Workshop on best practices for EU-SILC revision, −

Slides:



Advertisements
Similar presentations
Conducting of EU - SILC in the Republic of Macedonia, 2010 REPUBLIC OF MACEDONIA STATE STATISTICAL OFFICE State Statistical Office of Republic of Macedonia.
Advertisements

1 Developments to the Family Resources Survey Family Resources Survey (FRS) and the EU-Statistics on Income and Living Conditions (EU-SILC) Valerie Christian.
An example of longitudinal LFS weights
Riku Salonen Regression composite estimation for the Finnish LFS from a practical perspective.
1 1 Effects of attrition in the Norwegian Survey on statistics on income and living conditions Marit Wilhelmsen Statistics Norway Q European Conference.
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.
M. Fall, JP. Lorgnet et alii 26/02/2010 Individual Dynamics of Poverty, a study tackling changes in poverty in France via the SILC survey.
ICVS IN SLOVENIA Tatjana Škrbec. Content of presentation  Short history  Crime victim survey 2001 within SORS  Methodology and content of questionnaire.
Introduction Since 1995, the Municipality of Firenze designed a quarterly labour force (LF) survey, parallel to that of ISTAT, to cope with the unavailability,
USAGE OF ADMINISTRATIVE DATA IN EU-SILC SURVEY Signe Bāliņa University of Latvia.
5-4-1 Unit 4: Sampling approaches After completing this unit you should be able to: Outline the purpose of sampling Understand key theoretical.
ICES 2007 Labour Cost Index and Sample Allocation Outi Ahti-Miettinen and Seppo Laaksonen Statistics Finland (+ University of Helsinki) Labour cost index.
Building Wave Response Rates in a Longitudinal Survey: Essential for Nonsampling Error Reduction or Last In - First Out? Steven B. Cohen Fred Rohde and.
Eurostat Strategic issues: EU-SILC Agenda item 3.2 DSS meeting November 2013.
COMBINING SURVEY AND ADMINISTRATIVE DATA IN THE ITALIAN EU-SILC EXPERIENCE: POSITIVE AND CRITICAL ASPECTS National Institute of Statistics - Italy Claudio.
Introduction to Survey Research
Handling Attrition and Non-response in the 1970 British Cohort Study
EU-SILC Survey Process in the Czech Republic presentation for EU-SILC Methodological Workshop November 7th Martina Mysíková, Martin Zelený Social.
Peter Linde, Interviewservice Statistics Denmark
Martina Mysíková, Štěpán Tourek, Martin Zelený
Introduction to Survey Data Analysis
Conducting of EU - SILC in the Republic of Macedonia, 2010
The second wave of the new design of the Dutch EU-SILC: Possibilities and challenges Judit Arends.
Regression composite estimation for the Finnish LFS from a practical perspective Riku Salonen.
Planning the change to a targeted survey design
The effects of rotational design and attrition
Non-Response Bias in Income Data
Item 5.3 Feasibility studies
Data Collection and Sampling
Effect of Panel Length and Following Rules on Cross-Sectional Estimates of Income Distribution: Empirical Evidence from FI-SILC Marjo Pyy-Martikainen Workshop.
Nonresponse adjustments and calibration: a comparison between two methods to weight the Labour Force Survey Tania Borg Principal Statistician Labour Market.
Richard Heuberger, Nadja Lamei Statistics Austria
Market Research Sampling Methods.
Task Force on Victimization Eurostat, October 2011 Guillaume Osier
Sampling issues related to the implementation of EDSIM/ESIHSI
Effects of attrition on longitudinal EU-LFS estimates
3.6 Regional dimension of the poverty and exclusion indicators
Emilio Di Meglio and Emanuela Di Falco (EUROSTAT)
Directors of Social Statistics Board (DSSB) 4-5 December 2017
WORKSHOP ON CORE VARIABLES
The change of data sources in the Spanish SILC
« LFS series breaks with the adoption of the IESS FR How is Statistics Portugal planning to tackle the issue? 13th Workshop on Labour Force Survey Methodology.
Telling Canada’s story in numbers Marie-Josée Major
Istat - Structural Business Statistics
Panel care in the Austrian EU-SILC
ESDS Workshop on best practices
New Techniques and Technologies for Statistics 2017  Estimation of Response Propensities and Indicators of Representative Response Using Population-Level.
Marie Reijo, Population and Social Statistics
Implementing mixed mode questionnaire in FI-SILC
Sampling and estimation
How the Affordable Care Act Has Improved Americans’ Ability to Buy Health Insurance on Their Own Findings from the Commonwealth Fund Biennial Health Insurance.
3.3 Modernisation of Social Statistics
Directors of Social Statistics (DSS) 1-2 Mars 2018
Task Force on Small and Medium Sized Enterprise Data (SMED)
2.1 - IESS Framework Regulation update And implementation for SILC
SILC implementing regulation Overview and main text
EU-SILC Tracing Rules Implementation: Pros and Cons
FROM SCHOOL TO LABOUR MARKET PROJECTS IN ISRAEL Dalit COHEN-LERNER
2.7 Annex 3 – Quality reports
Mode effects in mixed-mode data collection WP2
Quick statistics - how to deal with quality?
Deciding the mixed-mode design WP1
Meeting of the EHIS Technical Group Luxembourg January 2012
Adaptive mixed-mode design WP1
STEPS Site Report.
ACTIVE LABOUR MARKET POLICIES AT THE INTERNATIONAL LEVEL
ACTIVE LABOUR MARKET POLICIES AT THE INTERNATIONAL LEVEL
Item 4.1: Annual labour market flows
LAMAS Working Group June 2019
Stratification, calibration and reducing attrition rate in the Dutch EU-SILC Judit Arends.
Presentation transcript:

Ongoing and future work to secure representativeness of longitudinal data in EU-SILC Workshop on best practices for EU-SILC revision, 15.− 16.5. 2019 Marie Reijo, Kaisa-Mari Okkonen

Content The Finnish EU-SILC The 4-year longitudinal component: sample size Unit non-response and sampling correction by calibration Representativeness of longitudinal component: Estimates accuracy: Variance of the estimates Estimates bias due to attrition Household splits impact Results and conclusions Actions for improving representativeness 15 May 2019

The Finnish EU-SILC 4-year rotating panel survey since 2010 Integrated cross-sectional and longitudinal components since 2010 2-phase stratified sampling design Sample units: individuals representing target persons; S-R in household Tracing: initial sample persons in-scope followed over waves, not followed after non-response Sample size (gross sample) for longitudinal component in the 1. wave: n=2500 in 2004−2009, n=5000 since 2010− Accepted sample size for 4-year panels larger since 2010 Extra sample for 2.−4. waves: not used for 4-year balanced panel 15 May 2019

The 4-year longitudinal component: sample size Figure. Accepted sample size for the balanced 4-year panel 15 May 2019

The 4-year longitudinal component: unit non-response and attrition Figure. Unit non-response and attrition for the 4-year panel, calculated from the 1. wave net sample 15 May 2019

Unit non-response correction by calibration Calibration data from total register statistics by Statistics Finland Cross-sectional component, 1.− 4. waves: Standard demographic, region and income variables Number of AROP persons was added for calibration from 2016 and persons by educational level from 2017 onwards Longitudinal component: 1. wave: cross-sectional unit non-response correction 2.−4. waves: data on longitudinal development of persons by sex*age groups 15 May 2019

Representativeness of longitudinal component (balanced 4-year panel) Estimates accuracy: Since 2013 smaller standard errors than earlier due to larger samples Estimates bias due to attrition in 2.−4. waves in 2018 The attrition is selective: e.g. 1. wave at-risk-of-poverty rate (sample) of persons is higher among those attrited and belonged to target population Sample persons All persons Weight RB050w1 RB064 Whole panel 16,8 11,9 Non attrition 2.-4. wave 14,8 10,5 11,6 Attrition 2.-4. wave 21,5 15 May 2019

Representativeness of longitudinal component (balanced 4-year panel) Figure. Coverage of certain income components compared with the total registers statistics in 2018 by age groups, % 15 May 2019

Representativeness of longitudinal component (balanced 4-year panel) Figure. Personal total disposable income estimates in 1.−4. waves of the 2018 balanced panel by sample and the total register data, 25th and 50th percentiles 15 May 2019

Sample persons attrition Longitudinal component (balanced 4-year panel): household splits impact Splitted household’s co-residents (n=150) are not followed: Source FI-SILC 2018, calculated from those persons who belonged to target population; Household splits are more common among sample-persons households that attrited. Samplew4 Attrition w2-w4 Splits Sample persons attrition Sample persons 2081 123 864 Co-residents 16+ 1688 150 1241 All persons 16+, n 3769 273 2105 15 May 2019

Longitudinal component (balanced 4-year panel): household splits impact Splitted household’s households sample members and co-residents by age group: Source FI-SILC 2018, calculated from those persons who were belonged to target population Agew1 Sample persons Co-residents Sample persons 16-17 6,2 17,6 7,8 18-24 15,6 26,4 24,2 25-64 62,2 48,4 55,4 65+ 14,9 7,6 12,6 Total 100,0 Persons 16+, N 17196 20076 115627 Persons 16+, n 123 155 Weight RB050w1 RB064 15 May 2019

Results and conclusions Representativeness of 2.− 4. waves for rotational cross-sectional component is good after calibration Instead, longitudinal component (4-year panel) includes weaknesses because of non-random attrition Calibration by age and sex does not correct change patterns between waves In calibration we follow the Doc065 guidelines Tracing co-residents in household-splits would increase costs and impact operational difficulties: no worth in relation to expected benefits for representativeness. Nationally we use longitudinal data from Total Statistics on Income Distribution (TSID) 15 May 2019

Actions for improving longitudinal component representativeness The first wave checked and improvements involved in calibration Further actions planned to be implemented primarily for interviewed data collection: Prioritizing fieldwork for those units that are difficult to contact in 1st wave sample of 2019: priority: younger sample persons in low income strata priority: not in group 1 or 3 priority: older sample persons in high income strata Fieldwork ongoing, results seen later Introducing CAWI interview as a pilot in 2020 Finding optimal field work strategy for population subgroups to maximize response rates (especially young sample persons) Testing the use of substitutes in the 2020 as part of web pilot Other actions to be considered next ? Calibration ? 15 May 2019

Expectations in the future Response rates are decreasing rapidly 1st wave: from 64.1 % in 2015 to 57.1 % in 2018 Increasing difficulties in contacting by CATI Goals of data collection set by data collection unit of Statistics Finland are not set on population group basis Communicating data quality needs and criteria to data collection unit is difficult, criteria for many purposes, e.g. longitudinal component FI-SILC production is facing multiple challenges in the next years Pressure to implement CAWI in FI-SILC soon after piloting Overall response rate and representativeness should be improved ? Large changes in register data sources: changes in tax register, real-time income register 2020 onwards Income register will have impact on survey and whole production process 15 May 2019