Asking sensitive questions in surveys is prone to mode effects. To analyze the consequences of switching from single to mixed mode designs, we randomly assign survey participants to a single mode (CATI) and a mixed mode (CATI and Web-based) survey. Both surveys contain the same set of items based on already established instruments from German health surveys conducted by the Robert Koch Institute. The instruments cover sensitive items such as subjective well-being, health behavior and self-reported illness, augmented with questions about confidentiality concerns. Because of randomization differences in means are informative on the net impact of switching to mixed mode. Applying a Blinder/Oaxaca decomposition method from econometrics to the field of survey methodology, we are able to separate (1) selection effects due to differences in non response rates from (2) mode effects due to differences in item non response and measurement error. Our study design allows us to draw conclusions about the relative strength of selection and mode effects without the use of external validation data. As the data collection will start in April 2011 we will be able to provide preliminary results by July. The project is funded by the German Robert Koch Institute (FKZ 1362/1-978).