Identifying Outliers in Response Quality Assessment by Using Multivariate Control Charts Based on Kernel Density Estimation

Abstract When monitoring industrial processes, a Statistical Process Control tool, such as multivariate Hotelling T 2 chart is frequently used to evaluate multiple quality characteristics. However, research into the use of charts for survey fieldwork–essentially production process in which data sets collected by means interviews are produced–has been scant date. In this study, using from eighth round European Social Survey Belgium, we present procedure simultaneously six response indicators and identifying outliers: with anomalous results. The integrates Kernel Density Estimation (KDE) chart, so that historical “in-control” or reference assumption parametric distribution not required. total, 75 outliers (4.25%) iteratively removed, resulting an in-control set containing 1,691 interviews. mainly characterized having longer sequences identical answers, greater number extreme against expectation, lower item nonresponse rate. validated ten-fold cross-validation comparison minimum covariance determinant algorithm criterion. By providing method obtaining data, findings go some way toward monitor quality, identify problems, provide rapid feedbacks during fieldwork.