14.6: The interview

As much as possible, a face-to-face interview should approximate to a conversation between the interviewer and respondent and must not be an interrogation. Good rapport between the two is vital, and the onus is entirely on the project team to ensure this.

6.1 Who, where, and when

In studies of children, the best informant regarding their health or behaviour is likely to be their mother or guardian. Only as a last resort should someone else be interviewed for this purpose. This may necessitate repeated visits to a household, until the mother or guardian is at home. Other than for children, proxy informants should be avoided, if possible.

The choice of the place of interview will be influenced by logistic considerations and the nature of the information to be collected. Usually, the place will have to be chosen for the convenience of the respondent, rather than for that of the interviewer. Privacy will be easier to ensure in a hospital or a clinic than in a village setting, but special arrangements may be made to ensure greater privacy in a village. For example, an interview might be conducted slightly away from the house under a shady tree. If interviews are to be conducted in homes, as far as possible, the time should be chosen to fit the convenience of the residents. If possible, they should be consulted, or at least informed, in advance regarding when an interview will be scheduled. Preliminary investigations, before the main survey, should be made to ascertain when the most convenient time will be for most participants. In rural communities, during planting or harvesting seasons, evening interviews may be preferred. But, if interviews take place after dark, poor lighting may be a problem, and attention to clear printing and a well-spaced layout for the questionnaire becomes even more important (as well as the provision of torches and batteries to interviewers). In some areas, security after dark may also be a significant problem, and interviewing at that time may be inadvisable.

6.2 Non-response

Steps that can be taken to ensure data completeness within a single questionnaire are discussed in Chapters 16 and 20. Here, we discuss the problem of non-response where a trial participant is either not seen or refuses to take part in the trial or in a particular data collection ‘round’ or survey.

Non-respondents in a study are rarely representative of the rest of the study population. They are a self-selected group, and thus their exclusion will usually introduce bias into the results of a survey, but the degree to which that has occurred is not usually directly measurable. Thus, if a high proportion of the target population for interviews are not interviewed, the valid interpretation of the results from those who are interviewed, and in particular the generalization of these results to the whole community, may be open to serious question. Therefore, great care must be taken to ensure that the response rate is high. This may be achieved in several ways. First, the questions included in the questionnaire should be thoroughly tested in a pilot study, so that any that a significant proportion of respondents cannot, or will not, answer adequately are eliminated. Second, an appropriate explanation of the survey should be given to study participants in advance, and any false suspicions they have about the motives or intentions of the investigators must be dispelled. Third, interviewers must be selected who are persistent, yet polite, and who will probe for a correct response to a question and not accept a ‘don’t know’ response too readily. Fourth, interviewers must be instructed to call back repeatedly if a house is empty or a respondent is away, before abandoning an interview. Their work schedule should take into account the need for such return visits.

Systems should be put in place to monitor the non-response rate within a trial on an ongoing basis, so that steps can be taken to attempt to decrease this, before it is too late. The non-response data should be disaggregated by the interviewer, the trial team, and other important groupings, where appropriate, such as language, location, etc., and all outliers investigated carefully. For example, in a trial of human papillomavirus vaccination within schools, it was discovered that the non-response rate was substantially higher for one field team than for the other. This turned out to be due to the way the team members were introducing themselves and the trial within the schools—something that could be changed, and the problem was quickly solved.

Even in the most well-conducted surveys, a 100% response rate is rare. Indeed a 100% response rate should be viewed with some suspicion! As much information as possible should be obtained about non-respondents, where necessary from proxy informants, so that the characteristics of non-responders for which information is available (for example, age and sex) may be compared with that available on responders. This may give clues to the extent of possible biases resulting from their exclusion.