Understanding variability in performance indicators

It is very unusual for a performance indicator to remain constant over any reasonable time period. When assessing performance, we need to know whether the differences seen from one period to the next are a sign of real change or are merely the result of variation that can be expected. Wheeler (1993) is a very readable book that suggests practical ways to understand and interpret variability in data. Wheeler argues that the output from any managed process will always display some variability, which means that performance through time must be interpreted very carefully. Wheeler provides several examples that clearly demonstrate the danger and difficulty in knowing whether apparent performance improvements are genuine or just random variation. This is an important question at all levels in the public sector, whether we are concerned with national economic performance or the micro performance of a single programme. For example, as this book is being written, economic commentators are sharing their views on the state of the UK economy. The UK’s Office of National Statistics has just published its estimate of growth in Gross Domestic Product (GDP) for the first quarter of 2010. The released figure, which may be later revised, is 1.1 per cent, which is larger than expected. Despite the excited comments of TV pundits and serious academics, no one seems to know whether this is a real improvement or just within the expected range of variation for this type of economic statistic.

Like other writers, Wheeler suggests that variation though time can be separated into two elements. The first is common cause variation, sometimes known as noise or random variation. It has many different causes that include poorly defined operating procedures, measurement '... errors and wear ...' and tear in equipment. In the case of many public services, we must add the sheer variability in the cases with which staff must deal. Common cause variation can be reduced and should be reduced to a minimum. However, doing so can be expensive and may not be worth it if the cost is excessive. Special cause variation, often known as the signal, is usually caused by a change in the system that is being monitored. It indicates a real shift in performance and its detection is vital to the proper use of performance indicators.