Levels of measurement 2.0
About this app
Rebuild 2.0 (preview): level of measurement as the set of admissible transformations — five statements hold or flip live, ladder of permitted statistics, sorting exercise with 19 variables.
A level as a set of transformations
Following S. S. Stevens (1946), a level of measurement is defined by the transformations that leave the information in a measurement unchanged. Nominal: any one-to-one relabelling. Ordinal: any strictly increasing function. Interval: y = a·x + b with a > 0. Ratio: y = a·x with a > 0. Absolute: only y = x. A statement is meaningful at a level if it survives every admissible transformation. If it fails under just one, it depended on an arbitrary choice such as the coding or the zero point.
What the app shows
Five constructed data sets are available: payment method, customer satisfaction, warehouse temperature, revenue per branch and team size, each with two groups A and B. Original values sit at the top, transformed ones below, joined by threads. Five statements about mode, median, mean, difference and ratio are checked live and marked as holding or breaking. With curved transformations a hollow triangle shows that the mean of the images differs from the image of the mean.
Things to try
Pick temperature and convert from °C to °F or kelvin: the claim “twice as warm” breaks. At ordinal level, drag the points freely or use a preset that spreads the top; the median survives, while the comparison of means can flip. If you pick a higher level than the variable supports, the app warns you. The ladder lists the statistics for each level, from the χ² test to the coefficient of variation. The quiz panel lets you practise with 19 variables.
Frequently asked questions
- What is an admissible transformation in measurement theory?
- An admissible transformation is a conversion of measured values that does not change their information, for example from euros to thousands of euros or from °C to °F. Which transformations are admissible depends on the level of measurement. At interval level they are linear functions y = a·x + b with a > 0.
- Why is 20 °C not twice as warm as 10 °C?
- The zero of the Celsius scale is set arbitrarily, so the ratio 20 : 10 is not a property of temperature. Converting to °F gives a different ratio. In kelvin, which has a true zero, 20 °C is only about 3.5 % more than 10 °C.
- Which statistics are allowed at each level of measurement?
- Nominal: mode, frequencies, cross tables, χ² test. Ordinal adds median, quantiles and rank correlations such as Spearman or Kendall. Interval adds mean, standard deviation, Pearson correlation and the t-test; ratio adds quotients, percentage changes, the coefficient of variation and the geometric mean.
- How does this differ from the first levels of measurement app?
- This version explains the levels through admissible transformations: you convert data and see which statements survive. It includes five data sets with presets such as °C → °F or k€ → €, a ladder of allowed statistics and a quiz with 19 variables.
Subject: Data & decisions
More from Data & decisions
- Anchoring effect
- Benford's law
- Berkson's paradox
- Cherry-picking: base period
- Cherry-picking: base period 2.0
- The three-cowboy duel (truel)
- The three-cowboy duel 2.0
- The three means
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps