The measurement of people's preferences has often been accomplished by the use of rating scales. For example, in much of the work investigating the change in visual amenity resulting from the reconstruction of roads, people's preferences for scenes were measured using a 'pleasantness' rating scale. This report considers two problems which arise from this rating procedure: (a) what reasons are there for deciding whether to use the mean or the median when a measure of central tendency is required, and (b) when a group measure of preference is required, what is the appropriate number of subjects to use (i.e. sample size)? data obtained from a semantic differential type of rating scale are at least ordinal and the median is probably the most appropriate measure of central tendency. Often the rating distributions depart markedly from normal or are greatly skewed. Under such circumstances the median will be a more representative measure of central location than is the mean. However, TRRL survey data showed that the difference between the mean and the median is unlikely to exceed half a scale unit, provided sample sizes of about 30 subjects are used. There is substantial variability in sample means and medians with very small samples, but small increases in sample size beyond about 30 lead to only small reductions in this variability. (Author/publisher)
Abstract