Levels of Measurement [AL]

Levels of Measurement

Not all numerical data are equal. The level of measurement of a variable — the mathematical properties of the scale on which it is measured — determines which statistical procedures are appropriate, which descriptive statistics are meaningful, and how results should be interpreted. Stanley Stevens (1946) proposed four levels of measurement: nominal, ordinal, interval, and ratio. In psychology, three are most commonly distinguished.

Nominal Level

Nominal data are categorical — scores represent named categories with no numerical ordering or quantitative meaning. The numbers (if used at all) serve only as labels. Examples: biological sex (1 = male, 2 = female); diagnosis category (1 = depression, 2 = anxiety, 3 = OCD); eye colour; type of attachment (secure, avoidant, ambivalent).

With nominal data: categories can be counted; the mode is the only valid measure of central tendency; the mean and median are meaningless; chi-squared is the appropriate inferential test for frequency comparisons.

Ordinal Level

Ordinal data are ranked — scores have a meaningful order (higher numbers indicate more of the measured attribute) but the intervals between consecutive ranks are not guaranteed to be equal. Examples: finishing position in a race (1st, 2nd, 3rd); ratings on a Likert scale (Strongly Disagree to Strongly Agree); a judge's ranking of essay quality.

With ordinal data: the order of scores is meaningful; the median and mode are valid measures of central tendency; the mean is technically inappropriate (because equal intervals cannot be assumed); non-parametric statistical tests (Mann-Whitney U, Wilcoxon, Spearman's rho) are appropriate.

Interval Level

Interval data have equal intervals between consecutive values, so differences between scores are directly comparable — but there is no true zero point (zero does not represent the complete absence of the attribute). Examples: temperature in Celsius or Fahrenheit; IQ scores; calendar dates.

With interval data: the mean, median, and mode are all valid; differences between scores are directly comparable; parametric statistical tests are appropriate. Ratios are not meaningful — 40°C is not 'twice as hot' as 20°C in any physical sense.

Ratio Level

Ratio data have equal intervals and a true zero point — zero represents the complete absence of the attribute, making ratios between scores meaningful. Examples: reaction time in milliseconds; number of words recalled; height in centimetres; heart rate in beats per minute.

Ratio is the highest level of measurement. All arithmetic operations are valid. 'Twice as fast' and 'half as many' are meaningful statements. In practice, interval and ratio data are often treated equivalently in psychology, and parametric statistical procedures apply to both.