Reliable qualitative data for safety and risk management

Many industries collect event data on technical, operational and human factors failures, which include short textual or 'qualitative' reports. These reports, whilst often rich in information about the motives and intentions of staff, offer a particular analysis problem, and consequently they often accumulate in filing cabinets with their potential as decision-making aids unrealized. In this paper we will argue that textual (or verbal) accounts are quite acceptable as evidence in safety management, provided they are dealt with rigorously and systematically. The watchword for what should or should not be accepted into risk models or safety databases is reliability. Reliable data generated from 'the things people say' should not be seen as inherently 'less worthy' than reliable data from any other source. In fact, once reliability is established, frequency output from coded discourse can be treated much as frequencies from any engineered system, and are just as amenable to statistical manipulation. The result of taking the time to analyse such data fully is an integrated risk management system where the 'soft' versus 'hard' data distinction is replaced by a 'reliable' versus 'unreliable' data model, this paradigm offering maximum benefit from data collected.