Written and maintained by CASRAI Editorial Board
Last updated
An ordinal survey question asks respondents to choose from answer categories that have a natural order (excellent, very good, good, fair, poor) but no known, equal distance between neighbouring categories. This guide gives examples of ordinal survey questions taken word for word from two US federal health surveys, shows what makes each one ordinal, and covers the coding and analysis mistakes that real data files invite.
What is an ordinal survey question?
An ordinal survey question is a closed-ended question whose response options can be ranked from lowest to highest (or least to most), where you can say that one answer is more than another but cannot say by how much. It sits at the ordinal level in the levels of measurement: more informative than a nominal category, less informative than a numeric count or amount.
Two features identify an ordinal survey question. The options have a fixed order, and the numeric codes attached to them are labels for rank, not measurements. “Good” coded 3 is better than “Fair” coded 4, but the step from “Good” to “Very good” is not known to equal the step from “Fair” to “Good”. Agreement scales are the best-known family; the Likert scale guide covers those. This page deals with the wider set of ordinal items that are not agreement statements.
Ordinal survey question examples from real CDC surveys
The items below are quoted from the CDC’s 2023 Behavioral Risk Factor Surveillance System (BRFSS) questionnaire and the 2026 National Health Interview Survey (NHIS) questionnaire, with wording and codes last checked against the questionnaires on 2026-10-29. They are fielded in national surveys, not invented teaching examples.
Self-rated health (five ordered categories)
BRFSS item CHS.01 asks, “Would you say that in general your health is…” with the options read aloud: 1 Excellent, 2 Very Good, 3 Good, 4 Fair, 5 Poor. NHIS item PHSTAT_A uses the same five categories: “Would you say your health in general is excellent, very good, good, fair, or poor?” This is the standard ordinal survey question example: the categories are ranked, and nothing says the gaps are equal.
Highest level of education (ordered brackets)
BRFSS item CDEM.05, “What is the highest grade or year of school you completed?”, has six substantive options: 1 Never attended school or only attended kindergarten; 2 Grades 1 through 8 (Elementary); 3 Grades 9 through 11 (Some high school); 4 Grade 12 or GED (High school graduate); 5 College 1 year to 3 years (Some college or technical school); 6 College 4 years or more (College graduate). The order is clear, but the brackets cover very different numbers of school years, so the codes cannot be treated as years.
Household income bands (ordered, unequal widths)
BRFSS item CDEM.15, “Is your annual household income from all sources…”, offers eleven bands from “Less than $10,000” (code 01) up to “$200,000 or more” (code 11). Band widths change as income rises: $5,000 wide for $10,000 to under $15,000, $10,000 wide for $25,000 to under $35,000, $25,000 wide for $50,000 to under $75,000, and $50,000 wide for $100,000 to under $150,000.
Worry (three ordered categories)
An NHIS item on paying medical bills asks, “Are you very worried, somewhat worried, or not at all worried?” with codes 1 Very worried, 2 Somewhat worried, 3 Not at all worried. Fewer categories, same logic.
Frequency of symptoms (four ordered categories)
The NHIS mental health items ask, “Over the last two weeks, how often have you been bothered by … Little interest or pleasure in doing things? Would you say not at all, several days, more than half the days, or nearly every day?” with codes 1 Not at all, 2 Several days, 3 More than half the days, 4 Nearly every day. Frequency wording is another common source of ordinal survey questions; survey question types places these among the other formats.
Ordinal vs numeric versions of the same question
The same topic can be asked at different levels of measurement, and the level decides the analysis. These pairs come from the BRFSS items above.
| Topic | Ordinal version | Other version |
|---|---|---|
| Income | CDEM.15: choose one of eleven bands | A respondent types an exact dollar amount (numeric) |
| Health | CHS.01: excellent to poor | CHD.01: number of days in the past 30 that physical health was not good, entered as a count from 01 to 30, with 88 meaning none (a count) |
Both health items are about health, but only the days count supports arithmetic such as a mean number of unhealthy days. The five-category rating supports medians, percentages and rank-based methods. More on closed formats is in closed-ended questions in research.
Formula: summarising an ordinal item
Ordinal items are summarised with counts, percentages, cumulative percentages and the median category, not the mean.
Cumulative percentage up to category k: Ck = 100 × (n1 + n2 + … + nk) / N
- nj is the number of valid responses in category j, with categories numbered in their natural order.
- N is the total number of valid responses (missing codes excluded).
- k is the position of the category you are accumulating up to.
- The median category is the first category whose Ck reaches 50 or more.
Worked example with real arithmetic
The counts below are invented for illustration. They are not from any survey. Suppose 200 respondents answered the five-category self-rated health item with valid responses.
| Category (code) | n | Percent | Cumulative n | Cumulative percent |
|---|---|---|---|---|
| Excellent (1) | 30 | 15.0 | 30 | 15.0 |
| Very good (2) | 62 | 31.0 | 92 | 46.0 |
| Good (3) | 66 | 33.0 | 158 | 79.0 |
| Fair (4) | 30 | 15.0 | 188 | 94.0 |
| Poor (5) | 12 | 6.0 | 200 | 100.0 |
Arithmetic: N = 30 + 62 + 66 + 30 + 12 = 200. C2 = 100 × 92 / 200 = 46.0, which is below 50. C3 = 100 × 158 / 200 = 79.0, which reaches 50. The median category is therefore Good: the 100th ranked respondent falls in the third category, since 92 respondents are in the first two and 158 in the first three.
The mean of the codes would be (30×1 + 62×2 + 66×3 + 30×4 + 12×5) / 200 = 532 / 200 = 2.66. That number exists, but it assumes each step on the scale is the same size, which an ordinal item does not promise. Report the distribution and the median instead.
Interpretation: reading guidance
| Result | How to read it |
|---|---|
| Median category | Half of respondents chose this category or a higher-ranked one, and half chose it or a lower-ranked one |
| Cumulative percent | The share at or above a point on the scale, such as “good or better” (79.0% in the example) |
| Percent in each category | The full distribution; the safest summary when categories are few |
| Mean of codes | Not meaningful on its own, because the gaps between categories are unknown |
Assumptions and common mistakes
- Non-substantive codes are not on the scale. In BRFSS CHS.01 the codes 7 (Don’t know/Not sure) and 9 (Refused) sit numerically after 5 (Poor); NHIS uses 7 Refused, 8 Not Ascertained and 9 Don’t Know. Left in, a median or correlation treats “Refused” as a worse health state than “Poor”. Recode them to missing before analysis.
- Treating bracket codes as quantities. Income code 06 is not twice income code 03. Band widths differ, so the codes carry rank only.
- Averaging without thought. A mean of codes assumes equal intervals. See the Likert guide for the debate about when this is tolerated.
- Collapsing categories after seeing the data. Merge categories according to a plan made before analysis, and state it.
- Wrong tests. Use rank-based tests or models built for ordered outcomes, such as the proportional-odds ordinal logistic regression, instead of a method that assumes a numeric outcome.
Writing and coding ordinal questions
- Order the options consistently from low to high or high to low, and keep the direction the same across a questionnaire.
- Make categories mutually exclusive and cover the whole range, as the income bands do with their open-ended top and bottom bands.
- Store the response order in your codebook and give “don’t know” and “refused” codes outside the 1 to k range, as the CDC files do, so they can be set to missing.
- Label every option, not just the end points, so respondents interpret the steps the same way.
How to report it (APA example)
Illustrative text using the invented counts above: “Self-rated health was measured with a single five-category item (1 = excellent, 5 = poor). The median response was good (Mdn = 3), and 79.0% of respondents (158 of 200) rated their health as good or better.”
How to calculate it in Excel, R or Python
Each snippet assumes valid answers are coded 1 to 5 and that 7 and 9 mean “Don’t know” and “Refused”.
Excel
=MEDIAN(IF(A2:A203<=5,A2:A203)) (array formula; Ctrl+Shift+Enter in older Excel)
=COUNTIFS(A2:A203,">=1",A2:A203,"<=3")/COUNTIFS(A2:A203,">=1",A2:A203,"<=5")R
x <- c(rep(1,30), rep(2,62), rep(3,66), rep(4,30), rep(5,12), 7, 9)
x[x > 5] <- NA
h <- factor(x, levels = 1:5, ordered = TRUE,
labels = c("Excellent","Very good","Good","Fair","Poor"))
round(cumsum(prop.table(table(h))) * 100, 1)
median(as.integer(h), na.rm = TRUE) # 3 = GoodPython (pandas)
import pandas as pd
codes = pd.Series([1]*30 + [2]*62 + [3]*66 + [4]*30 + [5]*12 + [7, 9])
codes = codes.where(codes <= 5) # 7 and 9 become missing
counts = codes.value_counts().sort_index()
print((counts.cumsum() / counts.sum() * 100).round(1))
print(codes.median()) # 3.0 = GoodFAQ
What are ordinal questions in a survey?
They are closed-ended questions whose answer options have a clear rank order without known equal spacing, such as excellent to poor health or income bands.
Is a Likert question ordinal?
Each Likert item has ordered response categories, so it is ordinal at the item level. Whether a summed Likert scale is analysed as ordinal or approximately interval is a separate decision, discussed in the Likert scale guide.
Is household income an ordinal question?
It is ordinal when asked in bands, as in BRFSS CDEM.15. Asked as an exact dollar amount, it is a numeric (ratio-level) question.
Can I average an ordinal survey question?
You can compute a mean of the codes, but it assumes equal steps between categories. Medians, percentages and cumulative percentages describe ordinal data without that assumption.
What are more examples of ordinal variables?
The categorical variable examples guide lists ordinal, nominal and binary variables used in research.
Sources
- Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System, 2023 Questionnaire (items CHS.01, CDEM.05, CDEM.15, CHD.01). https://www.cdc.gov/brfss/questionnaires/pdf-ques/2023-BRFSS-Questionnaire-508.pdf (date checked: 2026-10-29).
- National Center for Health Statistics. 2026 National Health Interview Survey questionnaire, document version 03 June 2026 (item PHSTAT_A, medical-bill worry items and the mental health items on how often respondents were bothered by problems). https://ftp.cdc.gov/pub/Health_Statistics/NCHS/Survey_Questionnaires/NHIS/2026/EnglishQuest.pdf (date checked: 2026-10-29).
