National Statistics Methodology Advisory Committee
Fourth meeting, 20 March 2003

The following work in progress papers were considered.

Paper 1: Methodology for selective editing
ONS has been developing methods for selective editing for business surveys in order to improve the efficiency of data editing processes without reducing data quality. Selective editing seeks to concentrate the checking of suspicious data on cases that will have a material effect on the survey estimates. This paper describes the process of implementing these methods on a variety of surveys and sketches out a general process for developing a selective editing methodology.

Committee Conclusions:

• Selective editing is only appropriate for business surveys;
• It is unrealistic to try to build a general selective editing system, it may be better to have a general philosophy rather than a general method;
• It may be worth considering the use of probabilities of selection for editing, rather than a rigid threshold.

Paper 2: Integrating data on work-related ill health from different sources to measure progress against targets
The Health and Safety Executive is committed to providing robust measures of progress against national Government targets for improving health and safety at work. One strand of this is the integration of data from different sources to assess whether the target for work-related ill health has been met. The starting point for this is a model of how work-related ill health is caused, perceived and then reported and attributed to work causes. This paper examines a 'fully quantitative' approach and a 'semi-quantitative' approach to this problem.

Committee Conclusions:

• It is important to use consistent measurement methods across the target period;
• Opportunities for the measures to be manipulated should be minimised;
• The 'fully quantitative' approach could be developed further by looking at data fusion and meta-analysis models;
• Given the complexities, it might be better to record a measure of change from each source and combine these changes into an overall judgment, as proposed in the 'semi-quantitative' approach.

Paper 3: Implementation of small area estimates
There is strong demand for information about income at the small area level. Following the decision not to include a question on income in the 2001 Census, ONS undertook work to explore alternative means of providing income data for small areas. This paper gives a brief overview of modelling techniques developed by the Small Area Estimation Project within ONS methodology group, to produce synthetic estimates for wards. The paper then describes how the methodology has been implemented to develop ward-level estimates of average household income for England, Wales and Scotland.

Committee Conclusions:

• It would be possible to extend the basic methodology to include group specific random effects;
• The requirement for the bias adjustment due to back-transformation could be avoided by simulating from the distribution on the transformed scale then back-transforming and computing estimates;
• The distributions of covariate values can be examined to investigate if London values are grouped at either extreme;
• The possibility that variability in income may be different in London should be examined;
• Some spatial smoothing could be introduced using the nearest neighbour approach;
• The first principle component should be used in the stability analysis.

Paper 4: Strategy for sample design and estimation within ONS
The Sample Design and Estimation centre in ONS is responsible for methods for sample design, allocation and selection; use of frames; estimation; outlier detection and treatment; and variance estimation within ONS. The centre currently does not have sufficient resources to be able to contribute to all these areas. This paper describes plans to increase resources over the next three years and proposes a new way of working for the centre.

Committee Conclusions:

• It may be beneficial for the centre to advise on sample design and estimation across the GSS in the future, but it is important to create a strong SD&E centre in ONS before considering other departments;
• In the future there will probably be a shift to making greater use of administrative data, rather than using sample surveys;
• It is important to have policies for training and retaining methodologists.

Next meeting due 29 September 2003

This page last revised: Tuesday, 24 June 2003