Which of the following is NOT a level of measurement?
You’ve probably seen a list of “nominal, ordinal, interval, ratio” and wondered if there’s a trick. In practice, the question often crops up in stats classes, research design workshops, or when a client asks why a certain test isn’t appropriate for their data. Let’s break it down, clear the confusion, and figure out the odd one out.
What Is a Level of Measurement?
Levels of measurement are the categories that describe how we can treat variables statistically. Think of them as the “rules of the game” for data. They tell you what kind of math you can safely perform—averages, percentages, correlations, or you’re stuck with counts or ranks.
There are four classic levels:
| Level | Key Feature | Example |
|---|---|---|
| Nominal | Pure categories, no order | Blood type, gender, city names |
| Ordinal | Ordered categories, but gaps unknown | Likert scale (strongly agree → disagree) |
| Interval | Ordered with equal intervals, no true zero | Temperature in Celsius or Fahrenheit |
| Ratio | Ordered with equal intervals AND a true zero | Weight, height, income, distance |
Each step up adds a new property. That’s why you can calculate a mean on an interval scale but not on an ordinal one—because the intervals aren’t guaranteed to be equal.
Why It Matters / Why People Care
You might ask, “I’m just collecting numbers, why does it matter if I call it interval or ratio?” The answer is simple: the level dictates what statistics are valid.
- Wrong level → wrong analysis. Using a mean on ordinal data can mislead you about central tendency.
- Wrong level → wrong interpretation. A ratio scale’s zero means “none,” so a value of 0 is meaningful. On an interval scale, zero is arbitrary.
- Research credibility. Peer reviewers and funding bodies check that you’re using the correct tests. A mislabel can raise red flags.
In practice, most people get the first two levels right. It’s the interval vs. Which means ratio distinction that trips up a lot of analysts. That’s where the “not a level” question usually hides Worth keeping that in mind. And it works..
How It Works (or How to Do It)
Let’s walk through each level with a practical lens. If you can’t decide which category your data falls into, you’re probably looking at the interval vs. ratio confusion Easy to understand, harder to ignore. Took long enough..
Nominal
- What you can do: Count frequencies, create mode, run chi‑square tests.
- What you can’t do: Compute mean, median, or any operation that assumes order.
- Real‑world example: Survey responses like “Yes,” “No,” “Maybe.”
Ordinal
- What you can do: Rank data, calculate median, use non‑parametric tests (Mann‑Whitney, Kruskal‑Wallis).
- What you can’t do: Treat the differences between ranks as equal; you can’t safely compute a mean.
- Real‑world example: Pain scale from 1 (none) to 10 (worst imaginable).
Interval
- What you can do: Compute mean, standard deviation, correlations.
- What you can’t do: Treat zero as a meaningful absence.
- Real‑world example: Calendar dates (year, month, day). The gap between 1999 and 2000 is the same as between 2000 and 2001, but 0 AD isn’t “no years.”
Ratio
- What you can do: All of the above plus percent change, fold change, coefficients of variation.
- What you can do: Compare ratios meaningfully because zero means “nothing.”
- Real‑world example: Body weight in kilograms. A weight of 0 kg means the person has no body mass.
Common Mistakes / What Most People Get Wrong
-
Assuming interval data is always ratio
Many think “if I can calculate a mean, it’s a ratio scale.” That’s false. Temperature in Celsius is interval, not ratio, because 0 °C isn’t the absence of temperature. -
Ignoring the true zero in ratio data
Forgetting that zero is meaningful can lead to misinterpretation. Here's a good example: saying “income doubled from $20k to $40k” is fine, but saying “income doubled from $0 to $20k” is nonsense because you can’t have half of zero But it adds up.. -
Using parametric tests on ordinal data
A common error is feeding Likert-scale data into a t‑test. Non‑parametric alternatives are safer unless you have strong evidence the data approximate interval properties. -
Treating dates as interval
Calendar dates are ordinal (you can order them) but not interval because the spacing between years isn’t uniform—leap years, month lengths, etc.
Practical Tips / What Actually Works
- Check the zero: Ask “does zero mean nothing?” If yes, it’s ratio; if no, it’s interval or lower.
- Ask about equal intervals: If you can’t guarantee the same difference between categories, it’s not interval.
- Use the right test:
- Nominal → chi‑square, Fisher’s exact.
- Ordinal → Mann‑Whitney, Spearman.
- Interval/Ratio → t‑tests, ANOVA, Pearson correlation.
- Label your variables: When writing a report, state the level of measurement. It signals to reviewers you understand the data’s nature.
- When in doubt, be conservative: If you’re unsure whether a scale is interval or ordinal, default to non‑parametric methods.
FAQ
Q1: Is a Likert scale interval or ordinal?
Most researchers treat it as ordinal because the distances between points aren’t guaranteed equal. Some argue it can be approximated as interval if you have enough points and justify it statistically.
Q2: Can I calculate a mean for ratio data?
Absolutely. Ratio data support all arithmetic operations, including mean, median, mode, standard deviation, and percent change Not complicated — just consistent. No workaround needed..
Q3: What about “cardinal” or “categorical” terms?
Those are just alternative names for nominal (categorical) and ratio (cardinal) scales. In modern stats, we stick to the four classic levels.
Q4: Does the level of measurement affect sample size calculations?
Yes—because the choice of test (parametric vs. non‑parametric) influences power and required sample size.
Q5: Is “ordinal” the same as “ranked data”?
Essentially, yes. Ranked data are a subset of ordinal data where the only information is order, not magnitude.
Closing
When you’re staring at a data set and thinking “what’s the right way to analyze this?Remember, the goal isn’t just to crunch numbers—it’s to let the data speak accurately. * If yes, you’re in the ratio world; if no, keep asking about equal intervals. ” start with the simple question: *Does zero mean none?Even so, once you nail the level, the rest of the analysis falls into place. And that starts with understanding the scale they’re on Simple as that..
Final Thoughts
Understanding the level of measurement isn’t just an academic exercise—it’s the linchpin that keeps your entire analytical pipeline honest. A single mis‑labelled variable can cascade into wrong tests, misleading effect sizes, and, ultimately, faulty conclusions that can ripple through peer review, grant applications, and policy decisions.
Easier said than done, but still worth knowing.
- Audit your data: Before you even write a single line of R or SPSS syntax, run a quick audit. List each variable, note its conceptual meaning, and decide its measurement level.
- Document everything: In your codebook or data dictionary, write the measurement level next to the variable name. Future you (and any collaborators) will thank you.
- When in doubt, test: If you’re unsure whether a variable is truly interval, run a normality test (Shapiro–Wilk, Kolmogorov–Smirnov) and a visual check (histogram, Q–Q plot). If the data are heavily skewed or have obvious gaps, lean toward non‑parametric methods.
- Keep your assumptions visible: Every statistical test has assumptions. Make them explicit in your methods section. If you’re using a t‑test on Likert data, justify the interval assumption or explain why you opted for a median‑based alternative.
A Quick Decision Flow
| Question | Answer | Recommended Approach |
|---|---|---|
| Does zero mean “none” or “nothing”? But | ||
| No | ||
| Can you only rank observations? | Yes | Ratio scale → parametric tests, compute means, use regression. |
| No | ||
| Is the variable purely categorical? | Yes | Interval scale → parametric tests, Pearson correlation, ANOVA. |
| No | ||
| Are the distances between categories equal? | Yes | Ordinal scale → Mann‑Whitney U, Kruskal–Wallis, Spearman. |
Wrap‑Up
The four classic measurement levels—nominal, ordinal, interval, and ratio—are more than a taxonomy; they are the compass that guides every statistical decision you make. By rigorously determining a variable’s level, you:
- Choose the correct inferential tools that respect the data’s structure.
- Avoid misleading metrics (e.g., mean of a nominal variable).
- Enhance reproducibility by making your analytic choices transparent.
- Increase statistical power by matching the test’s assumptions to the data.
Remember: the simplest rule of thumb is to ask whether zero has a meaningful interpretation. From there, the rest of the decision tree follows. Treat your data with the respect it deserves, and the numbers will lead you to sound, credible insights.
