Estimation of Dutch census tables Jacco Daalmans History
Estimation of Dutch census tables Jacco Daalmans
History 2
Background – Virtual Census – Some Sources: ‐ Population Register (PR) ‐ Social security registers ‐ Labour force Surveys 3
Contents – Compilation method – Weighting – Results 4
1) Linkage Population Register Other datasources Compilation 2) Micro‐integration Correction of inconsistencies at unit level 3) Weighting Consistent estimation of tables 5
Weighting Variables 300, 000 Units 16, 500, 000 Sample Surveys Registers Demographic + Socio economic variables Except: education and occupation 6 Education Occupation
Repeated weighting: Basics – Developed by Statistics Netherlands – Numerical consistency ‐ Succesfully applied to 2001 Census, but many estimation problems for 2011 census ‐ Therefore repeated weighting was extended 7
Zero cell problem Categories that exist in the population, but that do not appear in a sample survey, e. g. “ 99 years old people”. 8
Solution – zero cell problem Epsilon method: – Filling in artificial ‘ghost values’ – Prevents estimation problems – But creates new problem: Implausible estimates. 9
Solution – implausible results – Step 1: Auxiliary tables • One or two‐dimensional • No epsilon method – Step 2: Demanded tables • High‐dimensional • Consistent with auxiliary tables (step 1). • Epsilon method 10
Solution – implausible results – Step 1: Auxiliary tables: (e. g. Sex x Education Age x Education etcetera) – Step 2: Demanded tables (e. g. Sex x Age x Citizenship x Education) 11
Results – All required tables have been estimated succesfully – All tables fully consistent: mutually and with registers – Little deviation from other publications (LFS) – Drawback: No variances 12
Thank you Very much! 13
- Slides: 13