Abstract

The current scope of statistics produced from LFS data is limited by its sample size and administration costs, as well as the fact it is becoming increasingly harder to collect data from respondents who are less and less willing to participate in surveys. However, users of labour statistics, both national and international, expect more detailed statistics, including regional breakdowns, that cannot be directly produced from the survey.

This problem can be overcome by the application of small area estimation methods, which make it possible to obtain reliable estimates at lower levels of spatial aggregation or for more detailed domains. Direct estimation is usually the starting point for SAE techniques. If the sample size for a given area/domain is not sufficient to produce reliable direct estimates, such an area/domain is regarded as “small”. As indicated by Molina and Rao (2015), the problem of low estimation precision can be solved by exploiting auxiliary information from other sources of data, like censuses, administrative registers or big data.

One commonly used technique in small area estimation is known as Structure Preserving Estimation (SPREE). It was first proposed by Purcell and Kish (1980), and later developed by Zhang and Chambers (2004) and Luna Hernandez (2006). SPREE estimators retain the association structure from the contingency table calculated using a source of auxiliary data while recreating marginal totals of reliable direct estimates from a sample survey. The main advantage of this approach is that the resulting estimates at a lower level of spatial aggregation add up to direct estimates at a higher level of aggregation, which is a very important property in the case of official statistics.

SPREE estimation is commonly used in different countries e.g. to produce labour market or population statistics. For example, it is used by Jobs and Skills Australia to produce regular statistics about the number of unemployed persons, the unemployment rate and the economically active population at lower level of spatial aggregation.

The main purpose of the presentation is to demonstrate the possibility of applying a SPREE estimator and its generalisations to estimate the number of people representing the main labour market categories (employed, unemployed and economically inactive) for domains in which the number of sampled representatives in the LFS is not sufficiently large to produce direct estimates with an acceptable error of estimation. The proposed approach relies on data from administrative registers, which are used as a source of auxiliary information. The authors will also discuss possible applications of SPREE to produce estimates for non-standard areas, such provincial capital cities or urban functional areas.