01 · The Question
Do You Still Need to Know Nominal, Ordinal, Interval, and Ratio?
If you have taken a research methods or introductory statistics course, you have probably encountered four familiar labels: nominal, ordinal, interval, and ratio. You may also have been told that identifying the correct level determines which statistics you are allowed to use.
The framework is still widely taught because it captures genuinely important distinctions. A category code is not the same kind of information as a ranking, and a ranking is not automatically a measurement with equal numerical intervals.
Where things become less straightforward is the leap from those distinctions to rigid rules about statistical analysis. The four levels remain useful, but modern statistical practice is more nuanced than simply locating a variable in one box and reading the permitted test from a chart.
03 · What You Need to Know
What the Four Levels Actually Tell You About Your Data
The Framework Comes From Stevens' Theory of Measurement
The familiar classification is associated with psychologist S. S. Stevens, who presented the nominal, ordinal, interval, and ratio typology in his 1946 paper On the Theory of Scales of Measurement. The basic idea is that different measurement scales preserve different kinds of relationships among their values.
The framework is often presented hierarchically. Nominal measurement distinguishes values. Ordinal measurement adds meaningful ordering. Interval measurement adds meaningful equality of differences. Ratio measurement adds a meaningful zero that permits ratio statements.
That hierarchy is a useful starting point because it asks a more fundamental question than “Are these numbers?” It asks: What do these values actually represent?
| Level |
What the values tell you |
Simple example |
What you should not assume |
| Nominal |
Whether observations belong to the same or different categories |
Academic department |
That categories have a meaningful numerical order |
| Ordinal |
Category or value order |
Class rank |
That adjacent positions are equally far apart |
| Interval |
Order and meaningful equal differences |
Temperature in degrees Celsius |
That zero represents absence or that ratios are meaningful |
| Ratio |
Order, equal differences, and a meaningful zero |
Elapsed time |
That the variable is automatically suitable for every statistical procedure |
Nominal Measurement Is About Categories, Not Magnitude
A nominal variable classifies observations into categories without imposing a meaningful order. Examples might include academic discipline, type of institution, experimental condition, or country of residence.
Researchers often assign numbers to these categories for data entry. Department A might be coded 1, Department B as 2, and Department C as 3. Those numbers are labels. They do not imply that Department C has “more department” than Department A or that the distance between codes 1 and 2 has substantive meaning.
This is an important reminder that numeric coding does not automatically create quantitative measurement.
Ordinal Measurement Adds Order but Not Known Equal Distance
Ordinal values can be meaningfully ranked. If participants are classified as low, medium, and high on some ordered characteristic, “high” is above “medium,” which is above “low.”
What ordinal measurement does not establish is that the distances between those positions are equal. The difference between first and second place, for example, need not be the same as the difference between second and third.
This distinction becomes especially relevant for ordered response categories. A participant choosing “strongly agree” expresses a higher position on the response continuum than someone choosing “agree,” but the labels themselves do not establish that the psychological distance between adjacent categories is constant.
Interval Measurement Makes Differences Meaningful
An interval scale has ordered values with equal units, making differences interpretable. The classic example is temperature measured in degrees Celsius.
The difference between 10°C and 20°C is the same temperature difference as between 20°C and 30°C. However, zero degrees Celsius does not mean an absence of temperature. Consequently, a temperature of 20°C is not meaningfully “twice as hot” as 10°C simply because 20 is twice 10.
Interval measurement therefore supports statements about differences without necessarily supporting meaningful ratios.
Ratio Measurement Adds a Meaningful Zero
Ratio measurement has the properties of interval measurement plus a meaningful zero corresponding to an absence of the measured quantity in the relevant sense. This makes ratio comparisons meaningful.
For elapsed time, for example, zero seconds can represent no elapsed time. Twenty seconds is twice ten seconds in duration. Similar reasoning applies to many measurements of length, mass, counts, and other quantities, although the exact interpretation always depends on how the variable has been defined.
The presence of the numeral zero is not sufficient. The question is whether zero has the required substantive meaning for the scale.
Levels Describe the Meaning of Values, Not How Sophisticated They Look
A variable does not become interval or ratio merely because it contains many decimal places. Nor is a variable necessarily nominal because categories are represented with words.
Suppose a researcher codes satisfaction as 1 = dissatisfied, 2 = neutral, and 3 = satisfied. The values are ordered, but assigning the numbers 1, 2, and 3 does not by itself establish equal psychological distance between the categories.
Conversely, a ratio variable remains ratio-level information even if researchers later group it into categories. Categorizing age into 18–24, 25–34, and 35–44, for example, changes the information available in the analyzed variable. The original ages contain more quantitative information than the grouped categories.
What About Likert Items?
This is where the neat classroom framework often meets a less neat methodological debate.
An individual Likert-type item with ordered response categories such as “strongly disagree” through “strongly agree” is ordinarily treated as ordinal because the categories are ordered but equal spacing between adjacent responses is not established merely by assigning consecutive numbers.
A multi-item score formed from several such items raises a different question. Researchers commonly analyze summed or averaged scale scores using methods that treat the resulting score as approximately continuous, particularly when the measurement model and empirical properties support that practice. Methodological literature has long debated when this is defensible.
Watch Out
Do not treat “Likert item” and “Likert scale score” as automatically equivalent measurement problems. An individual ordered response and a composite score created from multiple items may require different reasoning about measurement and analysis.
This is also why the choice between single-item and multi-item measurement can matter beyond questionnaire length.
Does the Level of Measurement Determine the Statistical Test?
Not by itself.
Stevens linked scale types to permissible transformations and statistical operations, and introductory decision charts often translate this into rules such as “ordinal data require nonparametric statistics.” Those rules can be useful safeguards for beginners, but they can become too rigid.
The suitability of an analysis also depends on the research question, study design, sampling structure, distribution of the observations or residuals where relevant, measurement model, sample size, and assumptions of the statistical method. Modern models can also analyze categorical and ordinal outcomes directly.
Thus, identifying the level of measurement is a starting point for analysis, not a substitute for understanding the statistical model.
The Framework Has Been Influential, but It Is Not Beyond Criticism
Stevens' classification has had enormous pedagogical influence, but the relationship between measurement scales and permissible statistical procedures has been debated for decades. Researchers have questioned whether the traditional categories are sufficient for all measurement situations and whether statistical procedures should be prohibited solely on the basis of scale classification.
That criticism does not make the four levels useless. They remain an accessible way to ask whether values represent categories, order, equal differences, or meaningful ratios. Problems arise when the framework is treated as the whole theory of measurement or as an automatic statistical algorithm.
Your Operationalization Can Change the Level of Information You Have
The level of measurement is not always an intrinsic property of the broad construct. It often depends on how you operationalize the construct into a measurable variable.
Age can be recorded in years, grouped into ordered age bands, or dichotomized according to a threshold. Those choices produce variables with different informational properties even though they all concern age.
This is another reason to preserve information when there is no substantive reason to discard it. Converting a rich quantitative variable into coarse categories may simplify presentation, but it can also remove information that cannot later be recovered.
