The debate on the use of appropriate statistical analyses for behavioral data is far from new, with various literature quoting theories dating back to around 1874 by the famous statistician Sir Francis Galton.1 It has however, foreseeably evolved through the ages and is now a compelling topic in the field of psychiatric medicine in the analysis of psychiatric rating scale data. Parametric statistical tests are the major methods used to analyze psychiatric rating scale data, however this is majorly viewed as methodologically incorrect.2 The issue lies, as one may already assume, in the fact that performing inappropriate statistics will discredit and invalidate the data at hand, rendering the research impractical. With the contemporary expansion of pharmacological therapy and research in the psychiatric field, now more than ever it is paramount to determine the most pragmatic standardized approach for analyzing this type of data. Based off the current available literature, this paper will discuss the argument of the levels of measurements for psychiatric rating scale data, the implications of inappropriate statistical use, and the best statistical approach for analyses.
Psychiatric rating scales are useful in assessing and determining descriptions of psychiatric disorders, diagnostic severity, and change from therapeutic interventions i.e. treatment efficacy, in clinical practice and especially in research. Just as with research of general medical practice, psychiatric data must be assessed by statistical analysis. This requires psychiatric rating scale data to be categorized in the appropriate scales of measurement to be assessed by appropriate statistical analyses. The three observational scales of measurement in research include: nominal, ordinal, and interval; each increasing in complexity and therefore a higher level of analysis. Nominal data describes a characterizing quality of a value. It labels if the value quality is present or not, but is not limited to two categories. For example, a question with the answers yes or no, a person’s gender, or hair color, is categorized as nominal data. This type of data does not measure or quantify; it is simply a qualitative description. The second scale of measurement is ordinal data. It is also known as order, or categorical data, because of how it can be categorized into an order or ranking. This allows the data to be measurable, however this ordering and ranking is only measurable in relation to itself, i.e. higher or lower, faster or slower, better or worse, and is not considered to have equidistant measurements between each difference. The most common examples of this include percentiles, Likert scales, ratings of experience, age groupings, and disease or symptom severity. Lastly, there is interval data, also known as numerical data. Interval data is used for observations that have equal intervals in the values being measured. They have a universally accepted equidistant...