What is it about?
In studies that collect data at many points in time, some groups of participants have consistently lower levels of responsiveness/higher levels of missing data—particularly participants from disadvantaged and socially marginalised populations. With respect to these people, data are not missing at random. This paper presents a simulation study showing that in these studies, measures of responsiveness over time can be used as additional variables in imputation or missingness models to greatly reduce the associated bias. The paper also includes some introductory sections that describe approaches to inverse probability-weighting for missing data that are compatible with non-monotone missing data.
Featured Image
Why is it important?
People from disadvantaged backgrounds or who experience adverse childhood experiences are much more likely to have missing data or to be excluded from population-based surveys. The methods described in this paper were developed to improve the reliability of statistics and research about social disadvantage and childhood adversity.
Perspectives
The introductory sections on approaches to inverse probability-weighting for non-monotone missing data are somewhat academic, given that the subsequent simulation showed them to be much more sensitive than multiple imputation to violations of the missing at random assumption (i.e. when data are not missing at random; the context for which they had been developed). The real take-home message is that in studies with many points of data collection, there is a valuable opportunity to measure participants responsiveness over time, and these measures of responsiveness may be used to reduce bias/increase the plausibility of the missing at random assumption.
Mr James C Doidge
University of South Australia
Read the Original
This page is a summary of: Responsiveness-informed multiple imputation and inverse probability-weighting in cohort studies with missing data that are non-monotone or not missing at random, Statistical Methods in Medical Research, March 2016, SAGE Publications,
DOI: 10.1177/0962280216628902.
You can read the full text:
Contributors
The following have contributed to this page
