The accuracy of growth rates with classification errors

In this paper, an analytical method and a bootstrap method are developed to evaluate the effect of classification errors on the bias and variance of estimated domain growth rates. The approaches are illustrated by a case study on the effect of errors in the classification of businesses by main economic activity (NACE code) on quarterly turnover growth rates for the Dutch car trade sector.

Producing reliable, undisputed statistical figures is the backbone of national statistical institutes. When administrative data are used in the production of statistics, the accuracy is often mainly determined by so-called non-sampling errors. For statistics that are published by domain, classification errors in the domain codes are an important example of a non-sampling error. Quantifying the effect of non-sampling errors on statistics is often difficult in practice.

In previous work we have developed approaches to evaluate the effect of classification errors on the bias and variance of estimated domain totals, both analytically and by means of a bootstrap method. In this paper, we extend both approaches to estimated growth rates. Here, a more complicated model for classification errors is needed to account for the relation between errors made at different points in time. An important practical question is how to estimate the unknown parameters of this classification error model. We illustrate this in detail for a case study on the effect of errors in the classification of businesses by industry (NACE code) on the bias and variance of quarterly turnover growth rates for the Dutch car trade sector.