Social researchers tend to assume high levels of opinionatedness on the part of respondents, so that little attention is paid to establishing whether or not respondents actually have had the necessary experience upon which an assumed opinion or belief could be based. This is a problem because it has been found that one cannot rely on respondents to indicate when questions are of marginal relevance to them. Respondents typically do their best to answer every question put to them — even questions that they have difficulty understanding or relating to — either by cognitively adjusting the question so they can answer it or by falling back on contextual clues and general attitudes to formulate an appropriate answer (see, e.g., Schuman and Presser, 1980; Bishop et al, 1980; Belson, 1981; Strack and Martin, 1987).

A question or question component that is explicitly offered to a respondent, either to establish the relevance of the question to the respondent, or to emphasise that it is acceptable for the respondent to not answer the question, is generally referred to as a ‘filter’. A variety of filters have been invented ranging from the simple ‘Don't know’, ‘Can't recall’ and ‘No opinion’ response options, through the ‘Undecided’, ‘Not sure’, ‘It depends’ and ‘Neutral’ alternatives, to sets of questions that attempt to establish the extent to which an opinion or belief is based on experience of or knowledge about the topic.