Nonresponse in sample surveys
'The only time that suits me is 7 o'clock on Sunday morning,' said the
'OK, I'll be there,' answered the interviewer.
'You really want this interview, don't you? Well, you'd better come in and let's get it over with.'
'This man was not interested in being interviewed at all,' said one interviewer to the other. 'And then he finally gave in, and then there was no stopping him once he started talking about politics and the European Union'.
Survey researchers in the Netherlands seem to have resigned themselves to
low response rates in random sample surveys, despite the fact that the response
rate is generally seen as the main indicator of survey quality. This study
demonstrates that high response rates are feasible in the Netherlands and
describes how these have been attained. Rather than focusing on the rate of
response, the aim of this study is to show when nonresponse can cause bias, and
to investigate the causes of and the reasons behind nonresponse. It presents
the dangers of focusing on socio-demographic and socio-economic background
characteristics and ignoring the reasons why these may be correlated with
response behaviour, and also discusses the importance of distinguishing between
noncontact and noncooperation.
Even when response rates have been enhanced substantially, the question remains as to whether higher response rates actually decrease nonresponse bias. In order to investigate this assumption, data have to be available on the response process and on final nonrespondents. This study presents instruments for collecting these data and shows not only that higher response rates do not necessarily result in better data quality, but also that respondents who require the greatest fieldwork efforts are not necessarily similar to final nonrespondents.
Based on an extensive overview of the literature and detailed empirical analyses, it is recommended that data collection be approached with the same scientific rigour as sampling and statistical analysis, and that the aim should be to minimize nonresponse bias rather than to maximize the response rate. Hunting for additional respondents may simply result in more of the same; whereas setting aside part of the funding to obtain information about nonrespondents and reluctant respondents provides the means to really open up the black box.