Why Ordered Categories Need Careful Preparation
Many variables have values that can be placed in a meaningful order but do not measure amounts on a scale with known, equal steps. A satisfaction response might range from “Very dissatisfied” to “Very satisfied.” The first response is lower than the last in satisfaction, but that ordering alone does not tell us how much satisfaction changes between adjacent responses.
Preparing this kind of variable means preserving two things: the response labels that explain what each value means, and the documented order that expresses how the levels rank. It also means resisting a tempting but unsupported shortcut: treating the levels as if they were measured quantities merely because software stores them as numbers.
A nominal categorical variable has labels but no inherent ranking, such as blood type. For an ordinal variable, the ranking is part of the information. For example, sorting “moderate” before “mild” would lose or contradict the intended meaning. But an ordered ranking is not the same thing as a quantitative scale: placing one level above another does not establish that the gap between those levels equals the gap between any other adjacent pair.
Record the Meaning and the Order
The order should come from the question wording, the measurement instrument, or another clear definition of the categories—not from alphabetical order or the numerical order in which software happens to display labels. Keep the original response meaning visible and document the order explicitly. If numerical codes are used for storage, include a codebook that maps each code to its label and rank.
| Code | Recorded label | Documented rank |
|---|---|---|
| 1 | Very dissatisfied | 1, lowest satisfaction |
| 2 | Dissatisfied | 2 |
| 3 | Neither satisfied nor dissatisfied | 3 |
| 4 | Satisfied | 4 |
| 5 | Very satisfied | 5, highest satisfaction |
The codebook says what the codes stand for and what order they represent. It does not say that a one-unit increase in the code represents a fixed increase in satisfaction. The values 1 through 5 are identifiers for ordered response levels unless additional measurement justification supports treating them as quantitative scores.
Worked Example: Documenting a Satisfaction Scale
In a hypothetical service survey, respondents choose one of five answers: “Very satisfied,” “Satisfied,” “Neither,” “Dissatisfied,” or “Very dissatisfied.” The data file stores these answers as text. Prepare a clear ordered representation without claiming that satisfaction is measured in equal units.
Step 1: Identify the variable. The variable is each respondent’s satisfaction response. Its values are categories, not measured amounts such as minutes or dollars.
Step 2: Establish the order. The question’s response options imply a direction from least to greatest satisfaction: “Very dissatisfied,” “Dissatisfied,” “Neither,” “Satisfied,” “Very satisfied.” This order should be recorded as the category order rather than inferred from the alphabetic spelling of the labels.
Step 3: Preserve the labels. A codebook could assign ranks 1 through 5 in that order, while retaining the response text. For instance, code 1 means “Very dissatisfied” and code 5 means “Very satisfied.” A reader can then interpret a code by consulting the documented mapping.
Conclusion. The prepared variable is ordinal: its levels have an interpretable order. The coding does not establish that the difference between ranks 1 and 2 is the same size as the difference between ranks 4 and 5.
Order Does Not Guarantee Equal Intervals
The reason to be cautious can be shown directly. Suppose four hypothetical responses have categories “Low,” “Medium,” and “High,” with two responses in Low, one in Medium, and one in High. Codes 1, 2, and 3 represent the correct order. The numerical mean of these codes is:
Now keep the same category order but use the codes 1, 2, and 10 for Low, Medium, and High. The mean becomes:
Both coding schemes preserve exactly the same ordering: Low is below Medium, which is below High. Yet the numerical means differ substantially. The change came only from the spacing chosen for the codes, not from any change in the four responses. Therefore, the order alone cannot determine a meaningful numerical mean or the distances between response levels.
This argument does not mean that numerical summaries or models are never used with rating scales. It means that such use needs a stated rationale beyond the fact that the responses can be ranked. During data preparation, keep the distinction clear: store the categories and their order accurately, and do not silently turn ranks into purported amounts.
Worked Example: Interpreting a Five-Point Rating
A hypothetical questionnaire asks respondents to rate a clinic’s waiting-room comfort from 1, “Very uncomfortable,” through 5, “Very comfortable.” One respondent selects 2 and another selects 4. What can be concluded from these values?
The response options have an order, so the person selecting 4 chose a category ranked above the category selected by the person choosing 2. The values identify the ordered levels.
The subtraction \(4-2=2\) is an arithmetic fact about the codes, but the scale definition does not establish that the second respondent’s comfort is exactly two equal units higher. Nor does it establish that the change from 2 to 3 is the same size as the change from 3 to 4.
A correct preparation records the labels, their order, and the coding map. A careful interpretation states that the response of 4 is a higher rating category than the response of 2, without claiming a quantified difference in comfort.
Choose an Order That Matches the Variable
Some variables have a natural order that is clear from their definitions. For example, symptom severity categories may be “None,” “Mild,” “Moderate,” and “Severe.” Educational attainment categories may also be rankable when the categories are defined to reflect completed levels. The exact order still needs to be documented, especially when labels are abbreviated or codes are stored without the full wording.
Other categories can look ordered while the intended ranking is uncertain. A set of product names is not ordinal just because someone lists them in a sequence. Similarly, response categories such as “Often,” “Sometimes,” and “Rarely” need their direction recorded; displaying “Often” last does not change its meaning. If the order cannot be established from the question or variable definition, do not invent one during preparation.
Worked Example: Preparing Severity Categories
A hypothetical clinic form records a symptom as one of four categories: “Moderate,” “None,” “Severe,” or “Mild.” The entries appear in this order in the raw file. Prepare the category order and explain what it represents.
Step 1: Use the meaning of the levels. The response categories describe increasing symptom severity, so their intended order is “None,” “Mild,” “Moderate,” “Severe.”
Step 2: Document a mapping. A codebook may assign rank 1 to “None,” 2 to “Mild,” 3 to “Moderate,” and 4 to “Severe.” It should also retain the original labels so that rank 3 remains interpretable as “Moderate.”
Step 3: State the limit of the order. Rank 4 represents a more severe category than rank 3, but the categories do not specify how much more severe. The ranking alone does not make severity a quantitative measurement with equal intervals.
Conclusion. The raw display order is not the substantive order. A documented category order makes the intended ranking explicit while preserving the categorical nature of the responses.
Do Not Confuse Ordered Labels With Quantitative Measurements
A variable is not quantitative merely because its values contain digits. Consider age bands recorded as “0–17,” “18–39,” and “40 or older.” The bands can be ordered from younger to older, but a category code such as 1, 2, or 3 is not an individual’s age. The category “40 or older” also contains a range of ages rather than a single measured value. If the underlying age in years is available and relevant, it is a different variable from the age-band category.
Likewise, a response stored as the number 1 may mean “Strongly disagree,” not one unit of an underlying quantity. Its meaning depends on the survey’s codebook. Data preparation should preserve this distinction so that later summaries and analyses are not based on an accidental interpretation of storage format.
Worked Example: Ordered Age Bands Versus Age in Years
A hypothetical file contains one column called “Age group” with values “Under 18,” “18–39,” and “40 or older.” Another file records each person’s actual age in completed years. Explain how the variables differ and how to prepare the age-group column.
The “Age group” variable is ordinal because the groups can be ranked from younger to older. A documented order is “Under 18,” then “18–39,” then “40 or older.” A codebook may assign ranks 1, 2, and 3 in that order, while retaining the text labels.
The code 3 does not mean that a person is 3 years old, and the difference between codes 1 and 2 is not a measured age difference. The category “40 or older” does not even specify one particular age. By contrast, the separate age-in-years variable records a quantitative measurement with units of years.
The two columns therefore answer different questions: one records an age category, and the other records an age measurement. They should not be treated as interchangeable simply because the categories have a natural order.
Common Mistakes and a Preparation Check
- Using alphabetic order as the response order. Alphabetizing “Severe,” “Mild,” and “Moderate” does not represent increasing severity. Use the documented meaning of the levels.
- Assuming numbered codes are measurements. A code of 4 can identify a category without representing four units of anything. Consult the codebook and retain the label mapping.
- Assuming adjacent ranks are equally far apart. A rank establishes which category comes first, not the size of a difference. Do not infer equal intervals from the sequence of codes.
- Changing the substantive order while recoding. If larger codes are meant to indicate higher satisfaction, assigning the largest code to “Very dissatisfied” reverses the stated direction. Check the mapping against the response wording.
- Treating a missing response as the lowest category. A blank response does not mean “Very dissatisfied,” “None,” or any other endpoint. Keep nonresponse distinct from the ordered response levels.
Before using an ordered categorical variable, check that its labels are interpretable, its ranking has a defensible source, and its direction is stated. Confirm that any numerical codes map to the intended levels. Most importantly, do not let the storage format make a stronger measurement claim than the response scale supports.
Check Your Understanding
Use the distinction between ordered categories and quantitative measurements to answer these questions.
- A survey records “Rarely,” “Sometimes,” and “Often.” What order should be documented, and what does that order establish?
- A rating scale codes “Poor,” “Fair,” and “Good” as 1, 2, and 3. Does this coding alone establish equal intervals? Explain.
- A file stores “Mild,” “Moderate,” and “Severe” as 1, 2, and 3. What information should a codebook include?
- Why might the mean of ordinal category codes change under a different coding that preserves the same category order?
- A blank satisfaction response is encountered. Why should it not be assigned the lowest satisfaction rank?