Levels of Measurement: Nominal, Ordinal and Interval

Levels of measurement decide which statistical test you are allowed to run, and AQA devotes 25 to 30 per cent of the entire Psychology assessment to research methods. Get the level wrong and the test that follows is wrong too.

Every set of data a psychologist collects sits somewhere on a ladder. At the bottom, numbers are little more than labels: a 1 for each participant who chose the red door and a 2 for each who chose the blue one. At the top, numbers behave the way they do in maths lessons, where the distance from 10 to 20 is genuinely the same as the distance from 40 to 50. Where your data sits on that ladder is its level of measurement, and it governs almost everything that happens next.

This matters practically. The level of measurement determines which statistical test you should use, which measure of central tendency is honest, and whether calculating a standard deviation is meaningful or merely arithmetic. AQA names it explicitly as one of the two factors affecting the choice of test, alongside experimental design (AQA, 2021). It is the sort of topic that looks like vocabulary and behaves like a foundation.

The scheme itself comes from one paper. In 1946 the Harvard psychologist S. S. Stevens published four pages in Science proposing that all measurement falls into four classes, which he named nominal, ordinal, interval and ratio. He was writing to settle a seven-year argument inside a British Association for the Advancement of Science committee about whether human sensation could be measured at all. Eighty years on, his four categories are still the ones on the specification, minus one.

What Are the Levels of Measurement in Psychology?

The levels of measurement in psychology are nominal, ordinal and interval, three categories that describe how much mathematical information a set of numbers actually carries. Nominal data sorts observations into named groups. Ordinal data puts them in rank order. Interval data places them on a scale with equal, fixed units. AQA’s specification lists exactly these three under section 4.2.3.2 and does not examine a fourth (AQA, 2021).

Stevens (1946) described the scheme as cumulative, and that is the single most useful thing to understand about it. Each level can do everything the level beneath it can do, plus one extra operation. Nominal data lets you determine whether two things are equal or different. Ordinal data adds the ability to say which is greater. Interval data adds the ability to say by how much. Ratio data, the fourth level, adds the ability to say how many times greater.

Because the ladder is cumulative, it also runs downhill freely and uphill never. Reaction times measured in milliseconds are interval data, and you can always throw information away by ranking them fastest to slowest, or by sorting them into “under 300 ms” and “over 300 ms”. You cannot do the reverse. Once participants have simply told you which of three doors they chose, no amount of analysis will recover a distance between red and blue.

Notice the final column. The level of measurement decides which measure of central tendency is defensible, which is why AQA’s mathematical requirements annex asks students to select “which measure of central tendency is most appropriate for a given set of data” (AQA, 2021). That is a levels-of-measurement question wearing a different hat.

Nominal Data: Counting Into Categories

Nominal data is data sorted into named categories that have no order, where the only number you can produce is a frequency count. The word comes from the Latin nomen, meaning name, and that is genuinely all the numbers are doing: naming. If you code “male” as 1 and “female” as 2, the 2 is not larger than the 1 in any sense that survives contact with the data. Swap the codes and nothing about the study changes.

Examples of nominal data in psychology

• Attachment type in the Strange Situation: secure, insecure-avoidant, insecure-resistant. A child is in one category or another.
• Obedience outcome in Milgram’s procedure: obeyed to 450 volts, or did not.
• Type of conformity displayed: compliance, identification or internalisation.
• Diagnosis: whether a participant meets the criteria for a phobia or does not.
• Handedness, eye colour, first language, nationality, and any other descriptive characteristic recorded as a group.

Nominal data is what a tally chart produces, and it is what a great deal of observational and content analysis work produces once the coding is done. Behavioural categories in a structured observation are nominal: each time a child shares a toy, you add one to the “sharing” column, and the column totals are the data.

What you can and cannot do with nominal data

With nominal data you can count cases, report the mode and test whether the distribution across categories differs from chance. Stevens (1946) listed exactly these in his original table: number of cases, mode and contingency correlation. You cannot calculate a mean, a median, a standard deviation or a range, because none of those operations mean anything when the numbers are labels. A mean attachment type of 1.7 is not a finding. It is a category error.

The statistical test that goes with nominal data is Chi-Squared, which compares observed frequencies against the frequencies you would expect if there were no association. When a question hands you a contingency table of counts, the level of measurement is nominal and Chi-Squared is almost always the answer.

Strengths and limitations of nominal data

The strength of nominal data is that it is quick to collect, easy for a second researcher to replicate and almost impossible to misinterpret. Two observers coding the same behaviour into the same categories will usually agree, which makes inter-rater reliability straightforward to establish. Nominal measurement also captures variables that genuinely have no order, and forcing them onto a scale would be a distortion rather than an improvement.

The limitation is that nominal data throws away almost everything. Two children classified as insecure-avoidant may be very different from each other, and the category records none of that. Because the numbers carry so little information, nominal data has less statistical power than ranked or measured data drawn from the same participants, so a real effect is easier to miss. It is the least sensitive level, and if a more informative measure is available, using nominal is a choice that needs justifying in an evaluation answer.

Ordinal Data: Ranking Without Equal Gaps

Ordinal data is data that can be placed in a meaningful rank order, but where the gaps between adjacent positions are not known to be equal. You know that first beat second, and second beat third. You do not know whether first beat second by a hair or by a mile, and you have no reason to assume the two gaps are the same size.

That single missing piece of information, equal intervals, is what separates ordinal from interval and what most exam answers fail to say out loud. “Ordinal data is ranked” earns little on its own. “Ordinal data is ranked, but the intervals between ranks are not equal, so the mean would be misleading” is the version that scores.

Examples of ordinal data in psychology

• Finishing position in any task: first, second, third.
• Rating scales where participants rate anxiety from 1 to 10, or agreement from “strongly disagree” to “strongly agree”.
• Military rank, school year group, social class, and similar ordered categories.
• Well-being and involvement scales used in early years settings, such as the five-point Leuven Scale, where level 4 is higher than level 3 but not by a measurable amount.
• Degree classifications, Likert item responses, and any measure built from subjective judgement rather than a physical unit.

Stevens (1946) was blunt about how common this is, noting that most of the scales psychologists use widely and effectively are ordinal ones. Psychology measures constructs that have no ruler: anxiety, self-esteem, attachment security, job satisfaction. Numbers get attached to them, but the numbers are ranks dressed up as measurements.

What you can and cannot do with ordinal data

With ordinal data you can report the median and percentiles, and you can run non-parametric tests that work on ranks: Wilcoxon for related designs, Mann-Whitney for independent designs, Spearman’s rho for correlations. You should not report the mean, because averaging positions treats unequal gaps as if they were equal. The median sits at the midpoint regardless of how the gaps are spaced, which is exactly why it survives at this level and the mean does not.

Stevens made the point more sharply than most textbooks do. Means and standard deviations calculated on ordinal data, he wrote, are in error to exactly the extent that the successive intervals on the scale are unequal in size. If the gaps happen to be nearly equal, the error is small. If they are wildly uneven, the average is close to meaningless, and nothing in the arithmetic will tell you which situation you are in.

Strengths and limitations of ordinal data

The strength of ordinal data is that it captures variables psychology genuinely cares about but cannot measure physically. There is no unit of anxiety, yet ranking participants by how anxious they report feeling produces usable, analysable data. Ordinal measurement also retains far more information than nominal, so it detects effects that categories would miss, and it keeps analysis available where a physical scale simply does not exist.

The limitation is the unequal intervals themselves, and everything that follows from them. The mean and standard deviation are off limits, so the most powerful parametric tests are unavailable. Rating scales are also vulnerable to subjectivity: one participant’s 7 out of 10 is another’s 5, and neither number is anchored to anything outside that person’s head. Converting raw scores to ranks discards magnitude as well, so two participants who differ enormously can end up adjacent in the rank order.

Interval Data: Equal Units, Arbitrary Zero

Interval data is data measured in fixed, equal units, where the distance between any two adjacent points is the same anywhere on the scale, but where zero is a convention rather than a true absence. The gap between 10 and 20 is identical to the gap between 60 and 70. What you cannot say is that 20 is twice 10, because the zero point was placed by agreement rather than found in nature.

Temperature is the standard illustration, and Stevens (1946) used it himself. Celsius and Fahrenheit both have equal intervals and both have an arbitrary zero, chosen for convenience. Twenty degrees Celsius is not twice as hot as ten, and the proof is that converting both to Fahrenheit destroys the ratio while preserving the intervals. Any scale where you can add a constant without breaking the measurement is an interval scale, not a ratio one.

Examples of interval data in psychology

• Reaction time in milliseconds, the workhorse of cognitive psychology.
• Time taken in seconds to complete a task or solve a problem.
• Number of words correctly recalled from a list, and other counts of behaviour.
• Heart rate, galvanic skin response, cortisol concentration and other physiological measures.
• Standardised test scores such as IQ, which are built and normed so the units behave as if they were equal.

Several items on that list, notably word counts and times in seconds, are strictly ratio data: recalling zero words really is recalling nothing. Because AQA stops at interval, exam answers treat them as interval and nothing is lost. The specification’s three-level scheme simply folds ratio into interval rather than pretending it does not exist.

What you can and cannot do with interval data

Interval data unlocks almost the whole statistical toolkit. The mean, the standard deviation, the standard error, t-tests and Pearson’s r all become available, and these are the parametric tests, the ones with the most statistical power. What remains off limits is any statement of ratio: you cannot say an IQ of 140 is twice the intelligence of an IQ of 70, because IQ has no true zero and nobody has defined what zero intelligence would be.

Strengths and limitations of interval data

The strength of interval data is precision and power. Equal units mean the mean is honest and the standard deviation describes real spread, so parametric tests apply and small effects become detectable. Interval measurement is also objective in a way rating scales are not: a millisecond is a millisecond whoever is holding the stopwatch, which makes replication straightforward and removes the researcher’s judgement from the measurement itself.

The limitation is that genuinely interval measurement is hard to obtain for the things psychology most wants to study. You can time a response precisely, but timing a response is not the same as measuring a thought. Interval data also tends to be reductionist: reducing memory to a word count captures something real while ignoring what was remembered and why, a trade-off explored further in the debate on holism versus reductionism. Physiological measures carry the same risk, standing in for an emotional state they only partly reflect.

Ratio Data and Why AQA Leaves It Out

Ratio data is interval data with a true zero, where zero means a genuine absence of the thing being measured and ratios between values are therefore meaningful. Height, weight, time in seconds and number of items recalled are all ratio: zero centimetres is no height at all, and 40 items really is twice 20. AQA’s specification names only nominal, ordinal and interval, so ratio never appears in a mark scheme (AQA, 2021).

The reason is not an oversight. Stevens (1946) observed that ratio scales of psychological magnitudes are rare, though not unknown. Psychology measures constructs, and constructs almost never have a defensible zero. What would zero anxiety be? Zero extraversion? Zero attachment? Physics has absolute zero and true weightlessness; psychology has scales that were centred somewhere convenient and calibrated against a population.

Statistically, nothing is lost. Every test that works on interval data works on ratio data, so the distinction changes no analytical decision a student will ever make. That is exactly why AQA collapses the two. If you are revising from a general statistics textbook or an American source, you will meet four levels and should simply know that the fourth behaves identically to the third for every purpose the exam has.

It is still worth being able to explain the difference, because “interval data has an arbitrary zero whereas ratio data has a true zero, so only ratio data supports statements like twice as much” is a clean, correct answer if a question ever reaches for it.

Nominal vs Ordinal vs Interval: The Two Boundaries That Matter

Almost every mistake with levels of measurement happens at one of two boundaries: the line between nominal and ordinal, and the line between ordinal and interval. Learning what sits on each side of those two lines is more useful than memorising three definitions.

Nominal vs ordinal: is there an order?

The nominal-ordinal boundary is decided by one question: would reordering the categories lose information? Eye colour has no order, so listing blue before brown is arbitrary and the data is nominal. Degree classification does have an order, so listing a 2:1 before a first would be wrong, and the data is ordinal.

The trap is that both often look like words rather than numbers, which tempts students to call anything non-numerical nominal. “Strongly disagree, disagree, neutral, agree, strongly agree” is made of words and is unmistakably ordinal, because the sequence is part of the meaning.

Ordinal vs interval: are the gaps equal?

The ordinal-interval boundary is decided by whether the units are equal and fixed, and this is where most marks are lost. Both levels produce numbers that can be ranked, so ranking is not the test. The test is whether the gap between 3 and 4 is the same size as the gap between 7 and 8.

A reliable shortcut is to ask where the units came from. If they came from a physical instrument such as a clock, a scale, a thermometer or a counter, the data is interval. If they came from a person making a judgement, the data is ordinal. A stopwatch produces equal seconds by construction; a participant rating their mood out of ten does not produce equal mood units, however evenly the numbers are printed on the page.

What Level of Measurement Is It? Twelve Worked Examples

The quickest way to get secure with levels of measurement is to work through variables that trip people up. Each of the following is a question people genuinely search for, answered with the reasoning rather than just the label.

Age

Age in years is interval data, and strictly ratio, because years are equal units and zero years means no time elapsed since birth. Age recorded as a band, such as 18 to 25 or 26 to 35, is ordinal, because the bands are ordered but need not be equal in width. Age recorded as “child, adolescent, adult” is also ordinal. The variable does not have a fixed level; how you record it does.

Temperature

Temperature in Celsius or Fahrenheit is interval data, because the degrees are equal but zero is a convention rather than an absence of heat. Temperature in Kelvin is ratio, because zero Kelvin is a genuine absolute zero. This is the textbook example of the interval-ratio distinction and the reason 20°C is not twice as warm as 10°C.

A Likert scale response

A single Likert item, such as one statement rated from strongly disagree to strongly agree, is ordinal data. The response options are ordered, but nothing guarantees that moving from “disagree” to “neutral” is the same psychological distance as moving from “agree” to “strongly agree”. For A level purposes, treat a rating scale as ordinal. The research debate about whether summed Likert scales can be treated as interval is covered below.

Gender

Gender is nominal data. The categories are names with no inherent order, and any numerical codes assigned to them are arbitrary labels. The only meaningful statistics are frequency counts and the mode. This holds however many categories are offered, since adding options changes the number of groups but not the absence of order between them.

IQ score

IQ is treated as interval data in practice, though it sits on a genuine fault line. IQ tests are standardised so that scores are normally distributed with a mean of 100, which makes the units behave as though they were equal. But there is no true zero, so an IQ of 140 is not twice an IQ of 70. Stevens (1946) put intelligence among the ordinal scales that try to approximate interval ones, and noted it is not necessary to define what zero intelligence would mean. For exam answers, interval is the expected response.

Reaction time

Reaction time in milliseconds is interval data, and strictly ratio. Milliseconds are equal units produced by a clock, and zero milliseconds would mean no time at all. Reaction time is the cleanest interval variable in psychology and the reason cognitive studies can use parametric tests so routinely.

Height and weight

Height and weight are ratio data, treated as interval at A level. Centimetres and kilograms are equal units, and zero of either is a true absence. Both are the standard illustrations of ratio measurement in general statistics textbooks, which is why they come up so often in searches even though the AQA specification never asks about ratio.

Exam mark and class position

An exam mark out of 100 is interval data; the resulting class position is ordinal data. This pair shows the ladder working downhill in a single step. Two students on 71 and 68 are separated by three marks and by one rank; two students on 68 and 41 are separated by 27 marks and also by one rank. Converting marks to positions discards the size of the gap, which is precisely what ordinal data lacks.

Attachment type

Attachment type is nominal data. Secure, insecure-avoidant and insecure-resistant are named categories with no order, so the data is a set of frequency counts and the mode is the only average available. This is why studies of cultural variations in attachment report percentages in each category and analyse them with Chi-Squared.

Number of words recalled

Number of words recalled is interval data, and strictly ratio. Each word counts the same as every other and recalling zero words is a real outcome, so means and standard deviations are appropriate. Memory studies that report an average recall of 7.3 items are relying on exactly this.

Social class and educational level

Social class and highest qualification are ordinal data. Both are ordered, and neither has equal gaps: the distance between GCSE and A level is not the same quantity of education as the distance between a bachelor’s and a doctorate. They look like categories, which invites the nominal answer, but the order carries meaning and that makes them ordinal.

Whether a participant obeyed

Whether a participant obeyed or did not is nominal data, with just two categories. Any yes-or-no outcome is nominal, including whether a participant conformed, completed the task, or showed a particular behaviour. Binary variables are the simplest nominal data there is, and they are extremely common in social influence research.

The Likert Scale Problem: Ordinal or Interval?

A Likert scale response is ordinal data for A level purposes, but whether it can be treated as interval in real research has been argued over for more than fifty years and the argument is not settled. Knowing the shape of that debate is genuinely useful, because it explains why published psychology papers routinely do something your exam would mark as wrong.

The strict position is that rating scales are ordinal and must be analysed with non-parametric tests. Jamieson (2004) argued this directly in Medical Education, pointing out that the intervals between Likert response options cannot be presumed equal and that means, standard deviations and parametric tests are therefore inappropriate. On this view, reporting an average agreement of 3.7 is a mistake dressed as precision.

The pragmatic position is that it usually does not matter. Carifio and Perla (2008) replied in the same journal, drawing a distinction between a single Likert item, which is clearly ordinal, and a Likert scale built by summing several items, which behaves much more like an interval measure. Norman (2010) went further, reviewing the evidence and concluding that parametric methods are robust to violations of their assumptions, including with ordinal Likert data, and can be used without fear of arriving at the wrong answer.

Knapp (1990) had already framed the middle ground, arguing that the controversy turns on the difference between measurement and statistics: the level of measurement constrains what your numbers mean, not what arithmetic a computer will perform on them. Stevens (1946) conceded something similar, allowing that calculating means on ordinal data is technically improper but attracts a kind of pragmatic sanction because in many instances it produces useful results.

For an exam answer, the safe line is short: a rating scale is ordinal, so use a non-parametric test. If you want the evaluative extra, add that researchers often treat summed rating scales as interval because parametric tests are robust to the violation, and that this remains contested. That single sentence shows awareness of live methodological debate, which is exactly what higher-band AO3 marks reward.

A harder criticism goes after the scheme itself. Velleman and Wilkinson (1993) argued that Stevens’s four categories are actively misleading as a guide to choosing an analysis, because the level of measurement describes how a variable was constructed rather than dictating which arithmetic is permitted afterwards. Michell (1997) pressed from the opposite direction, questioning whether psychology has ever established that its constructs are quantitative at all, which would make the interval label an assumption rather than a finding. Neither argument changes what an exam expects, but both explain why the topic feels tidier on a revision card than it is in practice.

There is a nice irony buried in Stevens’s own table. He listed rank-order correlation under the interval row, not the ordinal one, remarking that the statistic assumes equal intervals between successive ranks and so strictly calls for an interval scale. Spearman’s rho, the test every A level student is taught to use because the data is ordinal, was classified by the man who invented the levels as requiring interval data. The scheme has always been less tidy than the textbooks make it look.

Test Yourself: Sort the Variables

The only reliable way to learn levels of measurement is to classify variables until it becomes automatic. Work through the twelve below, choose a level for each, and read the explanation whether you were right or wrong. The explanations carry the reasoning that earns the marks.

How Levels of Measurement Choose Your Statistical Test

The level of measurement is one of two things that decide your statistical test, the other being the experimental design. AQA states this directly, listing the factors affecting the choice of test as level of measurement and experimental design, and naming the seven tests students must be able to select between: Spearman's rho, Pearson's r, Wilcoxon, Mann-Whitney, related t-test, unrelated t-test and Chi-Squared (AQA, 2021).

Put those two factors on a grid and every test has exactly one home. Down one side, what you are looking for: a difference between related samples, a difference between independent samples, or an association. Across the top, the level of measurement. Nine cells, seven named tests plus the sign test for related nominal data.

Read down the interval column and every entry is a parametric test. Read down the nominal and ordinal columns and every entry is non-parametric. That is the practical consequence of the whole topic: interval data earns you the more powerful tests, and dropping to a lower level costs you power. A fuller walkthrough of the grid, including the parametric assumptions and worked examples for each cell, is in the guide to choosing a statistical test.

Once you have run the test, the level of measurement stops mattering and statistical significance takes over: you compare your calculated value against a critical value and decide whether to reject the null hypothesis. Levels of measurement get you to the right door; significance tells you what is behind it.

Levels of Measurement and Measures of Central Tendency

The level of measurement determines which average is legitimate: the mode for nominal data, the median for ordinal data, and the mean for interval data. This was in Stevens's original 1946 table and it is the reason AQA's mathematical requirements ask students to select the most appropriate measure of central tendency for a given set of data (AQA, 2021).

The logic is about what survives a change of scale. The mode is simply the most common category, so it survives any relabelling. The median is the middle case, so it survives any change that preserves order, however uneven the spacing. The mean is the balance point, so it only survives changes that preserve the size of the gaps, which is to say it needs equal intervals to exist in the first place.

The same argument applies to measures of dispersion. Ordinal data supports percentiles and the interquartile range, because both are positional. Interval data additionally supports the range and the standard deviation, because both depend on real distances. A standard deviation calculated on rank positions is arithmetic performed on numbers that were never measuring a quantity.

The graph column is worth noticing, because it is a quiet source of lost marks. A histogram has continuous bars that touch, which only makes sense when the x-axis is a continuous interval scale. A bar chart has separate bars with gaps, which is right for discrete categories. Drawing a histogram of attachment types is the same mistake as calculating their mean, in a different notation.

How to Remember Nominal, Ordinal and Interval

The simplest way to remember the levels of measurement is to notice that the names describe themselves. Nominal shares a root with name, and nominal data names things. Ordinal shares a root with order, and ordinal data puts things in order. Interval is about the intervals between values, which at this level are finally equal. Three names, three meanings, no mnemonic required.

If you do want an acronym, the common one is NOIR, which gives all four levels in ascending order: Nominal, Ordinal, Interval, Ratio. It has the advantage of being a real word and the disadvantage of including a level AQA never examines, so use it to keep the sequence straight and remember that the exam stops at I.

The version that actually helps in an exam is not a mnemonic at all but a pair of questions, asked in order. First: can the values be put in a meaningful order? If no, it is nominal and you are finished. Second: are the gaps between values equal? If no, it is ordinal. If yes, it is interval. Two questions, applied in that sequence, classify any variable in seconds.

Common Mistakes in Exam Answers

Levels of measurement questions are usually worth one to three marks, and the marks are lost in a small number of predictable ways. Each of the following costs marks on papers every year.

1. Saying "ordinal because it is ranked" and stopping. Interval data can be ranked too. The mark is in the second half of the sentence: the intervals between ranks are not equal.
2. Calling anything made of words nominal. "Strongly agree" is a word and is ordinal. Ask whether reordering the categories would lose meaning, not whether they are numbers.
3. Treating a rating scale as interval because it uses numbers. Printing 1 to 10 on a page does not make the psychological distance between 3 and 4 equal to the distance between 8 and 9.
4. Naming the level but not the consequence. Questions usually want the level in order to get the test or the average. Say both: "ordinal, so Wilcoxon" or "nominal, so the mode".
5. Offering ratio as the answer. It is not on the specification, and where a variable is genuinely ratio, interval is the expected response.
6. Forgetting that the same variable can change level. Reaction time in milliseconds is interval; the same participants ranked fastest to slowest are ordinal; sorted into "fast" and "slow" they are nominal. How the data was recorded decides the level, not what was studied.

That last point is the one that separates a confident answer from a memorised one. The level of measurement is a property of the data, not of the variable, and a researcher chooses it when they design the measure. Choosing a lower level than necessary throws away information and power, which is a legitimate methodological criticism to make in an evaluation, and one that connects to wider questions about how psychology measures what it claims to measure, from cultural bias in standardised instruments to the trade-offs of the case study method.

Conclusion

Levels of measurement look like terminology and behave like infrastructure. Nominal data names, ordinal data ranks, interval data measures in equal units, and each level inherits everything the one below it can do while adding one new ability. That ladder decides which average is honest, which graph is correct and which of the seven named statistical tests you are entitled to run.

Two questions settle almost every case. Can the values be put in a meaningful order? Are the gaps between them equal? Nominal answers no to both, ordinal answers yes then no, interval answers yes to both. Ratio, the fourth level Stevens described in 1946, adds a true zero that psychology rarely has and AQA does not examine.

Get into the habit of naming the level before doing anything else with a set of data, and the rest of the research methods topic becomes considerably more straightforward. It is the first question a psychologist asks about their numbers, and for good reason: everything downstream depends on the answer.

Frequently Asked Questions

What are the four levels of measurement?

The four levels of measurement are nominal, ordinal, interval and ratio, proposed by S. S. Stevens in 1946. Nominal names categories, ordinal ranks them, interval adds equal units, and ratio adds a true zero. A level Psychology examines only the first three, because psychological variables almost never have a defensible zero point.

What does nominal, ordinal and interval mean?

Nominal means data sorted into named groups that have no order, such as eye colour. Ordinal means data placed in rank order where the gaps are not equal, such as finishing position. Interval means data measured in equal, fixed units without a true zero, such as temperature in Celsius. Each term describes how much mathematical information the numbers carry.

Is a Likert scale nominal or ordinal?

A Likert scale is ordinal. Its response options run in a clear order from strong disagreement to strong agreement, which rules out nominal, but there is no guarantee that consecutive options are separated by equal psychological distances, which rules out interval. Researchers sometimes analyse summed Likert scales as interval data, a practice that remains contested.

What level of measurement is age?

Age depends entirely on how it was recorded. Age in years is interval, since years are equal units. Age in bands such as 18 to 25 is ordinal, since the bands are ordered but need not be equal in width. Age as "child, adolescent, adult" is also ordinal. The variable has no fixed level of its own.

How do you remember nominal, ordinal, interval and ratio?

Use the acronym NOIR, which spells a real word and lists the levels in ascending order. Alternatively, lean on the names themselves: nominal relates to name, ordinal to order, interval to equal intervals, and ratio to meaningful ratios. Both are faster in practice than a rhyme, and the second needs nothing memorised.

What is the difference between ordinal and interval data?

Equal gaps are the difference. Both ordinal and interval data can be placed in order, so ordering is not what separates them. Interval data is measured in units of identical size, which permits the mean, the standard deviation and parametric tests. Ordinal data is not, so the median and non-parametric tests are used instead.

What are five examples of nominal data?

Five examples of nominal data in psychology are attachment type, whether a participant obeyed or refused, type of conformity displayed, eye colour, and first language. Each sorts participants into named groups that carry no order, so the data is a set of frequency counts and the mode is the only average that applies.

Why is ratio not included in A level Psychology?

Ratio is excluded because it changes nothing statistically and psychology rarely achieves it. Every test that works on interval data works identically on ratio data, so the distinction never alters a decision a student makes. Stevens himself noted that ratio scales of psychological magnitudes are rare, since constructs such as anxiety have no meaningful zero.

What level of measurement is IQ?

IQ is treated as interval in A level answers. Intelligence tests are standardised against a population so that scores behave as if the units were equal, which permits means and parametric tests. There is no true zero, however, so no ratio statement holds: an IQ of 140 is not twice an IQ of 70.

References

AQA. (2021). AS and A-level Psychology specification (7181, 7182), version 1.2. Assessment and Qualifications Alliance.

Carifio, J., & Perla, R. (2008). Resolving the 50-year debate around using and misusing Likert scales. Medical Education, 42(12), 1150-1152.

Jamieson, S. (2004). Likert scales: How to (ab)use them. Medical Education, 38(12), 1217-1218.

Knapp, T. R. (1990). Treating ordinal scales as interval scales: An attempt to resolve the controversy. Nursing Research, 39(2), 121-123.

Michell, J. (1997). Quantitative science and the definition of measurement in psychology. British Journal of Psychology, 88(3), 355-383.

Norman, G. (2010). Likert scales, levels of measurement and the "laws" of statistics. Advances in Health Sciences Education, 15(5), 625-632.

Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103(2684), 677-680.

Velleman, P. F., & Wilkinson, L. (1993). Nominal, ordinal, interval, and ratio typologies are misleading. The American Statistician, 47(1), 65-72.

Further Reading and Research

Recommended Articles

• Which Statistical Test to Use in Psychology: A Decision Guide - the full grid of design against level of measurement, with an interactive picker.
• Standard Deviation in Psychology: Definition and Calculation - why this measure of dispersion needs interval data to mean anything.
• What Is a Correlation Coefficient? Pearson's r and Psychology Statistics - the interval-level correlation, and how it differs from Spearman's rho.

Suggested Books

• Discovering Statistics Using IBM SPSS Statistics by Andy Field
  • The standard undergraduate statistics text in British psychology, written with unusual humour. Its early chapters on types of data and choosing a test cover the same ground as this article at greater depth.
• Measurement in Psychology: A Critical History of a Methodological Concept by Joel Michell
  • A scholarly account of how psychology came to accept Stevens's definition of measurement, and why the author thinks that acceptance was a mistake. Demanding, but the definitive treatment of the question.
• Research Methods and Statistics in Psychology by Hugh Coolican
  • A comprehensive methods textbook used across A level and undergraduate courses, with clear worked examples of each level of measurement and the tests attached to it.

Recommended Websites

• AQA AS and A-level Psychology specification
  • The primary source. Section 4.2.3.2 lists the three levels of measurement examined, and section 4.2.3.3 names the seven statistical tests students must choose between.
• PubMed record for Norman (2010), at pubmed.ncbi.nlm.nih.gov/20146096
  • The abstract of the most-cited defence of using parametric statistics on Likert data, and the clearest statement of the pragmatic position in the ordinal-interval debate.
• PubMed record for Stevens (1946), at pubmed.ncbi.nlm.nih.gov/17750512
  • The catalogue entry for the original four-page paper in Science that created the whole scheme, useful for citing the source properly rather than second-hand.

Kathy Brodie

Kathy Brodie is an Early Years Professional, Trainer and Author of multiple books on Early Years Education and Child Development. She is the founder of Early Years TV and the Early Years Summit.

Kathy's Author Profile

To cite this article please use:

Early Years TV Levels of Measurement: Nominal, Ordinal and Interval. Available at: https://www.earlyyears.tv/levels-of-measurement-psychology-nominal-ordinal-interval/ (Accessed: 20 September 2026).