The Building Maintenance Cost Information Service (BMCIS) offers a comprehensive and rigorous framework for collecting data about the running costs of buildings. Nevertheless, it is pitched at such a level of detail that the amount of data collected and analysed may be constrained. This paper describes the deveopment and testing of a novel technique which reduces the amount of data to be collected without any unacceptable reduction in utility. It draws on the principle of cost-significance to create a simple model of maintenance and operating costs (together called running costs) from a rare and consistent set of data for 20 buildings at York University. The model contains only 11 elements, yet can predict the total running costs of each of four categories of building to an accuracy of about 21 2%. It can also predict annual costs to about 7%, despite variations in the periodicity of costs such as painting and insurance. The model was tested using the jacknife method and on virgin data. It proved to be extremely robust, predicting the running costs of 12 new buildings to within 5%. The model offers a simple framework for collecting and analysing reliable and consistent data on running costs.
