Abstract

Asking sensitive questions in surveys is prone to mode effects. In a randomized experiment we compare a single mode CATI to a mixed mode survey where respondents can chose between a CATI and Web-based mode. Our main focus lies on the consequences of the additional Web-survey option on data quality. Indicators of data quality are the respondent`s tendency towards social desirable answers, satisficing behavior and item non-response. In both modes identical questionnaires are used covering sensitive items such as health behavior and self-reported illness and including auxiliary variables e.g. about item sensitivity and affinity to the internet. However differences in means of sensitive items between single and mixed mode surveys can also be due to differences in unit non-response induced by the Web-option. Therefore we propose a decomposition method. It utilizes statistical matching techniques to estimate the relative strength of selection and mode effects even without external validation data .