Correspondence between survey and admin data on quarterly turnover
Use of robust regression analysis and of mixture models to determine quarterly differences between survey and administrative turnover
Statistics Netherlands aims to benchmark monthly turnover growth rates based on sample survey data onto quarterly growth rates based on a combination of Value Added Tax (VAT) and survey data. Those growth rates are published for a range of economic sectors and underlying industries. Preliminary benchmarking results over 2015 showed that monthly growth rates were adjusted downwards in the first quarter of the year while they were adjusted upwards in the fourth quarter for the industries within the sector Retail trade.
In the current paper we investigated whether there are systematic differences between the quarterly outcomes of the survey data versus those of the VAT data. To this end, we applied a robust linear regression model as well as a linear mixture model to turnover data from 2014 and 2015. We indeed found some evidence for systematic differences, but the effects were limited. We recommend to extend the mixture model towards a more explanatory model that can describe effects of multiple groups within the population.