[Choice and consequence: measurement level determines the statistical tool-box]

Abstract

Various types of data are used in health and behavioural sciences. Some of the variables, such as blood variables, can be measured by standardised, calibrated methods while others are based on subjective judgements from an expert or from the patients. There is a strong link between the measurement process and the choice of statistical toolbox. The statistically important measurement levels are; categorical, ordinal, quantitative discrete and quantitative continuous data. Special attention should be paid to non-negative quantitative data, and to ordinal data which are very common in medicine. In this paper the properties of the measurement levels are given and their consequences on the choice of statistical methods for description and analysis.