Nominal scale
Categories with no order. Numbers, if you use them, are name tags: coding Chrome as 1 and Safari as 2 ranks nothing.
- Examples
- eye color, blood type, marital status, race, yes or no
- You can
- count, compare shares, find the mode
Levels of measurement: nominal, ordinal, interval and ratio
Every variable sits at one of four levels of measurement. Nominal data names categories, ordinal data ranks them, interval data adds equal steps, and ratio data adds a true zero. The level decides which statistics make sense.
By Michael Hodge, BSc Psychology Updated September 2026
The four levels of measurement, with examples
S. S. Stevens named the four levels in 1946, and statistics courses still teach them in his order.1 Each level has every property of the one before it, plus one more.
Ordinal scale
Categories in a meaningful order, with gaps of unknown size. You know first beat second, not by how much.
- Examples
- education level, income bracket, star ratings, class rank, clothing size
- You can
- all of that, plus the median
Interval scale
Numbers with equal steps but no true zero. Differences mean something. Ratios do not, so 20°C is not twice as warm as 10°C.
- Examples
- temperature in Celsius or Fahrenheit, calendar year, IQ scores, SAT scores
- You can
- add, subtract, take the mean
Ratio scale
Equal steps and a true zero, where zero means none of the thing. Ratios finally work: 80 kg really is twice 40 kg.
- Examples
- height, weight, age in years, income, time taken, number of children
- You can
- every statistic, ratios included
Psychology courses often teach the same ladder as four properties: identity (the values differ), magnitude (they have an order), equal intervals, and an absolute zero. Nominal data has identity only. Ratio data has all four.
Nominal vs ordinal vs interval vs ratio: how to tell them apart
Three yes or no questions sort any variable. Ask them in order and stop at the first no.
- 1Do the answers have a meaningful order?No: nominal. Yes: go to question 2.
- 2Are the gaps between values equal?No: ordinal. Yes: go to question 3.
- 3Does zero mean none of the thing?No: interval. Yes: ratio.
Nominal vs ordinal: is there an order?
Shuffle the categories. If nothing is lost, the variable is nominal: blood types A, B, AB and O read just as well in any order. If shuffling breaks something, it is ordinal. Small, medium and large only make sense one way round.
Marital status is the classic nominal example. Education level is the classic ordinal one.
Ordinal vs interval: are the steps equal?
A 1 to 5 satisfaction rating looks like evenly spaced numbers. Nobody has shown that satisfied to very satisfied is the same distance as neutral to satisfied, so the rating is ordinal.
On an interval scale the steps are equal by construction. One degree is one degree anywhere on a thermometer.
Interval vs ratio: is there a true zero?
A true zero means none of the thing. Zero kilograms is no weight, so 80 kg is twice 40 kg. On an interval scale, zero is a point someone chose: 0°C is where water freezes, not the absence of heat.
Convert 10°C and 20°C to Fahrenheit and you get 50°F and 68°F. The "twice as warm" claim disappears, while the 10 degree difference survives. Kelvin starts at absolute zero, so it is a ratio scale.
Is it nominal, ordinal, interval or ratio? Look up a variable
The 52 variables people ask about most, each with its level and the reason. Type a variable to filter the list.
| Variable | Level | Why |
|---|---|---|
| Age in years | Ratio | Zero is a real starting point, and 40 is twice 20. |
| Age group, such as 18 to 24 | Ordinal | The groups are ranked, but their widths differ. |
| Gender or sex | Nominal | Categories with no built-in order. |
| Weight in kilograms or pounds | Ratio | Zero means no weight, so 80 kg is twice 40 kg. |
| Income in dollars | Ratio | Zero income is a true zero, and $80,000 is twice $40,000. |
| Income bracket | Ordinal | Ordered bands, usually of different widths. |
| Race or ethnicity | Nominal | Categories with no order. |
| Height | Ratio | Equal units and a true zero. |
| Temperature in Celsius or Fahrenheit | Interval | Equal degrees, but zero is not the absence of heat. |
| Temperature in kelvin | Ratio | Zero kelvin is absolute zero, so ratios hold. |
| Time taken, in seconds or minutes | Ratio | Zero means no time passed, and 10 minutes is twice 5. |
| Time of day | Interval | Equal minutes, but midnight is not an absence of time. |
| Body mass index (BMI) | Ratio | Calculated from weight and height, both ratio variables. |
| BMI category, underweight to obese | Ordinal | Ranked groups of unequal width. |
| Marital status | Nominal | Single, married, divorced and widowed have no order. |
| Blood type | Nominal | A, B, AB and O are labels with no order. |
| Blood pressure (mmHg) | Ratio | Equal units measured up from a true zero. |
| Yes or no answer | Nominal | Two categories with no order. |
| Education level, high school to doctorate | Ordinal | Ranked, but the steps between levels are not equal. |
| Years of schooling | Ratio | A count of years with a true zero. |
| Single Likert item, strongly disagree to strongly agree | Ordinal | Ranked answers with gaps of unknown size. |
| Likert scale score, several items added together | Interval | A multi-item score is conventionally analyzed as interval. |
| Satisfaction, very dissatisfied to very satisfied | Ordinal | Ranked categories. |
| Star rating, 1 to 5 | Ordinal | Ranked, and nobody has shown 4 to 5 stars equals 1 to 2. |
| Likelihood to recommend, 0 to 10 | Ordinal | One rating item. More points do not make equal steps. |
| Pain score, 0 to 10 | Ordinal | A self-rating, not equal units of pain. |
| IQ score | Interval | Treated as equal units, but an IQ of 0 does not mean no intelligence. |
| SAT or other standardized test score | Interval | Scaled scores with no true zero. |
| Credit score | Interval | Scores run from 300 to 850. There is no zero. |
| Calendar year | Interval | 2024 minus 2014 is 10 years, but year 0 is a convention. |
| Eye color | Nominal | Labels with no order. |
| Favorite color or brand | Nominal | Preferences with no order. |
| Country or state of residence | Nominal | Places, not amounts. |
| Religion | Nominal | Categories with no order. |
| Political party | Nominal | Categories with no order. |
| Zip code | Nominal | A number used as a label. Its average means nothing. |
| Phone number | Nominal | Digits used as a name tag. |
| Jersey number | Nominal | Number 10 is not more of anything than number 5. |
| Class rank or finishing position | Ordinal | First beat second, but by how much is not recorded. |
| Letter grade, A to F | Ordinal | Ranked grades with uneven gaps. |
| Clothing size, S, M, L | Ordinal | Ranked labels. |
| Military or job rank | Ordinal | Ranked, with unknown gaps between ranks. |
| Socioeconomic status, low to high | Ordinal | Ranked groups. |
| Cancer stage, I to IV | Ordinal | Ranked stages. Stage IV is not twice stage II. |
| Number of children | Ratio | A count. Zero means none. |
| Hours of sleep | Ratio | Zero hours is none, and 8 hours is twice 4. |
| Heart rate, beats per minute | Ratio | A count per minute with a true zero. |
| Distance traveled | Ratio | Zero means no distance. |
| Price paid | Ratio | Zero is free, and $20 is twice $10. |
| Number of correct answers on a test | Ratio | Zero correct is a true zero. |
| Reaction time in milliseconds | Ratio | Equal units and a true zero. |
| Sales revenue | Ratio | Zero means no sales. |
No variable in the list matches. Run it through the three questions above instead.
Showing 18 of 52 variables
What you can calculate at each level
Each level can use every statistic allowed at the levels below it. So interval data supports more calculations than nominal data, never fewer.
| Level | Average | Spread | Typical tests | Chart |
|---|---|---|---|---|
| Nominal | Mode | Counts and percentages | Chi-square, Fisher's exact test | Bar chart sorted by size |
| Ordinal | Median, mode | Range, interquartile range | Mann-Whitney U, Kruskal-Wallis, Spearman correlation | Bar chart in the scale's own order |
| Interval | Mean, median, mode | Standard deviation, variance | t-test, ANOVA, Pearson correlation, regression | Histogram, box plot |
| Ratio | All of the above, plus the geometric mean | All of the above, plus the coefficient of variation | The same tests as interval | Histogram, box plot |
The tests follow UCLA's guide to choosing a statistical test.6 The chart column is the one people miss: nominal bars can be sorted by size, but ordinal bars must stay in the scale's own order, because the order is the information.
Is a Likert scale ordinal or interval?
A single Likert item is ordinal. A Likert scale, several items about the same idea added into one score, is usually analyzed as interval.
One Likert item
OrdinalThe support team solved my problem.
- Strongly disagree
- Disagree
- Neutral
- Agree
- Strongly agree
Report the median and the share choosing each answer.
A four-item Likert scale
Interval- Support replied quickly.1 to 5
- Support understood my problem.1 to 5
- Support solved my problem.1 to 5
- I would contact support again.1 to 5
Averaged into one support score, which you can summarize with a mean and standard deviation.
The reason is spacing. Nobody has shown that the step from agree to strongly agree equals the step from neutral to agree. As Susan Jamieson put it in Medical Education, the average of "fair" and "good" is not "fair-and-a-half."3
As Boone and Boone describe it, Rensis Likert analyzed the combined score, not single answers. Their guidance treats four or more items combined into one score as interval data, open to means, t-tests and regression.4 Some researchers go further: Sullivan and Artino report that parametric tests hold up well on Likert-type data.5
The habit that keeps you right either way: show the distribution first, then the median, and save the mean for multi-item scales. A 3.0 average can hide a room split between 1s and 5s.
Levels of measurement in survey questions
The level is set when you write the question, not when you analyze the answers. The same topic can be asked at any of the four levels.
| Level | The question | Answers | What you can report |
|---|---|---|---|
| Nominal | How did you get to work today? | Car, bus, train, bike, walked | The most common way, and each one's share |
| Ordinal | How long is your commute? | Under 15 min, 15 to 30, 31 to 60, over 60 | The median band, and the share in each band |
| Interval | What year did you start this commute? | 2019, 2022, 2025 and so on | Gaps between years, and the average year |
| Ratio | How many minutes is your commute, one way? | Any number, 0 upward | Any statistic: 60 minutes is twice 30 |
Three rules for your next survey
A multiple choice poll records nominal or ordinal answers. For an exact number such as age in years, a number field in the form builder records ratio data. For wording by topic, the survey question examples are grouped the same way you would plan a survey.
- 1Ask for the number when you can. Age in years can be grouped later. An answer of 25 to 34 can never become 29 again.
- 2Prefer counts to frequency words. "How many times did you exercise last week?" is ratio. "How often do you exercise?" is ordinal, and "often" means different things to different people.
- 3Decide up front whether a rating stands alone. One rating is ordinal. Four ratings of the same idea, averaged, make a scale you can analyze with a mean.
Levels of measurement quiz
Ten variables, one at a time. Pick a level and you see the reason straight away, right or wrong.
Stuck on one? The three questions sort any variable in under a minute.
Variable 1 of 10Score 0
Which level of measurement is this?
Age in years
Levels of measurement: common questions
Why do levels of measurement matter?
Because the level decides which numbers mean anything. Average a column of zip codes and you get a precise figure that describes nothing. Classify each variable first and the rest follows: the summary statistic, the chart and the statistical test. In quantitative research, a test picked for the wrong level answers a different question from the one you meant to ask.
Are nominal and ordinal data qualitative?
Yes. Nominal and ordinal data are categorical, often called qualitative. Interval and ratio data are quantitative. Discrete versus continuous is a separate split: number of children is ratio and discrete, because it moves in whole steps, while height is ratio and continuous, because any value in a range is possible.
Who created the four levels of measurement?
The Harvard psychologist S. S. Stevens, in a 1946 paper in Science called "On the Theory of Scales of Measurement."1 He matched each level to the statistics he considered permissible for it. Some statisticians think that matching is taught too rigidly: Paul Velleman and Leland Wilkinson argued in 1993 that the levels should guide an analysis rather than forbid one.2
What does SPSS call the levels of measurement?
SPSS uses three measurement levels: Nominal, Ordinal and Scale. Scale covers interval and ratio data together, and IBM's own examples are age in years and income in thousands of dollars.7 You set the level in the Measure column of Variable View.
Does a 0 to 10 scale make data interval?
No. More points give finer resolution, not equal spacing. Nothing shows that the step from 8 to 9 on a recommend-us question equals the step from 5 to 6, so a single 0 to 10 item is still ordinal. Many analysts treat long rating scales as roughly interval anyway. If you do, say so and show the distribution beside the mean.
What level is a "select all that apply" question?
It is not one variable. Each option becomes its own nominal yes or no variable, so a question with six options gives you six variables. That is also why the percentages add up to more than 100.
Is a percentage a ratio variable?
Usually. A percentage of a count, such as the share of people who voted yes, has a true zero, and 40 percent is twice 20 percent. It is also capped at 100, a limit the four levels have no name for.
Ask it at the level you need
Write the question, choose the answers, share one link. Poll Maker counts the responses and charts them as they arrive.
Keep reading
How accurate are polls?What a normal polling error looks like, and how to read a poll in two minutes.Nonresponse biasWhen the people who skip your poll change the answer, and how to spot it.
About this guide
Written by Michael Hodge, BSc Psychology (University of Wollongong). First published August 2024 and rewritten in September 2026. Each classification in the lookup is the textbook answer; where careful sources disagree, as they do on Likert data, both positions are given.
Sources
- 1.Stevens, S. S. (1946). On the Theory of Scales of Measurement. Science, 103(2684), 677-680. www.jstor.org/stable/1671815
- 2.Velleman, P. F., and Wilkinson, L. (1993). Nominal, Ordinal, Interval, and Ratio Typologies Are Misleading. The American Statistician, 47(1), 65-72. www.jstor.org/stable/2684788
- 3.Jamieson, S. (2004). Likert scales: how to (ab)use them. Medical Education, 38(12), 1217-1218. eprints.gla.ac.uk/59552/1/59552.pdf
- 4.Boone, H. N., and Boone, D. A. (2012). Analyzing Likert Data. Journal of Extension, 50(2), Article 48. commons.joe.org/joe/vol50/iss2/48/
- 5.Sullivan, G. M., and Artino, A. R. (2013). Analyzing and Interpreting Data From Likert-Type Scales. Journal of Graduate Medical Education, 5(4), 541-542. pmc.ncbi.nlm.nih.gov/articles/PMC3886444/
- 6.UCLA Office of Advanced Research Computing. Choosing the Correct Statistical Test in SAS, Stata, SPSS and R. stats.oarc.ucla.edu/other/mult-pkg/whatstat/
- 7.IBM. Variable measurement level. IBM SPSS Statistics 30 documentation. www.ibm.com/docs/en/spss-statistics/30.0.0?topic=view-variable-measurement-level
