Metric data is numerical data measured on a scale with equal intervals between values. It covers two of the four levels of measurement: interval data and ratio data. Because the gaps between values are meaningful, you can calculate averages, spread, and correlations with confidence.
That flexibility is why metric data sits at the center of marketing and business research. Income, sales, age, satisfaction scores, and test results all fall into this group.
In this guide, we’ll define metric data, show how it differs from non-metric data, and walk through examples, analysis methods, and common mistakes.
What is metric data?
Metric data is quantitative data where the distance between any two values is equal and meaningful. A rise from 20 to 30 means the same change as a rise from 70 to 80.
Statisticians also call it scale data, continuous data, or parametric-ready data. Software such as SPSS labels it “Scale.” The term covers two levels of measurement, first formalized by psychologist S. S. Stevens in On the Theory of Scales of Measurement in Science.
- Interval data has equal spacing but no true zero.
- Ratio data has equal spacing and a true zero.
If you can add, subtract, and average the values and the results mean something, you are working with metric data. If you can only count categories or rank them, you are not.
Metric vs non-metric data: Where is the line?
Metric data uses numbers that measure an amount. Non-metric data uses labels or ranks. The difference decides which statistics are valid.
Non-metric data splits into nominal and ordinal levels. Nominal data names categories, such as region or product type. Ordinal data ranks them, such as “low, medium, high,” without promising equal gaps.
| Level | Type | Order | Equal gaps | True zero | Valid summary |
|---|---|---|---|---|---|
| Nominal | Non-metric | No | No | No | Mode, counts |
| Ordinal | Non-metric | Yes | No | No | Median, percentiles |
| Interval | Metric | Yes | Yes | No | Mean, standard deviation |
| Ratio | Metric | Yes | Yes | Yes | Mean, ratios, coefficient of variation |
For a full walkthrough of all four levels, see the QuestionPro guide on levels of measurement.
Metric data vs metrics
The two terms sound alike but mean different things. Metric data describes how a variable is measured. A metric, such as churn rate or customer acquisition cost, is a number you track to judge performance. Most business metrics are built from metric data, which is why quantitative metrics and metric data often appear together.
- Metric data: A property of how a variable is measured.
- A metric: A tracked number, such as monthly churn rate.
What is the difference between interval and ratio data?
The only difference between interval and ratio data is the zero point. Ratio data has a true zero that means “none of the thing.” Interval data has an arbitrary zero.
- Interval: 0 is a position on the scale, not an absence.
- Ratio: 0 means nothing is present.
Interval data
This kind of data lets you add and subtract but not multiply or divide meaningfully. Temperature in Celsius is the classic case. 0°C does not mean “no heat,” so 40°C is not “twice as hot” as 20°C.
Other examples include calendar years, IQ scores, and SAT scores. You can find the average and standard deviation of any of them. You cannot say one score is double another. The QuestionPro page on interval data covers its properties in more depth.
Ratio data
Ratio data supports every arithmetic operation, including ratios. A household earning $100,000 earns twice as much as one earning $50,000. Zero dollars means no income at all.
Common ratio variables include:
- Age and height
- Time spent on a task
- Units sold and order value
The U.S. Bureau of Labor Statistics reports the unemployment rate as a ratio-scale figure because zero means nobody is unemployed. You can read more about it in the QuestionPro guide to ratio data.
Metric data examples from business and surveys
The fastest way to spot metric data is to ask whether the number measures an amount on a fixed scale. These examples show how common it is.
Ratio examples
- Annual household income, a core variable in the U.S. Census Bureau’s income data
- Monthly revenue per customer
- Minutes spent on a checkout page
- Number of support tickets per week
- Weight of a shipped package
Interval examples
- Temperature in Celsius or Fahrenheit
- Standardized test scores
- Time of day on a 24-hour clock
- Calendar year of purchase
A worked example
Imagine a coffee chain surveying 400 customers. The survey asks for weekly spend in dollars (ratio data) and overall satisfaction on a 0 to 10 scale. The team can report average weekly spend, compare it across store locations, and test whether higher satisfaction goes with higher spend. This example is illustrative, not real data.
Are rating scales metric data?
Individual rating items are technically ordinal, but researchers often treat them as interval. That choice is the most debated point in survey analysis.
- Purist view: Psychometric scales are ordinal.
- Pragmatic view: Well-built scales behave like interval data.
The purist view says a 1 to 5 agreement item only ranks answers. No one can prove the gap between “agree” and “strongly agree” equals the gap between “neutral” and “agree.” The pragmatic view, common in marketing research, treats well-built rating scales as interval so researchers can use means and multivariate methods.
Evidence supports the pragmatic view in many cases. In a widely cited paper, Norman’s review of Likert scales and parametric statistics argues that parametric tests hold up well with Likert-type data, even with small samples and skewed responses.
Which scales usually count as metric?
- Summed Likert scales: Adding several items into one score is usually treated as interval.
- 0 to 10 scales: Satisfaction, likelihood to recommend, and perceived economic conditions are commonly treated as interval. The Net Promoter Score is built on a 0 to 10 item.
- Single Likert items: Safest as ordinal, though many analysts still report means.
The Likert scale examples page shows common formats, including 5, 7, and 9 point versions.
How do you decide whether to treat data as metric?
Ask three questions in order. The answers place any variable on the right level of measurement.
- Can the values be ranked?
If not, it is nominal. Stop.
- Are the gaps between values equal and meaningful?
If not, it is ordinal. Use medians and rank-based tests.
- Is there a true zero?
If yes, it is ratio. If no, it is interval.
For rating scales, add a fourth check. Does the scale have at least five points, clear labels, and consistent wording? If so, treating it as interval is defensible. If the scale is short, uneven, or unlabeled, stay with ordinal methods.
Write your decision into the analysis plan before you open the data. That keeps the choice from being driven by which test gives the better result.
How to analyze metric data
Start with a description, then move to testing. Metric data supports the widest range of statistics, so the order matters more than the technique.
- Describe first.
- Check shape and outliers second.
- Test last.
Step 1: Describe the distribution
Report these four figures for every metric variable:
- Mean and median
- Range
- Standard deviation, the typical distance from the mean
- Sample size
A large gap between mean and median signals skew. Plot a histogram before you trust any average.
Step 2: Check for outliers and shape
Income, spend, and time-on-task are usually right-skewed. A few very large values pull the mean up. Look at the data visually and decide how to handle extreme values.
Step 3: Match the test to the question
| Question | Common test | Non-parametric option |
|---|---|---|
| Do two groups differ? | Independent t-test | Mann-Whitney U |
| Do three or more groups differ? | One-way ANOVA | Kruskal-Wallis |
| Are two variables related? | Pearson correlation | Spearman correlation |
| Can one variable predict another? | Linear regression | Rank-based regression |
Parametric tests assume data is roughly normal and observations are independent. When the data breaks those assumptions badly, use the non-parametric option in the right-hand column.
How do you transform and group metric data?
Transformations reshape skewed data so it meets test assumptions. Grouping turns numeric answers into bands for easier reporting.
Ratio data can be log-transformed or square-root-transformed to pull a long right tail closer to a normal curve. Report results on the original scale where you can, so readers can interpret them.
Numeric responses can also be split into quartiles, deciles, or terciles. A tercile split cuts respondents into low, middle, and high thirds. This works only for numeric answers, not for categories. Grouping discards detail, so keep the raw values and create the bands as a new variable.
How to collect metric data in surveys
Good metric data starts with question design. The format you pick fixes what level of measurement you get.
- Numeric input: Ask for exact amounts, such as age or monthly spend. This gives true ratio data.
- Slider or 0 to 10 scale: Capture strength of opinion with finer steps.
- Constant sum: Ask respondents to split 100 points across options.
- Rating matrix: Collect several items on one scale, then sum them into a composite score.
- Open numeric ranges: Avoid wide bands like “$50,000 to $100,000” when you need exact values. Bands turn ratio data into ordinal data.
Plan these formats before fielding. Rich scale questions give strong insight, but a poor design cannot be fixed afterward.
How QuestionPro helps you capture and analyze metric data
QuestionPro survey software supports numeric input, slider, constant sum, and rating matrix questions, so you choose the measurement level when you build the survey. Responses export to Excel or SPSS for deeper statistical work.
Inside the platform, cross-tab reports let you compare average scores across segments, such as age group or region. That helps you test whether a metric result holds across audiences before you export anything. Check your plan’s feature list for the exact analysis tools you have.
Common mistakes when working with metric data
Most errors come from forcing the wrong level onto a variable. These are the ones to avoid.
- Averaging numeric codes for categories.
Coding “red = 1, blue = 2” and averaging the codes is meaningless.
- Computing ratios on interval data.
“Twice as hot” fails on Celsius.
- Ignoring skew.
A mean income without a median hides the real picture.
- Binning ratio data too early.
You lose detail you cannot recover.
- Mixing scales.
Combining a 1 to 5 item and a 0 to 10 item without standardizing distorts totals.
- Skipping the assumption check.
A t-test on heavily skewed data can mislead.
Choosing the right level before you collect a single response
Metric data is powerful because its numbers carry real meaning. Equal intervals let you average, compare, and model. A true zero lets you state ratios. The skill lies in knowing which of your variables earn those privileges and which only look numeric.
Decide the measurement level while you write the question, not after the responses arrive. Pick the format that matches the analysis you plan to run.
Before you field any survey, confirm three things:
- Each variable has a stated level of measurement.
- Each planned test fits that level.
- You will keep the raw values for reanalysis.
Frequently Asked Questions (FAQs)
Almost. Quantitative data is any numeric data, including counts and codes. Metric data is the subset where intervals between values are equal and meaningful. A zip code is numeric but not metric because averaging zip codes tells you nothing.
A single 5-point item is strictly ordinal, because equal spacing cannot be proven. Researchers often treat summed multi-item scales as interval. The safest practice is to state your assumption and report medians alongside means.
Age in years is ratio data. Zero means no time lived, and a 40-year-old is twice as old as a 20-year-old. If you group age into bands such as 18 to 24, the result becomes ordinal.
With large samples, the t-test tolerates moderate non-normality. With small samples or heavy skew, switch to a Mann-Whitney U test or transform the data first. Always check a histogram before choosing.
Excel handles descriptive statistics and basic tests. SPSS, R, and Python cover regression, ANOVA, and non-parametric methods. Survey platforms export raw responses in these formats, so you can build the questionnaire in one tool and analyze in another.
