1. Introduction
The Millennium Cohort Study (MCS) is the fourth in a series of internationally
renowned cohort studies in the UK. It includes 18818 babies in 18552 families born
over a 12 month period and living in selected UK wards at age 9 months. Areas with
high proportions of Black and Asian families, disadvantaged areas and the three
smaller UK countries are all over-represented in the sample which is
disproportionately stratified and clustered. The first and second sweeps took place
when the cohort members were 9 months and 3 years old. In addition, at sweep two,
families who were living in sampled areas with a child of the appropriate age but who
were not located at the first sweep were introduced. These “new families” tend to be
more mobile than those already part of the MCS. Partners were interviewed
whenever possible and detailed questions about individual and family income were
included in both sweeps.
There are four ways in which income data can be missing. There was unit non-
response at sweep one such that the response rate then was 72%. There was
further partner non-response; the partner response rate at sweep one, among
respondent families with partners, was 88%. In addition there was item non-
response for income: about 6% of main respondents and 6% of partners did not
provide income data at sweep one. Moreover, there was attrition between sweeps
one and two: 79% of eligible cases responded at sweep two. The correlates of unit
and partner non-response at sweep one are set out in Plewis (2004); the evolution of
the sample from sweep one to sweep two is described in Plewis and Ketende (2006).
The paper will address the following questions:
(i) Are there (a) within household and (b) within individual correlations for missing
income data?
(ii) Is a female interviewer more successful than a male interviewer in getting
responses to income questions from main respondents and their partners?
(iii) Is there a systematic tendency for income data to be missing at sweeps one
and two over and above what we know about unit and partner non-response?
(iv) Is attrition at sweep two related to (a) family income at sweep one; (b) the
failure to provide income data at sweep one?
The paper will conclude by considering the implications for statistical modelling and
future data collection of our findings on the patterns and correlates of income non-
response at both sweeps.
1.1 Background Literature
Unlike the previous UK cohort studies, MCS has been designed with a focus on
social and economic data rather than health data. As a consequence the quality of
the MCS dataset could be reduced by the failure of some participants to report their
income. If income non-response were truly random then it would merely result in a
loss of precision in any statistical analysis based on complete cases. However,