Web surveys, being self-administered, ensure more privacy for the respondent and there is considerable evidence that web surveys result in less social desirable answers. However, the absence of an interviewer may also be seen as a disadvantage, as there are no interviewers to motivate respondents and probe them.

In this paper we report on an experiment in which a particular type of question format, i.e. how the “do not know” option is presented and whether or not a probe is used, affects answers to web and telephone surveys.

In interview surveys, usually a ‘do-not-know’-option is not explicitly offered to a respondent, but interviewers can accept it. It is considered good practice to train interviewers in using a probe after an initial ‘do-not-know’ to reduce item-nonresponse.

In web surveys designers are hesitant to offer an explicit do-not-know option for fear of encouraging respondents to choose this option as a quick answer. One the other hand, not accepting do-not-know and issuing an error message insisting on an answer, may lead to either irritation and more break-offs or to guessing and less valid answers.

Using the interactivity of the web, we emulated friendly interviewer probing behaviour in a probability based Internet panel. The questionnaire contained a series of questions, which in previous selfadministered (mail and web) surveys showed a high percentage of item-nonresponse. A two by two experimental design was used: (1) explicit offering of do-not-know vs. no do-not-know option, and (2) directly accepting a do-not-know vs. only accepting it after a friendly probe. As baseline for comparison a fifth condition was added with the ‘standard’ web option: an error message with no acceptance of continuation without an answer. The number of resulting do-not-know answers in each condition are

evaluated and compared with the results of a telephone survey on the same topic.