ABSTRACT
This chapter looks at the evidence on the quality of the data and discusses non-response bias. The combination of eliminating the smallest cells and incomplete information in other cells has meant that the numbers are so small, in absolute terms, as to call into question the reliability of the answers. Whenever variables are restricted there is the possibility of truncation bias. The problem arises because of the random error in the model. The actual model tested is unlikely ever to explain all of the variation in the dependent variable, and typically in cross-section work explains considerably under half. A related difficulty arises even when people who have worked zero hours are included in the sample. The hypothesis was that people in areas of high unemployment would be less willing to be interviewed about work than people in areas where unemployment was low.
