Jäckle et al (2011, this session) argue that differences in the characteristics of the question administration method are a cause of mode effects on measurement. One such characteristic is the communication channels by which questions and answers are transmitted between researcher/interviewer and respondent. For any particular survey question, a combination of channels may be used. There are three pieces of information to be communicated – the question, the response options, and the answer – and there are multiple possible channels for each. A respondent may perceive a survey question either visually or aurally, but the response options need not be perceived through the same channel (e.g. if an interviewer reads the question but the respondent has to read the response options on a card). And he or she may be requested to supply the answer either orally or by self-completion, the latter of which could involve either pen-and-paper or computer keyboard/mouse.

Using data from a large-scale controlled experiment involving CAPI, CATI and web treatments, we attempt to identify effects of communication channels on responses and the ways in which these effects may differ between types of question. Our study includes behavioural, satisfaction and attitude items. We also present comparative results from a

cognitive interviewing follow-up that provided additional understanding of the quantitative results. This study is funded by the UK Economic and Social Research Council.