Improving the timeliness of SILC SILC workshop 2018 - Warsaw
Why – three arguments? Because we could Because the upcoming regulation say we must do so SILC is the most important of the Eurostat Statistics – it should be available as soon as possible
What has been accomplished? Month of data delivery (cross sectional SILC) SILC 2012 N+21 (September N+1) SILC 2013 N+21 SILC 2014 N+20 SILC 2015 N+17 SILC 2016 N+18 (June N+1) SILC 2017 N+11 (November year N) SILC 2018 N+10 (October year N)
Identify bottlenecks Late availability of registers (mainly tax-data) Mostly only one person could work at SILC at anytime Manual time consuming tasks (Enter questionnaires, ISCO-coding etc.) Unnecessary complexity in programs Lack of structure and naming standards Other statistics and tasks had higher priority than SILC data processing
Solutions Faster income statistics – sometimes settle for earlier date of reference temporarily. Modularize and use double manpower, when data is available -> shorter production period Abandon Paper questionnaires Structure and document Impose naming standards, automate and remove hard-coding where possible. Always give data processing high priority
The old SILC calendar
The new SILC calendar - modularized
Intuitive folder structure
Automatization (1)
Automatization (2)
Automatization (3) - ISCO Respondent entered job-position ”I.E Nurse”. These were only coded manually for every respondent. We introduced: - Use of registers for currently employed Automated search for job-positions in the job database Improved software for the ones that need to be coded
Simple Proces documentation Q:\Systemdokumentation\A1 Personstatistik\K10\SILC
Catch up period is tough In order to catch up, you need to process SILC with less than a year in between each SILC-year. Extra manpower and/or less alternate tasks are required in the catch up period. We Got Inez on the job Replaced annual publication on incomes with less time consuming web-based documents
Future targets Cross-sectional / preliminary SILC delivered by N+10 Longitudinal/Final SIILC delivered N+11
Work ahead for Final SILC Longitudinal Weighting procedure: Reduce complexity, improve program structure, remove all remaining hard-coded parameters from input programs. Challenge: Small sample and attrition bias leads to low quality of longitudinal SILC. It’s so poor that production of longitudinal SILC should not get priority over Dissemination of income statistics Dissemination of cross-sectional SILC Increased efficiency should allow production of final SILC without causing significant delays to the dissemination.
Tips / summary Retain data as fast as possible without compromising to much on quality Modularize production – so more people can work on the production of SILC at once in busy periods Remove or improve on slow manual processes Be aware of trade offs between time spend and value of going into details.