National waste survey

Hadley Hunter and Peter Jordan were asked to provide statistical support to the Environment Agency for the first National Waste Survey of 20,000 companies. They developed methods to validate the data and identify implausible values for follow up. They developed innovative ways of grossing up the data on a regional and national basis to produce results for different purposes. They also developed and ran the survey analysis programs.

Fraud detection

Allan Cole developed systems to help HMRC risk assess applications for Working Families' Tax Credit by both employees and employers. He used statistical analysis to determine the combination of factors that indicated that an employee application was potentially fraudulent. In addition, he developed a statistical model to identify potentially fraudulent applications for funding by employers. Both systems have detected and prevented fraudulent applications entering the system and led to the prosecution of employees and employers making fraudulent claims.

Forecasting pipeline capacities

Projected increases in the number of households living in the South East of England will increase the demand for water in the region. Liz Archibald was asked to provide a model to forecast future water requirements in localised sub-areas to help South East Water identify the level and location of any necessary increases in pipeline capacity. Logging data were used to establish typical per capita usages in different localities. As well as satisfying typical demand, the infrastructure also needs to be able to cope with peak water demand. Estimates of the effects of increased metering and reduced leakage were also incorporated into the model, with the provision to vary these to see the effect of different assumptions.

Client comments: “We were very happy with the model, which gave us the detailed picture we needed, identifying exactly where these population increases were going to have greatest impact and when. This has enabled us to prioritise our planned investments. And vitally, the model is flexible enough to be used under different scenarios.”