1. Quality – 'Fitness for purpose'
The quality of a statistical product can be defined as the fitness for purpose of that product. More specifically, it is the fitness for purpose with regards the following dimensions: relevance, accuracy, timeliness and punctuality, accessibility and clarity, comparability, and coherence.
2. Quality Management – 'encompassing approach to quality work'
Quality management provides the organisation with an encompassing approach to quality work. It focuses on the full statistical process, and aims to improve quality and coordinate quality initiatives. It also encourages and promotes a culture of continuous improvement, self-assessment and quality reviews.
For the purposes of this work, 'quality management' includes:
the Principles governing our work, as set out in the UK Code of Practice for Official Statistics
the coordination of ongoing quality initiatives that ensure we are Code compliant (e.g. quality reporting, evaluation activities, etc)
work that builds on existing standards and best practice
the measurement, monitoring and management of data quality on a day to day basis
auditing of statistical processes
ensuring that all staff are sufficiently equipped to produce quality outputs (e.g. that appropriate training and guidance exist)
promoting a culture of systematic quality improvement work
3. Quality Reviews
Reviews may be internal or external and cover the processes or outputs of official statistics, or both. Review is part of the cycle of improvement and aims to identify areas for further examination and improvement. Reviews are a way to audit compliance with the Code of Practice Principles.
4. Quality Assurance – 'anticipating and avoiding problems'
Quality assurance covers all procedures focused on providing confidence that quality requirements will be fulfilled, and anticipating problems. It requires processes and systems in place that are planned and tested to perform under all conditions, and to self-correct or flag problems under exceptions. The goal of quality assurance is to prevent, reduce or limit the occurrence of errors in a statistical product and, therefore, to get it right first time.
Quality assurance is about creating evidence that errors have not slipped through. For example, testing survey questions to demonstrate that interviewers and respondents understand the concepts and vocabulary, providing 'other' boxes so that exceptional answers will not be forced into mis-codings, providing clear routing so that all relevant questions are asked (and not others), and including triangulations for consistency checking.
5. Quality Control – 'responding to observed problems'
Quality control is directed only at what can be measured and judged acceptable or not; if measurement is not possible, then quality control cannot be performed. Quality control is used to measure actual performance, compare it to standards and act on the difference, thus it only focuses on accuracy. Quality control is most commonly applied at the process stage of a survey to work that is typically performed by persons with various levels of training and ability, and where the task is repetitive and manual. It therefore applies to activities such as coding, data capture and editing.
Quality control is best used to define the quality of the process and identify causes of failure, which informs quality management on where to improve quality assurance so that the causes are mitigated and the process overall improved. For example, measuring the response rate to a survey should lead to investigation of reasons for 'missingness' which might suggest an alternative delivery or incentive to reply.
Quality control checks might be applied at the data capture and validation stages where automated checks compare new values to previous ones and throw out those that look suspect. The suspect values would then be looked at manually which would lead to quality improvements over time.
Process quality measures might be applied to certain processes as a quality control check (e.g. editing hit rates or response rates). Values would be compared back to those previously obtained from applying the measure, allowing a survey manager to react where values are dropping.