SAE 2016: Small Area Estimation in current statistical practice
SAE 2016 was attended by 125 participants from 34 countries. Over half of them represented a university while one-third worked for a national statistical office. The focus of this year’s programme was on implementation of the theory on small area estimation at NSIs. Small area estimation is defined as methods which use statistical models to produce reliable figures for detailed disaggregation. This is not possible through conventional (design-based) sampling techniques. This particular research field is becoming increasingly popular. Prof. Jan van den Brakel, who works for both CBS and UM: ‘Until now, research activities were predominantly confined to universities. It is time to start putting the theory into practice more often.’
Contributions by CBS
The potential benefits of this application were demonstrated at the conference by CBS researchers Harm Jan Boonstra and John Michiels. Their presentation showed how SAE was used in the CBS Labour Force Survey. CBS has been employing this method for several years, among others to produce monthly labour force data. Using the same method, annual labour force data can be broken down by municipality. Van den Brakel spoke to the attending participants about his research into alternative ways of producing monthly unemployment figures at the provincial level.
Other presentations carried topics including the use of big data to produce regional figures. The big data could be a direct source, but could also be auxiliary variables in a model which employs sampling data to produce accurate estimates. In Italy, for example, Twitter data are used to estimate household income more accurately, and GPS data to estimate traffic intensity and travel time. These applications are similar to what is being researched at CBS at the moment. Another valuable contribution was the presentation by Prof. Danny Pfeffermann, Director-General of the Israeli Central Bureau of Statistics. He argued that rising non-response rates are given too little attention but result in skewed data, and presented a method to adjust figures for this. Prof. Ralf Münnich of Trier University explained how poverty levels in border regions can be estimated by aggregating data from neighbouring countries.
Another impressive presentation contained an outline of what has been accomplished with small domain estimates in the United States, Van den Brakel says: ‘They have been working with SAE since the 1990s already, mainly in their production of data on unemployment, health and poverty.’ In Europe, the UK official statistics bureau stands out in applying SAE in official statistics: ‘They too have been using SAE to produce unemployment figures at the sub-regional level for some time.’
For the first time, the conference was co-organised by an NSI. The responses from participants were positive. ‘I am glad to see that the application of small area estimation has now come to the forefront; and we see more and more statistical bureaus becoming interested’, Van den Brakel says. Next year, the conference will be organised by the National School for Statistics and Information Analysis (ENSAI) in Paris, and CBS will again take part in the organisation.
Aims and scope
Small area estimation (SAE) plays an important role in the field of information provision about modern societies using survey sampling. SAE methods enable reliable estimation at detailed disaggregation levels where sample sizes are too small to apply direct estimation techniques.
The main purpose of organizing a conference on small area estimation is to assess the current state of development and usage of small area methodology. This conference continues a series of conferences on small area estimation that have been organized annually at different places around the world. It will give researchers and practitioners from all over the world an opportunity to exchange information or learn about state-of-the-art small area estimation techniques. So far small area estimation has predominantly been an academic area of research. The aim of this conference is to give more attention to applications and implementation in government agencies. Our intention with this conference is to have several invited sessions on focused research areas including a session on applications in official statistics.