The fifth seminar in the CLOSER Longitudinal Methodology Series features talks from Dick Wiggins, Professor in Quantitative Social Science at the UCL Institute of Education, and Lisa Calderwood, Principal Investigator of Next Steps (previously known as the Longitudinal Study of Young People in England) and the Senior Survey Manager at the Centre for Longitudinal Studies.
About the CLOSER Longitudinal Methodology Series
The aim of the series is to highlight methodological innovations and expertise and in turn facilitate and encourage future collaborations and new research.
Dr Lisa Calderwood
The prevention of non-response in longitudinal surveys: an overview of the literature and some ideas for future research
This presentation is about the prevention of unit non-response in longitudinal surveys through improving survey practice. I will give a brief overview of the relevant theoretical and empirical literature on non-response in longitudinal surveys covering all of the main sources of non-response i.e. location, contact and co-operation and focusing primarily on evidence from major face-to-face longitudinal surveys in the UK. Recent developments such as the use of para-data, adaptive or response designs and targeted response inducement strategies will be discussed. I will highlight some gaps in our knowledge and suggest some ideas for future research and practice.
Professor Dick Wiggins
Co-author: Tarek Mostafa
The impact of attrition and non-response in birth cohort studies: a need to incorporate missingness strategies
This paper reveals the need to incorporate strategies to handle missing data by revealing the extent of attrition in the 1970 British Birth Cohort Study (BCS70) and how it affects sample composition over time. To begin with we illustrate the construction of inverse probability weights (IPWs) to adjust for the unit non-response. Secondly, we illustrate the impact of using IPWs and multiple imputation (MI) for an artificially constructed set of patterns of missingness for a substantive research question. Our findings show that when the predictive power of the response models is weak, the efficacy of using IPWs is undermined. Further, MI is effective in reducing the bias resulting from item missingness when the magnitude of the bias is high and the imputation models are well specified.