Attrition Bias in the Millennium Cohort Study: The Contribution of Family Mobility
Ian Plewis, Heather Joshi and Charles Ketende
Centre for longitudinal Studies, Bedford Group, Institute of Education, University of London
Abstract
The mobility of families, one of the phenomena which longitudinal surveys may be used to study, at the same time poses a real problem for their conduct. Permanent long-distance migration, local moves of address, or temporary circulation of subjects all are challenges to the task offinding the respondents for follow-up interviews. Because we are not always entirely successful in meeting this challenge the mobility of families is a major cause ofsurvey attrition.Examining the extent to which this is a cause for concern in terms of creating bias in the sample is therefore crucial for those involved in longitudinal analysis. In order to do this, as well as its intrinsic interest, we need to know as much as possible about who moves and why and how movers differ from those who stay put.
The issue of attrition bias is a particular concern for the Millennium Cohort because of the high propensity of families with young children to move home. The focus of this paper is thereforetwofold: Firstly, we discuss the methodological issues arising from the high mobility of cohort families in the Millennium Cohort Study and secondly, we examine techniques being used to locate movers.
The Millennium Cohort conducted its first survey during 2001-2 in selected electoral wards across the whole United Kingdom when the survey subjects were aged 9 months. Calculations indicate that around 7% of the target population were missed because of internal movement (Plewis et al., 2004). The second sweep of the survey (which collected data when the cohort members were around 3 years old) providesinformation on families who were missed in the first survey because their move into the target area was not recorded in time. These families have subsequently been found in administrative records. In addition we have information on the families who moved between the first and second sweeps (around one in 5).Using these data we are able to compare the characteristics of the movers to non-moversin order to gauge the effect this movement has on attrition bias.We can retrospectively describe some of those who were missing at Sweep 1 by using the ‘repair’ dataset. We can prospectively record the characteristics of those who are lost to follow up between Sweeps 1 and 2 and in both surveys we can compare the characteristics of included families who moved with those who are included and have not changed address over a given period. Moreover, we can compare refusals at Sweep 2 with those remaining in the cohort, in terms of their Sweep 1 characteristics and we also know something about refusals at Sweep 1. Consequently, we can estimate the contribution of mobility to overall attrition bias.
Maintaining the Cohort, despite the mobility of families is of major importance given the concerns of attrition bias. Therefore, the second part of this paper discusses a variety of methods employed to keep track of all families in the Millennium Cohort Study, even those who move. These include using linked administrative data from the Department of Work and Pensions and Health Service Registers to try to locate the mobile families. The paper will consider how successful these types of methods have been in locating the mobile families and the impact that this has had on reducing attrition bias.