Handling Attrition and Non-response in the 1970 British Cohort Study

Slides:



Advertisements
Similar presentations
Handling attrition and non- response in longitudinal data Harvey Goldstein University of Bristol.
Advertisements

Handling Missing Data on ALSPAC
Non response and missing data in longitudinal surveys.
Treatment of missing values
Grandparenting and health in Europe: a longitudinal analysis Di Gessa G, Glaser K and Tinker A Institute of Gerontology, Department of Social Science,
Some birds, a cool cat and a wolf
Latino fathers’ childbearing intentions: The view from mother-proxy vs. father self-reports Lina Guzman, Jennifer Manlove, & Kerry Franzetta.
NLSCY – Non-response. Non-response There are various reasons why there is non-response to a survey  Some related to the survey process Timing Poor frame.
Why is the international agenda here to stay? Dr Mary Stiasny.
How to deal with missing data: INTRODUCTION
Modeling Achievement Trajectories When Attrition is Informative Betsy J. Feldman & Sophia Rabe- Hesketh.
PEAS wprkshop 2 Non-response and what to do about it Gillian Raab Professor of Applied Statistics Napier University.
1. Fathers in the UK Millennium Cohort Study EUCCONET Workshop Vienna 24 February 2010 Lisa Calderwood Sub-brand to go here CLS is an ESRC Resource Centre.
1. Family change in the first five years of life: new evidence from the UK Millennium Cohort Study Lisa Calderwood Sub-brand to go here CLS is an ESRC.
Following lives from birth and through the adult years Examining the truth behind the myth of the 'the Monstrous Army on the March' Dylan.
1. Measuring the Impact of Universal Preschool Education and Care on Literacy Performance Scores. Tarek Mostafa Institute of Education – University of.
Data Collection, Harmonisation and Storage (An international perspective) Jon Johnson (CLS, Senior Database Manager) Sub-brand to go here CLS is an ESRC.
CHAPTER 6, INDEXES, SCALES, AND TYPOLOGIES
1 Introduction to Survey Data Analysis Linda K. Owens, PhD Assistant Director for Sampling & Analysis Survey Research Laboratory University of Illinois.
Experiences of managing Birth Cohort Data at CLS Jon Johnson (Senior Database Manager) Sub-brand to go here CLS is an ESRC Resource Centre based at the.
Handling Attrition and Non- response in the 1970 British Cohort Study Tarek Mostafa Institute of Education – University of London.
Imputation for Multi Care Data Naren Meadem. Introduction What is certain in life? –Death –Taxes What is certain in research? –Measurement error –Missing.
1 Introduction to Survey Data Analysis Linda K. Owens, PhD Assistant Director for Sampling & Analysis Survey Research Laboratory University of Illinois.
SW 983 Missing Data Treatment Most of the slides presented here are from the Modern Missing Data Methods, 2011, 5 day course presented by the KUCRMDA,
The Impact of Missing Data on the Detection of Nonuniform Differential Item Functioning W. Holmes Finch.
AN EXAMPLE OF COOPERATION & SOME WIDER ISSUES Ian Plewis (Bedford Group, Institute of Education) & Stephen Morris (Social Research Division, Department.
3.14 X AXIS 6.65 BASE MARGIN 5.95 TOP MARGIN 4.52 CHART TOP LEFT MARGIN RIGHT MARGIN ©TNS 2013 Are ‘better’ interviewers more successful at.
Missing Values Raymond Kim Pink Preechavanichwong Andrew Wendel October 27, 2015.
Item-Non-Response and Imputation of Labor Income in Panel Surveys: A Cross-National Comparison ITEM-NON-RESPONSE AND IMPUTATION OF LABOR INCOME IN PANEL.
Life course partnership status and biomarkers in mid-life: Evidence from the 1958 British birth cohort George B. Ploubidis, Richard J. Silverwood, Bianca.
DATA STRUCTURES AND LONGITUDINAL DATA ANALYSIS Nidhi Kohli, Ph.D. Quantitative Methods in Education (QME) Department of Educational Psychology 1.
Research and Evaluation Methodology Program College of Education A comparison of methods for imputation of missing covariate data prior to propensity score.
Attrition and Selection of alteri Respondents in the pairfam panel
Missing data: Why you should care about it and what to do about it
SESRI Workshop on Survey-based Experiments
Factors Influencing Response Rates to ALSPAC Questionnaires
James Law Professor of Language and Communication Science and Director
Understanding Non Response in the 1958 Birth Cohort
Antidepressant Use Among Working Age Canadians:
Rose Krebill-Prather, PhD
The Centre for Longitudinal Studies Missing Data Strategy
Obtaining information on non-responders: a development of the basic question approach for surveys of individuals Patten Smith (Ipsos MORI) Richard Harry.
SESRI Workshop on Survey-based Experiments
Maximum Likelihood & Missing data
Introduction to Survey Data Analysis
Multiple Imputation.
Multiple Imputation Using Stata
The second wave of the new design of the Dutch EU-SILC: Possibilities and challenges Judit Arends.
SESRI Workshop on Survey-based Experiments
The European Statistical Training Programme (ESTP)
Dealing with missing data
Does cognitive ability in childhood predict fertility
The effects of rotational design and attrition
Non-Response Bias in Income Data
Presenter: Ting-Ting Chung July 11, 2017
Effect of Panel Length and Following Rules on Cross-Sectional Estimates of Income Distribution: Empirical Evidence from FI-SILC Marjo Pyy-Martikainen Workshop.
The bane of data analysis
The European Statistical Training Programme (ESTP)
ELM DICIPE Mozambique Gaza, Nampula, and Tete Midline 2016
Non response and missing data in longitudinal surveys
Analysis of missing responses to the sexual experience question in evaluation of an adolescent HIV risk reduction intervention Yu-li Hsieh, Barbara L.
Inference for Sampling
Chapter 4: Missing data mechanisms
The European Statistical Training Programme (ESTP)
Rachael Bedford Mplus: Longitudinal Analysis Workshop 23/06/2015
An Introduction to the 1970 British Birth Cohort (BCS70)
Clinical prediction models
Implementation of the Bayesian approach to imputation at SORS Zvone Klun and Rudi Seljak Statistical Office of the Republic of Slovenia Oslo, September.
Chapter 5: The analysis of nonresponse
Workshop on best practices for EU-SILC revision, −
Presentation transcript:

Handling Attrition and Non-response in the 1970 British Cohort Study Tarek Mostafa Institute of Education – University of London

Challenges to Longitudinal Surveys Statistical analyses face a number of challenges: Unit non-response. Item non-response. Attrition over time in longitudinal surveys. Bias caused by missing data when missingness is not MCAR. Aim of the paper: Examine non-response in the 1970 British Cohort Study (BCS70) and explore the use of non-response weights and imputation to deal with attrition and item missingness

British Cohort Study 1970 BCS70 follows the lives of 17,000 people born in a single week in April 1970. Individuals were surveyed at birth and then at age 5, 10, 16, 26, 30, 34 and 38. The most recent wave at age 42 will be made available soon. BCS70 collects data on health, physical, educational and social development, and economic circumstances among other factors. Pattern Frequency Percentage Monotone 5,277 30.53 Non monotone 8,287 47.95 Non missing 3,720 21.52 Total 17,284 100.00

Response Categories Wave 1 Birth Wave 2 Age 5 Wave 3 Age 10 Participated 16,569 12,939 14,349 11,206 8,654 10,833 9,316 8,545 Contact later 715 2,859 1,122 3,372 4,872 507 631 Dead 567 587 597 697 747 793 807 No contact later 919 1,226 2,109 2,830 3,077 4,001 Temporary Emigrant 49 32 41 Permanent Emigrant 255 365 418 Refusal 141 1,292 1,334 687 Unproductive 541 803 2,048 Not issued 4,779 Total 17,284

Sample Composition Over Time

Sample Composition Over Time

Modelling Non-response in BCS70

Modelling Non-response in BCS70

Modelling Non-response in BCS70

Modelling Non-response in BCS70

How effective are weights and imputations. Two approached are used to deal with attrition and item non-response: Attrition and non-response weights. Imputation techniques. Attrition weights: weak predictive power, no solution to item missingness, constructed using restrictive models, reduction in sample size especially when using data from different waves. Imputations: Treatment of both unit and item non-response, can be tailored according to the need of the researcher. Both techniques require knowledge of the process behind missingness.

Simulation Study Use a substantive model with Dependent variable: literacy scores at age 16 Independent variables: gender, age 10 gross family income per week (wave 3) and highest parental qualification (wave 4). Simulation: 1- construct inverse probability weights for wave 4. These weights will adjust for attrition in wave 4. 2- On literacy scores, introduce 10% missing values completely at random. 3- We recode the father’s social class into a binary variable with two categories Manual and non-manual. 4- On income and highest qualification, we introduce 40% missing values if the father’s social class is manual and 10% if it is non-manual. 5- We don’t introduce any missing values on gender.

Simulation Study

Models to estimate Model 1: is estimated using the sample with complete cases (C) with non-response weights to adjust for the bias resulting from unit non-response (A-B). Model 2: is estimated with the sample with complete cases (C) but without applying the non-response weights. Model 3: is estimated using the simulation sample (D) with listwise deletion. Model 4: is estimated using the simulation sample (D) with unit non-response weights. Model 5: is estimated using 20 imputed datasets that restore the sample size back to (C). Model 6: is the most complete model and is estimated using 20 imputed datasets that restore the sample size back to (C) in conjunction with unit non-response weights.

Results

Results

Conclusions Men from lower social backgrounds and with less educated parents are less likely to respond. The predictive power of the models is weak. Non-response weights don’t improve the estimates or their standard errors by much when data loss is due to item missingness. random multiple imputations are efficient in reducing the bias resulting from item missingness both in terms of estimates and standard errors with some exceptions. The efficacy of weights and imputations in dealing with bias resulting from unit non-response and item missingness depends on the extent of bias and whether variables correlated with the probability of unit and item non-response can be found.

Thank you for your attention The Centre for Longitudinal Studies www.cls.ioe.ac.uk Tarek Mostafa T.Mostafa@ioe.ac.uk Institute of Education University of London 20 Bedford Way London WC1H 0AL Tel +44 (0)20 7612 6881 Fax +44 (0)20 7612 6126 Email info@ioe.ac.uk Web www.ioe.ac.uk