Quick answer
An ordinal variable is a categorical variable whose values have a meaningful order but no equal spacing, such as a 1–5 satisfaction score, plan tier or priority level. Because the gaps are unequal, an average can mislead: summarize it with the distribution, the median and the top-two-box share.
Formula: Top-two-box share = ratings in the two highest categories ÷ all ratings
What is an ordinal variable?
An ordinal variable is a categorical variable whose categories have a meaningful order, while the distances between neighboring categories are unknown or unequal. A satisfaction rating from 1 to 5, a T-shirt size, a support priority from P1 to P4 and an education level are all ordinal: you can say which value is higher, but not by how much.
That missing piece, equal spacing, is the whole point. Going from "dissatisfied" (2) to "neutral" (3) is not necessarily the same step as going from "satisfied" (4) to "very satisfied" (5). The numbers 1–5 are labels in order, not measurements. So an average of ratings can hide what customers actually said, and two teams with the same average score can have very different customers. Read ordinal data through its distribution, median and top-two-box share instead.
What does ordinal mean in statistics?
"Ordinal" is one of the four levels of measurement S. S. Stevens proposed in 1946. Each level adds one property to the previous one, and each property unlocks more arithmetic:
| Level | Order? | Equal gaps? | True zero? | Examples | Valid summaries |
|---|---|---|---|---|---|
| Nominal | No | No | No | Country, browser, payment method | Count, share, mode |
| Ordinal | Yes | No | No | CSAT 1–5, plan tier, priority P1–P4 | Above + median, percentiles, rank tests |
| Interval | Yes | Yes | No | Temperature in °C, calendar year | Above + mean, differences |
| Ratio | Yes | Yes | Yes | Revenue, session length, page views | Above + ratios ("twice as long") |
Nominal and ordinal variables are both categorical. Interval and ratio variables are numeric measurements. Ordinal sits in between: it carries more information than a label, less than a measurement. The broader map of types is in data types.
Examples of ordinal data
Ordinal data shows up wherever people rank, rate or bucket things:
- Satisfaction ratings: CSAT on a 1–5 scale, app store stars, "How did we do?" thumbs scales with three or more steps.
- Agreement scales (Likert-type items): strongly disagree, disagree, neutral, agree, strongly agree.
- Tiers and levels: free, pro and business plans; bronze, silver and gold loyalty levels; beginner, intermediate and advanced.
- Severity and priority: P1–P4 tickets, low/medium/high risk, bug severity.
- Binned numbers: age bands (18–24, 25–34), income brackets, company size (1–10, 11–50, 51–200 employees). The underlying number is a ratio variable, but once binned with unequal widths it becomes ordinal.
- Rank positions: 1st, 2nd and 3rd in a race or a search result list. The gap between 1st and 2nd says nothing about the gap between 2nd and 3rd.
Why averaging a 1–5 score misleads: a worked example
Sample data: two support teams each collect 100 satisfaction ratings in a month, on a 1–5 scale.
| Score | Team A | Team B |
|---|---|---|
| 1 (very dissatisfied) | 0 | 26 |
| 2 | 10 | 5 |
| 3 | 30 | 3 |
| 4 | 50 | 15 |
| 5 (very satisfied) | 10 | 51 |
| Total ratings | 100 | 100 |
Now compute the usual summaries:
- Mean. Team A: (0×1 + 10×2 + 30×3 + 50×4 + 10×5) ÷ 100 = 360 ÷ 100 = 3.6. Team B: (26×1 + 5×2 + 3×3 + 15×4 + 51×5) ÷ 100 = 360 ÷ 100 = 3.6. Identical.
- Median. With 100 ratings, the median sits between the 50th and 51st rating in sorted order. Team A's cumulative count reaches 40 at score 3 and 90 at score 4, so both middle ratings are 4: median 4. Team B's cumulative count reaches 49 at score 4, so the 50th and 51st ratings are both 5: median 5.
- Top-two-box share (scores 4 and 5). Team A: 50 + 10 = 60%. Team B: 15 + 51 = 66%.
- Bottom-two-box share (scores 1 and 2). Team A: 0 + 10 = 10%. Team B: 26 + 5 = 31%.
The mean says the teams are the same. The median and the top-two share say Team B is better. The bottom-two share says nearly a third of Team B's customers were unhappy, three times Team A's rate. Team A is consistent; Team B is polarized, delighting most customers and failing a large minority. Those call for different fixes, and the average hid both stories.
The rule that follows: report the distribution, and always pair the top-two share with the bottom-two share and the median. One number on its own is not enough for ordinal data. To turn your own rating counts into a CSAT score, use the CSAT calculator, then look at the bottom-two share next to it.
Which statistics work for ordinal variables?
| Question | Use | Avoid |
|---|---|---|
| What is typical? | Median, mode | Mean as the headline |
| How spread out? | Interquartile range, percentiles, full distribution | Standard deviation |
| What share is positive or negative? | Top-two and bottom-two box shares | A single "satisfaction index" without the split |
| Do two groups differ? | Mann–Whitney U (rank-sum) test | t-test on the raw codes |
| Does it change with another ordered variable? | Spearman rank correlation | Pearson correlation on the codes |
The Mann–Whitney U test ranks all observations from both groups together and asks whether one group's ranks run systematically higher. It uses only order, so it suits ordinal data. On the 200 ratings above it gives z ≈ −2.33 and p ≈ 0.02: the two teams' rating distributions differ even though their means are the same. A t-test on the codes would find no difference at all, because it compares means.
Means of ordinal scales are not banned. Many teams track an average CSAT over time, and with a fixed scale and stable distribution a moving average can flag a change. The risk is in comparing averages across groups whose distributions have different shapes, as in the example.
Common mistakes with ordinal variables
- Treating labels as equally spaced. Coding "poor, fair, good, excellent" as 1–4 and averaging assumes poor-to-fair equals good-to-excellent.
- Comparing averages across different scales. A 4.2 on a 5-point scale and an 8.4 on a 10-point scale are not the same score, and rescaling does not fix it because respondents use the scales differently.
- Hiding the denominator. "92% satisfied" from 25 ratings is far shakier than from 2,500; one customer switching from satisfied to unhappy moves the first figure by 4 points (23 to 22 of 25).
- Reversing a scale halfway. If 1 meant "best" in last year's survey and "worst" this year, every trend line breaks.
- Collapsing categories inconsistently. Calling 4–5 "satisfied" in one report and 5 only in another makes the numbers incomparable.
- Throwing away the order. Treating ordinal data as nominal is safe but wasteful; a rank-based test usually detects differences a chi-square test on the categories would miss.
How to raise satisfaction once you can measure it properly is covered in how to improve CSAT. Open-ended feedback, which is not ordinal at all, belongs to qualitative data.
Reading satisfaction scores in TapCub
The two-team example is exactly what a support lead sees when comparing agents. Here is how to apply the method to TapCub's numbers:
- Find the score. Live chat › Reports shows Satisfaction: a score from 1 to 5 that the visitor can give within 24 hours after a conversation is resolved, one score per conversation, averaged to one decimal. The same report breaks it down per agent under agent workload.
- Treat the average as a pointer, not a verdict. The report shows the mean, so two agents at 3.6 can look identical, like Teams A and B. Before ranking agents, check how many ratings each has and look at the individual scores in the Resolved list of Live chat › Inbox, where each resolved conversation shows its satisfaction score.
- Chart your own ordinal properties as a distribution. If you track ratings as an event property, All-platform analytics › Behavior › Behavior analysis has a distribution view for a numeric property: a histogram with each bucket's count and share, plus the mean and the median, so the shape of the scores is visible next to the average.
- Know the boundary. TapCub does not run surveys such as NPS or CES questionnaires; the chat satisfaction score is the rating it collects itself.
See also: What Is Ordinal Data? 8 Examples and Which Statistic to Use.
See also: What is categorical data? Types, examples and how to analyze it.
Sources
- Stevens, S. S. (1946). On the Theory of Scales of Measurement. Science, 103(2684), 677–680.
- Mann, H. B., & Whitney, D. R. (1947). On a Test of Whether One of Two Random Variables is Stochastically Larger than the Other. Annals of Mathematical Statistics, 18(1), 50–60.
- NIST/SEMATECH e-Handbook of Statistical Methods, 7.3.5 Do two arbitrary processes have the same central tendency? (Mann–Whitney rank-sum procedure).
- UCLA OARC Statistical Consulting, What is the difference between categorical, ordinal and interval variables?
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Data Analyst
Explores traffic sources, conversion funnels, and retention, using clear metric definitions and concrete examples to explain what the numbers mean.