Guidelines to statistical evaluation of data from rating scales and questionnaires.

Questionnaires and rating scales are commonly used to measure qualitative variables, such as feelings, attitudes and many other behavioural and health-related variables. There are different types of instruments ranging from single scales to multidimensional, multi-item questionnaires. The scaling of the responses can vary from the dichotomous alternatives “yes” and “no” to a mark on a line, as in the visual analogue scale (VAS). Numerical labels are commonly used for the recordings. Nevertheless, irrespective of the type of scaling, the item responses indicate only an ordered structure and not a numerical value in a mathematical sense. Such data are often called ordered categorical or ordinal (1–4). Statistical methods for data from rating scales must take account of the rank-invariant propertiesof ordinal data, whichmeans that the methodsmust be unaffected by a relabelling of the scale categories. Hence, statistical methods applicable to data from rating scales differ completely from the traditional methods for quantitative variables, since calculations based on adding or subtracting ordinal data are not appropriate. Sum scores of multi-item assessments, the mean value, standard deviation and calculation of differences for description of change in score do not have an interpretable meaning and must be avoided in the statistical evaluation of data from rating scales and questionnaires (4–6). Traditionally, in applied research, there is a temptation to treat data from rating scales as numerical on an interval level (4, 5). It should be emphasized, however, that data on an interval level are quantitative, which means that such data have the mathematical properties of well-deŽ ned size and equidistance, but the same variable does not have the same ratio when it is measured in different units (1). Hence, qualitative data could never gain the properties required for being treated as interval data. Statistical methods for quantitative data are valid only when data have the mathematical properties of well-deŽ ned size and distance, and conclusions drawn from such analyses are solely interpretable and reliable for quantitative data. However, quantitative data such as blood pressure could be treated as ordinal when categorized as “low”, “normal” and “high”. Such categorization changes the choice of appropriate statistical methods of analysis. GUIDELINES FOR STUDIES INCLUDING RATING SCALES AND/OR QUESTIONNAIRES