Modelling non-response in the National Child Development Study

(2006) - Denise Hawkes, Ian Plewis

Journal: Journal of the Royal Statistical Society
Link::
DOI:: 10.1111/j.1467-985X.2006.00401.x
Links::
Tags:: #paper #NCDS #MissingData
Cite Key:: [@Hawkes2006]

Abstract

Summary. There is widespread concern that the cumulative effects of the non-response that is bound to affect any long-running longitudinal study will lead to mistaken inferences about change. We focus on the National Child Development Study and show how non-response has accumulated over time. We distinguish between attrition and wave non-response and show how these two kinds of non-response can be related to a set of explanatory variables. We model the discrete time hazard of non-response and also fit a set of multinomial logistic regressions to the probabilities of different kinds of non-response at a particular sweep. We find that the best predictors of non-response at any sweep are generally variables that are measured at the previous sweep but, although non-response is systematic, much of the variation in it remains unexplained by our models. We consider the implications of our results for both design and analysis.

Notes

“There is widespread concern that the cumulative effects of the non-response that is bound to affect any long-running longitudinal study will lead to mistaken inferences about change.” (Hawkes and Plewis, 2006, p. 479)

“We find that the best predictors of non-response at any sweep are generally variables that are measured at the previous sweep but, although non-response is systematic, much of the variation in it remains unexplained by our models.” (Hawkes and Plewis, 2006, p. 479)

“Table 1. NCDS longitudinal target and observed samples, sweeps 0–6” (Hawkes and Plewis, 2006, p. 480)

“Most of this fall is accounted for by ‘non-response: other’. These are cases for which there are (a) no data (either a response or a refusal) for the current sweep but some data (either a response or a refusal) at a later sweep and (b) ‘temporary emigrants’ who were known to be abroad at the current sweep but who returned to Great Britain later.” (Hawkes and Plewis, 2006, p. 481)

“The distinction between wave non-response and attrition is important up to sweep 3 as the correlates of the two kinds of non-response are different. The attrition cases (2% of the cases) after sweep 2 are missing on a more systematic basis than the wave non-respondents (7% of the cases) at sweep 3. We also find that cases lacking all education data (15% of the total) at sweep 3 differ from attrition cases and wave non-respondents.” (Hawkes and Plewis, 2006, p. 485)

“It is clear that there are systematic differences between respondents and non-respondents at every sweep with a tendency for male cohort members and cases with lower educational attainments, less stable employment patterns and living in more disadvantaged circumstances to be more likely to be lost from the study. There is, therefore, no support for the position that NCDS data are, to use the terminology of Little and Rubin (1987), missing completely at random.” (Hawkes and Plewis, 2006, p. 489) Exceptionally important for my PhD, work on missing data.

“Although there is some evidence, most notably from the US Panel Study of Income Dynamics (Lillard and Panis, 1998), that the effects of attrition in longitudinal studies can be relatively benign, it cannot be assumed that this will always be so. Watson (2003) and Behr et al. (2003), both based on analyses of the European Community Household Panel, showed that the correlates of attrition vary from country to country and therefore it would be rash to assume that findings that hold for a panel study in the USA will necessarily apply to a UK birth cohort study.” (Hawkes and Plewis, 2006, p. 490)