It is common practice to adapt the format of a question to the mode of data collection. Multi-coded questions in self-completion and face-to-face modes tend to be transformed for telephone into a series of ’yes/no’ questions. Questions with response scales are often branched in telephone interviews, that is, converted into two or more questions, each with shorter response lists. There has been limited research into the impact of these format differences on measurement, particularly across modes. We analyse data from an experiment that contrasted these question formats in face-to-face, telephone and web surveys. The study also included a cognitive interviewing follow-up to further explore the quantitative findings.