There are four levels of measurement in research and statistics — nominal, ordinal, interval and ratio. They describe how much mathematical information a variable carries, from simple unordered categories (nominal) up to fully numeric measurements with a true zero (ratio). The level you are working with determines which descriptive statistics, charts and statistical tests are valid for your data.
Not all data is created equal. A postcode, a satisfaction rating, a temperature and a body weight all store very different kinds of information, and if you treat them the same way statistically you will get nonsense results. That is exactly what the levels of measurement help you avoid.
Understanding the four levels is one of the most practically useful things you can learn in a research methods course. It directly determines which type of variable you have, which statistics you can use, which charts make sense, and how carefully you need to word your conclusions.
The four levels of measurement.
Where The Framework Comes From
American psychologist Stanley Smith Stevens introduced this four-level classification in a 1946 paper in the journal Science. His core argument was that the mathematical operations permissible on any set of numbers depend on the measurement scale used. That single idea has shaped how statistics has been taught ever since, and it remains the standard framework in research methods textbooks today.
“…we may say that measurement, in the broadest sense, is defined as the assignment of numerals to objects or events according to rules.” — S. S. Stevens, ‘On the Theory of Scales of Measurement’, Science, vol. 103 (1946)
Stevens labelled his four scales nominal, ordinal, interval and ratio. The key insight is that they form a hierarchy: each level keeps every property of the level below it and adds one new property of its own. Knowing where your variable sits on this ladder tells you, at a glance, what you are allowed to do with it.
The framework is sometimes remembered by the mnemonic NOIR (Nominal, Ordinal, Interval, Ratio), which lists the four scales in order of increasing information. The first two are often grouped as categorical (or qualitative) data and the last two as numeric (or quantitative) data — a distinction worth keeping in mind, because most statistics software asks you to declare it before you can run an analysis.
Not sure which test fits your data?
ResearchProspect to the rescue!
Our statisticians match every variable to the right measurement level and analysis, so your results stand up to examiner scrutiny — explore our statistical analysis service.
The Four Levels At A Glance
The table below is the quickest way to tell the four levels apart. Ask four questions of any variable — are the values ordered? are the intervals equal? is there a true (meaningful) zero? and can you compute valid averages? — and the answers place it on exactly one level.
| Level | Ordered? | Equal intervals? | True zero? | Valid averages | Examples | Typical tests |
|---|---|---|---|---|---|---|
| Nominal | No | No | No | Mode only | Blood type, nationality, party voted for | Chi-square, mode, frequencies |
| Ordinal | Yes | No | No | Median, mode | Degree class, pain scale, single Likert item | Mann-Whitney U, Spearman’s rho, median |
| Interval | Yes | Yes | No | Mean, median, mode | Temperature in °C, IQ score, calendar year | t-test, ANOVA, Pearson’s r |
| Ratio | Yes | Yes | Yes | Mean (incl. geometric), median, mode | Height, weight, income, reaction time, age | t-test, ANOVA, regression, all of the above |
Each level builds on the one above it. Ratio has all the properties of interval, interval has all the properties of ordinal, and ordinal has all the properties of nominal. Move up the hierarchy and you gain information and statistical options; move down and you lose them.
The Four Levels Explained
1. Nominal level
Nominal data classifies observations into named categories that have no inherent order. You can tell whether two values are the same or different, but nothing more. “Nominal” comes from the Latin nomen, meaning name — the values are simply labels.
- Examples: nationality, blood type, eye colour, marital status, the political party someone voted for.
- What you can do: count how many fall in each category, report frequencies and percentages, and identify the mode (the most common category).
- What you cannot do: rank the categories, calculate a mean or median, or measure “distance” between them. If categories are coded 1, 2, 3, those numbers are arbitrary labels, not quantities.
2. Ordinal level
Ordinal data can be rank-ordered, but the gaps between ranks are not necessarily equal. You know which value is higher or lower, but not by how much.
- Examples: a degree classification (First, 2:1, 2:2, Third); a five-point Likert response from “strongly disagree” to “strongly agree”; finishing position in a race; a clinical pain scale.
- What you can do: order the values, report the median and mode, and use rank-based tests such as Mann-Whitney U or Spearman’s rho.
- What you cannot do: assume the distance from “agree” to “strongly agree” equals the distance from “neutral” to “agree”. Because intervals are not guaranteed equal, the mean is technically not justified for a single ordinal item.
3. Interval level
Interval data is ordered and has equal, meaningful intervals between values, but its zero point is arbitrary rather than a true “none”. This is the level students most often get wrong.
- Examples: temperature in Celsius or Fahrenheit, IQ scores, calendar years (e.g. AD).
- What you can do: add and subtract values, calculate the mean, standard deviation, and use t-tests, ANOVA and Pearson correlation.
- What you cannot do: form ratios. 20°C is not “twice as hot” as 10°C, because 0°C does not mean “no temperature” — it is just the freezing point of water. The zero is a convention, not an absence.
4. Ratio level
Ratio data has every property of interval data plus a true zero that genuinely means “none of the quantity”. This makes it the richest level of measurement.
- Examples: height, weight, age, income, distance, reaction time, number of children.
- What you can do: everything interval allows, plus form meaningful ratios. Because 0 kg means no mass, 80 kg really is twice 40 kg. The full toolkit — including the geometric mean and coefficient of variation — is available.
Key fact: the single test that separates interval from ratio is the true zero. If ‘zero’ means ‘none of the thing’ and you can say one value is ‘twice’ another, you have ratio data; if not, it is interval. — ResearchProspect statistics team
What Each Level Allows You To Do
The hierarchy is not academic hair-splitting — it decides which statistical test and which measure of central tendency are valid. This grid summarises the permissible operations.
| Operation | Nominal | Ordinal | Interval | Ratio |
|---|---|---|---|---|
| Count frequencies | ✓ | ✓ | ✓ | ✓ |
| Identify the mode | ✓ | ✓ | ✓ | ✓ |
| Rank or order values | ✗ | ✓ | ✓ | ✓ |
| Report the median | ✗ | ✓ | ✓ | ✓ |
| Measure distance between values | ✗ | ✗ | ✓ | ✓ |
| Calculate the mean & SD | ✗ | ✗ | ✓ | ✓ |
| Form ratios (“twice as much”) | ✗ | ✗ | ✗ | ✓ |
| Chi-square test | ✓ | ✓ | ✓ | ✓ |
| Mann-Whitney U / Spearman | ✗ | ✓ | ✓ | ✓ |
| t-test, ANOVA, Pearson’s r | ✗ | ✗ | ✓ | ✓ |
Note: in applied research, multi-item Likert scales (a summed set of items) are frequently treated as interval and analysed with means and t-tests, even though a single Likert item is strictly ordinal. This is a widely accepted convention — just state the assumption explicitly in your methodology.
A Worked Example Of Why It Matters
Suppose a survey codes political party preference as: 1 = Labour, 2 = Conservative, 3 = Liberal Democrats, 4 = SNP. Five respondents answer Labour, Lib Dem, Conservative, Labour, SNP — coded 1, 3, 2, 1, 4.
Mean = (1 + 3 + 2 + 1 + 4) ÷ 5 = 11 ÷ 5 = 2.2.
What does 2.2 mean? Nothing at all. Party preference is nominal data; the numbers are arbitrary labels, not quantities. Averaging them implies an order and equal spacing that simply do not exist — it is statistical nonsense dressed up in a decimal point.
Correct approach: report frequencies and percentages instead — Labour 40% (2/5), Conservative 20%, Lib Dem 20%, SNP 20%. The valid measure of central tendency is the mode (Labour). Visualise with a bar chart, and use a chi-square test if you want to examine whether party preference is associated with, say, region.
The lesson generalises: identifying the level of measurement first tells you which average, which chart and which test are legitimate — before you run a single line of analysis. The same logic applies in reverse when you read other people’s research: if a study reports a mean for a clearly nominal variable, that is a red flag that the analysis cannot be trusted.
Common Student Mistakes
- Treating a single ordinal Likert item as interval and reporting its mean without acknowledging the assumption.
- Assuming any numeric variable is automatically interval or ratio — always check whether the numbers are codes or genuine measurements.
- Confusing discrete vs. continuous with the levels of measurement — these are two separate dimensions of a variable.
- Running a t-test on a nominal outcome — a binary outcome needs logistic regression, not a t-test.
- Claiming “20°C is twice as warm as 10°C” — ratios are invalid for interval data because the zero is arbitrary.
