RSSeNEWS to be determined...

The Social Statistics Section met in June to discuss the use of administrative data to aid statistical analysis for small geographies in the inter-censal period. The problem with small geographical areas is that, whereas surveys provide reliable estimates at national or regional level, there may be few or even no observations for local areas. However, administrative data at local level are available and may be used as covariates to produce model-based estimates or on their own as a proxy for some topic of concern.

There has been growing interest in this field in which there are considerable challenges. Issues include model selection, validation by users and experts, and the future needs of data providers and users. The role of academic experts has been of particular importance.   

The first speaker was Stephen Bond of the Office for National Statistics (ONS), who spoke about analytical tools developed in the neighbourhood statistics programme. The toolkit is designed to help the user create and understand high quality statistics and analysis for studying small area changes over time. The analysis involves five stages:

• data selection
• understanding the context
• splitting the area into groups
• understanding how the groups are changing
• investigating changes for individual small areas. 

He discussed the difficulties in assessing changes over time for small geographies, including the need to use intrinsically variable administrative data and whether or not change is more appropriately measured in terms of rates or counts.   

Denise Silva and Jane Naylor spoke about the current models and future plans at ONS for small area estimation with regard to data on unemployment, income and the census. Work on small area estimation in the ONS started with support from experts at the University of Southampton and other international academics. Early projects were the small area estimation programme in 1998 and, at the European level, EURAREA.   

Applications of the techniques included quarterly updates of local area unemployment estimates being produced for 1996/97 onwards and estimates of income at ward level. In the specific case of unemployment for individual local authorities the model used as a dependent variable the International Labour Organisation’s definition of unemployment and as covariates the jobseekers allowance claimant count and data from the labour force survey. Another important project has been the estimation of unemployment rates at parliamentary constituency level.   

The speakers explained that the production of small area estimates is usually quite challenging because data sources are not always fit for purpose and data from surveys, administrative sources and the census may need to be combined. They stated that the solution involved combining strength across different data sources.   

Keith Dugmore, who runs Demographic Decisions, led the subsequent discussion by presenting the viewpoint of the user. He stressed the need for more analysis and new datasets. He also saw a need for tracking changes over time and reported on the current interest by major retailers in spotting changes in local markets. He was concerned about possible changes to the definitions of variables or geographical areas leading to lack of continuity.   

Report by Marta Haworth