Online Program

Return to main conference page

All Times ET

Program is Subject to Change

Wednesday, June 16
Wed, Jun 16, 1:30 PM - 3:30 PM
The Future of Establishment Surveys Is Easy as ABC: AI, Big Data, Cutting-Edge Survey Design

Methods for Estimating the Quality of Multisource Statistics (307947)

Ton De Waal, Statistics Netherlands (CBS) 
Sander Scholtus, Statistics Netherlands (CBS) 
*Arnout Van Delden, Statistics Netherlands (CBS) 

Keywords: bias, variance, output accuracy, representation errors, measurement errors, linkage errors

With the increasing availability of data, official business statistics are more often based on multiple data sources. Evaluating accuracy, i.e. bias and variance, of output based on multiple sources has therefore become an important topic. Estimating the accuracy is important to inform users about data quality and it can be a trigger to adjust processing steps when accuracy drops below an acceptable level. An inventory of methods to estimate output accuracy of multisource statistics has been made in the European project KOMUSO. The bias and variance of multisource statistics is affected by errors on the representation side (units and populations) and by errors on the measurement side. Additionally, when combining sources at micro level, linkage errors may occur. We will introduce recently developed methods to estimate bias and variance of outputs as affected by representation error, linkage error and measurement error, illustrated by examples for business statistics