Stevens' Four Levels of Measurement

Why it exists

Variables carry different kinds of information. A number can be a label, a rank, or a true quantity. Without a way to classify these differences, researchers apply the wrong math to the wrong data. Stevens created his framework to prevent category errors by matching statistical operations to the actual nature of the values.

The mental model

Think of the levels as permissions on a file system. Nominal data gives you the least permission; you can only check equality. Ordinal adds ordering. Interval adds meaningful differences. Ratio adds a true zero and meaningful ratios. Each level inherits the previous permissions and adds new ones. You cannot run operations that require permissions your data does not possess.

How it works

Stevens defined four levels that describe the nature of information within values assigned to variables. Nominal scales use values as names or categories. Ordinal scales use values to express rank or sequence. Interval scales use values where differences are meaningful but zero is arbitrary. Ratio scales use values where differences and ratios are both meaningful and zero indicates absence. This hierarchy means that as you move up the scale, the permissible mathematical and statistical operations increase.

When to use it

Use this framework when designing research instruments and choosing analytical methods in UX research. It guides whether you should report modes, medians, or means. It also helps you defend your methodology when stakeholders ask why you cannot average a set of user personas or why a Net Promoter Score calculation is mathematically controversial.

When not to use it

Do not treat the levels as rigid physical laws. The framework originated in psychology and has been criticized, extended, and rejected across disciplines. Some researchers argue that many real-world variables, especially aggregated survey items, do not fit cleanly into one level. Do not let the taxonomy override domain expertise or context-specific validity.

One canonical example

Imagine a UX survey asking three questions. First, which device did you use, phone, tablet, or desktop. This is nominal; you can count frequencies but not order the devices. Second, rate ease of use from one to five. This is ordinal; you know five is better than four but you cannot assume the gap between four and five equals the gap between one and two. Third, how many minutes did you spend on the task. This is ratio; you can say one user spent twice as long as another. Running an ANOVA on the device category or averaging the ease ratings without caution violates the level of measurement and corrupts the insight.