Identification of Falsifications in Survey Data

调查数据可能会被面试者篡改。在分析造假动机的基础上，我们开发、检验和应用了可用于识别调查数据中造假的多元统计方法。这些方法建立在伪造访谈的特定性质上，例如关于未回答问题的数量或数字的分布及其相互依赖。在应用分类方法时，使用这些标准对实际访谈数据进行分组。我们还考虑了启发式优化算法，以获得针对不同标准的最佳可能的聚类。此外，在事先掌握造假信息的情况下，使用判别分析来识别虚假访谈的典型特征。在分析统计方法的同时，我们开发了问卷的设计特征，从而增加了事后发现造假的机会。因此，调查数据的质量预计会有所改善，这既是因为事后查明了虚假面谈，也是因为加强了威慑作用。问卷设计和统计方法在实验环境中进行了测试。

Survey data might be subject to falsifications by interviewers. Based on an analysis of the motivation for such falsifications we develop, test and apply multivariate statistical methods, which can be used to identify falsifications in survey data. The methods build on specific properties of falsified interviews, e.g., with regard to the number of unanswered questions or the distributions of digits, and their interdependence. The grouping of actual interview data is performed using these criteria when applying clustering methods. We also consider heuristic optimisation algorithms for obtaining the best possible clusters for different criteria. Furthermore, if ex ante information on falsifications is available, discriminant analysis is used to identify typical properties of false interviews.In parallel to the analysis of the statistical methods, we develop design features of questionnaires, which increase the chance of ex post detection of falsifications. Thereby, the quality of survey data is expected to improve both due to the ex post identification of faked interviews and an increased deterrence effect. Questionnaire designs and statistical methodology are tested in an experimental setting.