# Variance estimation for complex indicators of poverty and inequality using linearization techniques

• Guillaume Osier Statistical Office of the European Communities (EUROSTAT), Luxembourg
Keywords: EU-SILC survey, non-linear statistics, influence function, standard error, confidence interval, design effect coefficient

### Abstract

The paper presents the Eurostat experience in calculating measures of precision, including standard errors, confidence intervals and design effect coefficients - the ratio of the variance of a statistic with the actual sample design to the variance of that statistic with a simple random sample of same size - for the "Laeken" indicators, that is, a set of complex indicators of poverty and inequality which had been set out in the framework of the EU-SILC project (European Statistics on Income and Living Conditions). The Taylor linearization method (Tepping, 1968; Woodruff, 1971; Wolter, 1985; Tille, 2000) is actually a well-established method to obtain variance estimators for nonlinear statistics such as ratios, correlation or regression coefficients. It consists of approximating a nonlinear statistic with a linear function of the observations by using first-order Taylor Series expansions. Then, an easily found variance estimator of the linear approximation is used as an estimator of the variance of the nonlinear statistic. Although the Taylor linearization method handles all the nonlinear statistics which can be expressed as a smooth function of estimated totals, the approach fails to encompass the "Laeken" indicators since the latter are having more complex mathematical expressions. Consequently, a generalized linearization method (Deville, 1999), which relies on the concept of influence function (Hampel, Ronchetti, Rousseeuw and Stahel, 1986), has been implemented. After presenting the EU-SILC instrument and the main target indicators for which variance estimates are needed, the paper elaborates on the main features of the linearization approach based on influence functions. Ultimately, estimated standard errors, confidence intervals and design effect coefficients obtained from this approach are presented and discussed.
Published
2009-12-23
How to Cite
Osier, G. (2009). Variance estimation for complex indicators of poverty and inequality using linearization techniques. Survey Research Methods, 3(3), 167-195. https://doi.org/10.18148/srm/2009.v3i3.369
Issue
Section
Articles