What Are Levels of Measurement?
Levels of measurement, also called scales of measurement, classify variables according to the mathematical properties of their values. Psychologist Stanley Smith Stevens proposed four levels in his 1946 paper: nominal, ordinal, interval, and ratio. Each level builds on the one below it, adding a new property. Nominal assigns labels. Ordinal adds rank order. Interval adds equal spacing between values. Ratio adds a true zero point. The level of measurement determines which statistical operations are valid, which visualizations make sense, and which conclusions you can draw from your data.
Why Levels of Measurement Matter in Research
Your measurement level dictates your analysis options. Running a mean on nominal data produces nonsense. Using a t-test on ordinal data violates assumptions that might (or might not) affect your conclusions. Stevens' framework gives you a quick way to check whether your analytical plan matches your data. According to a review published in Advances in Health Sciences Education, measurement-level errors account for roughly 20% of the statistical mistakes flagged during peer review of medical research manuscripts.
How the Four Levels Work
Nominal
Nominal measurement assigns observations to categories with no inherent order. The categories are mutually exclusive and exhaustive, but you can't rank them.
Properties: Identity (categories are distinct)
Examples:
- Blood type: A, B, AB, O
- Survey question: "Which department do you work in?"
- Customer segment: new, returning, churned
Valid operations: Counting, mode, chi-square test, frequency distribution
You can ask "which category has the most observations?" but not "which category is highest?" because the categories don't have a high-low dimension.
Ordinal
Ordinal measurement adds rank order to nominal categories. You know that one value is more or less than another, but you don't know by how much.
Properties: Identity + order
Examples:
- Education level: high school, bachelor's, master's, doctorate
- Customer satisfaction: very dissatisfied to very satisfied
- Pain scale: mild, moderate, severe
Valid operations: Everything valid for nominal, plus median, percentiles, Spearman correlation, Mann-Whitney U test
The gap between "mild" and "moderate" pain isn't necessarily the same as between "moderate" and "severe." This is why calculating means on ordinal data is technically inappropriate, even though it's widely practiced.
Interval
Interval measurement adds equal, defined distances between values. The difference between 10 and 20 is the same as between 80 and 90. But there's no true zero, zero doesn't mean "none."
Properties: Identity + order + equal intervals
Examples:
- Temperature in Fahrenheit or Celsius (0°F doesn't mean "no temperature")
- Year (the year 0 is arbitrary)
- IQ scores (0 IQ doesn't mean "no intelligence")
- Net Promoter Score (the 0-10 scale has equal spacing by design)
Valid operations: Everything valid for ordinal, plus mean, standard deviation, Pearson correlation, t-test, ANOVA
You can add and subtract interval values meaningfully, but ratios don't work. 80°F isn't "twice as hot" as 40°F.
Ratio
Ratio measurement has all the properties of interval measurement plus a meaningful zero point, where zero represents the complete absence of the measured quantity.
Properties: Identity + order + equal intervals + true zero
Examples:
- Income in dollars ($0 means no income)
- Age in years (0 means birth)
- Number of purchases (0 means none)
- Response time in seconds (0 means instant)
Valid operations: Everything valid for interval, plus geometric mean, coefficient of variation, and meaningful ratios (someone with $80K income earns twice as much as someone with $40K)
Comparison Table
| Property | Nominal | Ordinal | Interval | Ratio |
|---|---|---|---|---|
| Distinct categories | Yes | Yes | Yes | Yes |
| Rank order | No | Yes | Yes | Yes |
| Equal intervals | No | No | Yes | Yes |
| True zero | No | No | No | Yes |
| Central tendency | Mode | Mode, median | Mode, median, mean | Mode, median, mean |
| Dispersion | Frequency | Range, IQR | Std. Deviation, variance | All, plus CV |
| Example | Gender | Satisfaction rating | Temperature (°C) | Revenue ($) |
| Can say "twice as much" | No | No | No | Yes |
Gray Areas in Practice
The biggest debate in applied research is where Likert scales fall. Technically, a 5-point agreement scale is ordinal. But decades of research show that treating scales with five or more balanced points as interval produces results that rarely differ meaningfully from ordinal methods, as long as sample sizes are adequate.
Survey design can also blur the lines. A frequency question asking "How many times per week do you exercise?" with options of "0, 1-2, 3-4, 5+" is ordinal (unequal bins). The same question asking for an exact number produces ratio data. How you ask the question determines your measurement level.
When to Use Each Level
- Nominal for demographic segmentation, categorical grouping, and any variable where order doesn't apply
- Ordinal for attitudes, satisfaction, preferences, and ranked data where you need order but can't define exact distances
- Interval for scales with defined, equal spacing but no meaningful zero (temperature, standardized test scores, calendar dates)
- Ratio for counts, amounts, durations, and any variable where zero means the absence of the thing being measured
Common Mistakes to Avoid
- Treating zip codes or phone numbers as numerical data: they're nominal even though they contain digits
- Running parametric tests on small ordinal samples without considering non-parametric alternatives
- Assuming a numbered scale is automatically interval: assigning numbers 1-5 to response options doesn't create equal intervals unless the scale is designed and validated for that purpose
- Collecting ratio-level data as ordinal by unnecessarily grouping continuous variables into brackets (income ranges instead of exact amounts), which discards information you could have kept
How Quali-Fi Supports All Measurement Levels
Quali-Fi's survey builder supports all four measurement levels through 50+ question types. The platform automatically maps each question type to its measurement level and applies appropriate default analysis, mode and chi-square for nominal questions, median and non-parametric tests for ordinal, means and parametric tests for interval and ratio. Real-time dashboards adapt their visualizations based on data type, showing frequency tables for nominal data and distribution charts with summary statistics for higher-level measurements.
Frequently Asked Questions
Does the level of measurement affect which chart I should use?
Yes. Nominal data works best with bar charts and pie charts. Ordinal data suits bar charts with ordered categories, stacked bars, and diverging bar charts. Interval and ratio data support histograms, box plots, scatter plots, and line charts. Using the wrong chart type can misrepresent your data, a line chart for nominal data implies a trend that doesn't exist.
Can I change the level of measurement after data collection?
You can move down levels (convert ratio to ordinal by grouping into bins) but you can't move up. Once you've collected ordinal data, you can't create equal intervals after the fact. This is why it's important to collect data at the highest measurement level practical, you can always simplify later.
What level of measurement is age?
It depends on how you measure it. Age as exact years and months is ratio (has a true zero at birth, equal intervals, meaningful ratios). Age in grouped brackets (18-24, 25-34, 35-44) is ordinal. Always collect the most granular version your respondents will tolerate.