dirk
混合模式调查容易出现混杂模式效应和无应答误差。序贯混合模式设计，例如，跟踪一个模式与后续模式的非应答者（Dillman，2000）。因此，不同调查模式的样本构成可能有很大差异，妨碍就模式效应得出直接的结论。但是，即使在实验混合模式调查设计，这样的混淆可能会发生，如果无应答不受控制。只要不同的调查模式吸引不同类型的受访者，样本的可比性就会受到损害，而且很难过滤掉模式效应（Voogt & Saris，2005）。这些考虑为目前的研究奠定了基础。本研究旨在分析面对面和网络调查之间的模式效应，而不混淆他们与无回应偏见。这两种调查模式可以代表顺序混合模式调查的起点和终点。如果人口的互联网覆盖率足够高，可以进行网络调查，收集尽可能多的数据，然后再用其他更昂贵的调查方式对其余未答复者进行后续调查。面对面调查可以是这种调查设计的最后阶段。由于这种顺序混合模式调查混淆了模式效应和无应答差异，本研究采用了同时混合模式设计。然而，一个随机选择的部分受访者被要求回答最后36个问题，通过网络调查，而其他受访者只是完成了整个调查面对面。这种设计排除了不回答错误，因为被要求在网上完成调查表最后部分的大部分答复者都遵守了这一要求。一般的研究问题是（更昂贵的）面对面调查是否比网络调查产生更高质量的数据。虽然数据质量是一个广泛的概念，但它在这里被定义为常用的指标不知道（DK）答案，项目无响应，响应无差异和默许，一方面统称为“响应质量”，另一方面是社会期望。
Mixed mode surveys are prone to confounding mode effects with nonresponse error. Sequential mixed mode designs, for instance, follow up nonrespondents to one mode with a subsequent mode (Dillman, 2000). The sample composition consequently could differ to a large extent across survey modes, hampering straightforward conclusions about mode effects. But even in experimental mixed mode survey designs, such a confounding can occur if nonresponse is not controlled. As long as different survey modes attract different types of respondents, sample comparability is compromised, and it will be difficult to filter out mode effects (Voogt & Saris, 2005). These considerations set the stage for the current study. This study aimed to analyze mode effects between face-to-face and Web surveys, without confounding them with nonresponse bias. These two survey modes could represent the starting and the end point of a sequential mixed mode survey. Given a population with a sufficiently high Internet coverage, a Web survey could be deployed to collect as much data as possible, after which other, more expensive survey modes are used to follow up remaining nonrespondents. A face-to-face survey could be the final stage of such a survey design. Because such sequential mixed mode surveys confound mode effects with nonresponse differences, the current study utilized a concurrent mixed mode design.More specifically, all respondents were interviewed face-to-face in the current study. However, one randomly selected part of the respondents was asked to answer the final 36 questions via a Web survey, while the other respondents simply completed the entire survey face-to-face. This design precludes nonresponse error to the extent that a large proportion of the respondents asked to complete the final part of the questionnaire on the Web complies with that request. The general research question is whether (more expensive) face-to-face surveys generate higher quality data than Web surveys. While data quality is a broad concept, it is defined here in terms of the commonly used indicators do not know (DK) answers, item nonresponse, response nondifferentiation, and acquiescence, referred to collectively as ‘‘response quality’’on the one hand and social desirability on the other.