Combining Administrative data sources Statistics Belgium - - PowerPoint PPT Presentation

Robust Variable Estimation by Combining Administrative data sources

Statistics Belgium Guy.Vekeman@economie.fgov.be

Information sources and prior work:

• 1. MEETS ESSNet project on Use of Administrative Data in Business Statistics
• 2. Monobenificiary grant: “Improving the Robustness of a Ratio Imputation

Scheme”

Structural Business Statistics

• Admin data:

– VAT Tax data – Profit and loss accounts (P&L) – Social security data on employment & personnel cost

• Survey data: mainly breakdowns

– Cost items – Revenue specifications

• Combining the two => Statistical variables

Quality issues with admin. data

• VAT turnover/cost ≠ Accounting turnover/costs
• P&L acc’s for SME’s: TO/Csts not compulsory
• P&L acc’s for self-employed: not compulsory

Missing info in 2 non-survey years imputed:

• Previous survey data of the respondent
• Former (survey year) and new administrative data

SMEs: reconciliation of admin proxies derived from VAT with others from P&L acc’s

Para-data: monitoring process (For 3 ‘user groups’: )

• Statistical process analysts: ‘cryptic info’ on coherence checks/ &

frequency of use of program branches and subroutines

• For statisticians: E.g. end-of-process reporting on 1) ‘frequency of

anomalies’ in microdata or 2) origin/type of admin proxies used

• For methodologists: E.g. Distribution analysis on match between

VAT admin proxy and observed variables.

Obs COMMENT COUNT 10 FTE>50 and Pers.Cost exceeds Val.add. 629 23 Hours Worked/FTE < 1200 175 24 Hours Worked/FTE > 2200 13 25 Hours Worked/FTE > 2600 3

Better matching proxy for VAT ‘purchases’: bias removed, skewness & dispersion reduced …