Development of a method for ex-post identification of falsifications in survey data

Results of a research project dealing with ex-post detection of falsified data in surveys are reported. Based on an analysis of the motivation for falsifications we develop, test and apply multivariate statistical methods, which can be used to identify falsifications in survey data. The methods build on specific statistical properties of falsified interviews and their interdependence. The classification of interviewers is based on these indicators calculated for the data collected by each interviewer. In a first explorative phase we identify further attributes of questionnaires which are useful to detect interviewers producing falsified data. Among those attributes are both specific type of content, e.g. knowledge questions, and behaviour across several questions, e.g. reliability in multi item scales. It is explored to what extent these additional indicators improve the capability to distinguish between potential falsifiers and regular interviewers. The sensitivity of the results with regard to the number of interviews available is analyzed by means of bootstrap analysis. The results are discussed regarding methodological issues in development of our data driven approach for identification of falsified interview data as well as its potential application already during the field phase in real surveys.