Statistical Publication Notice
Evaluation of Statistical Techniques Used in the Scottish Index of Multiple Deprivation (SIMD) 2004
The Scottish Executive today released the evaluation of the statistical techniques in the SIMD 2004. The report can be found at www.scotland.gov.uk/simd2004evaluation. The Scottish Executive's Long Term Strategy for Measuring Deprivation included a commitment to evaluate the statistical techniques used in SIMD 2004. Five areas were highlighted for review: shrinkage, factor analysis, exponential transformation, weights for combining domains and confidence intervals on ranks. The SIMD 2004 methodology was developed by Oxford University and is being used by the four UK administrations to identify the most deprived areas within each country.
Over the summer we contracted the Robertson Centre for Biostatistics (RCB), University of Glasgow to conduct the evaluation. The evaluation has been used to decide whether we need to adjust any of the statistical techniques used in the construction of the SIMD 2006 which is due for release in October 2006. The SIMD 2004 methodology passed its 'health check'. The extensive RCB research is valuable and has helped answer many of the questions asked in the long term strategy.
We have considered the RCB research very carefully and after discussions at the Scottish Executive's Measuring Deprivation Advisory Group, we have agreed an outline methodology for SIMD 2006:-
- The SIMD 2006 will not include shrinkage. It was shown to have little effect on the resultant indices and by shrinking towards local authority averages, introduces a very small bias against data zones in otherwise less deprived areas.
- We will not be producing confidence intervals around the SIMD data zone ranks. The extensive RCB research compared the SIMD 2004 methodology with a range of other methods. It showed that each method identified essentially the same set of data zones as being most deprived.
- SIMD 2006 will continue to use Maximum Likelihood Factor Analysis and exponential transformations. It is recognised that the more advanced and complex factor analysis approach (recommended by RCB) could allow confidence intervals to be produced. However, given that the confidence intervals around the ranks is small, and that the current approach is working, we have decided to retain the current approach for continuity and operational reasons.
- The domain weights employed in SIMD 2003 and 2004 were arrived at through expert value judgement and are fit for purpose. The RCB research showed the set of deprived data zones identified was relatively insensitive to the choice of domain weight. This is not surprising given the high correlations between the SIMD domains. We will continue to use the expert value judgement approach in SIMD 2006.
We welcome any comments or suggestions that help us to improve SIMD. Robert Williams and Tracey Stead at the Scottish Executive can be contacted on 0131 244 0442 or email@example.com.