What Is a Nominal Scale? How to Distinguish and Use It in Surveys
What Is a Nominal Scale
You open your data file and see that GIOITINH is coded as 1 = Male and 2 = Female, or that KHU_VUC uses 1 = Hanoi, 2 = Da Nang, and 3 = Ho Chi Minh City. These are typical examples of a nominal scale.
A nominal scale assigns observations to different groups. The number in the data is only a code that identifies the group. It does not show that one group is greater, better, or farther along than another. Therefore, a value of 2 does not mean “more” than a value of 1.
You can think of this scale as a label attached to each record. The labels 1, 2, and 3 help SPSS store and process the data efficiently, but their underlying meaning is still the name of a group. If a survey asks respondents to select their gender, field of study, marital status, customer type, or channel through which they heard about a brand, the variable is usually measured with a nominal scale.
A nominal scale usually allows you to determine only two things: which group an observation belongs to and whether two observations belong to the same group. The appropriate operations are mainly counting frequencies, calculating percentages, creating crosstabs, and testing relationships between categorical variables. You should not calculate the mean of gender codes or region codes and interpret the result as a meaningful quantity.
The Role of a Nominal Scale in Quantitative Research
In a quantitative thesis, a nominal scale helps you describe the sample and bring categorical characteristics into your analysis model. For example, you can report the percentages of male and female respondents, the age groups represented in the sample, or the channels through which customers heard about a service.
Its second role is to serve as a control variable or grouping variable. You can compare the mean purchase intention score across gender groups, or test whether purchase intention differs across regions. In that situation, you do not calculate the mean of the nominal variable itself. You use it to divide the sample into groups or convert it into dummy variables as required by the model.
Its third role is to help you select the correct analysis. With two nominal variables, a crosstab and a Chi-square test are often appropriate for checking whether the distribution of one variable is related to the distribution of the other. With one nominal variable used to form groups and one score-based outcome variable, you can consider a t-test or ANOVA depending on the number of groups and the data conditions. The test should be selected based on the research question, variable type, and statistical assumptions, not simply on the variable name.
You should keep the measurement scale of a categorical variable separate from what is a questionnaire. A questionnaire is the instrument used to collect data, while a nominal scale describes how a variable is recorded and handled after respondents complete the questionnaire.
When designing the questionnaire, label the groups clearly and make sure the options do not overlap. For example, the age groups “18 to 24” and “25 to 34” are easier to code than “under 25” and “25 or older” if you are checking the boundaries. If someone might not belong to any listed group, add an appropriate option such as “Other” or “Prefer not to answer”, then decide in advance how to handle that value.
What Counts as Acceptable for a Nominal Scale
A nominal scale does not have a threshold such as Cronbach's Alpha, factor loading, or AVE. A nominal variable is suitable for use when its groups are clearly defined, each observation belongs to one group, the groups cover the relevant possibilities, and the coding remains consistent from the questionnaire to the data file.
The table below is a decision checklist. The Likert principles in the relevant row are linked to the origin of the Likert technique. The other criteria are data-design and coding conditions that you need to check in your own questionnaire.
| What to check | Decision rule | Source or basis |
|---|---|---|
| Group labels | Each code is linked to one specific label, and the same code is not used for two meanings | Coding rule for nominal variables in data design |
| Mutual exclusivity | Each respondent is assigned to only one group, unless the question allows multiple answers | Rule for classifying data |
| Coverage | The options include appropriate choices for real cases, such as “Other” or “Prefer not to answer” | Rule for designing categorical questions |
| Code order | Codes 1, 2, and 3 identify groups only and must not be used to infer greater or lesser values | Nature of a nominal scale |
| Likert scale | If the options express agreement from low to high, the scale is a Likert scale rather than a purely nominal scale | (Likert, 1932) |
| Multiple-choice variable | Store each selectable option as a separate 0/1 variable when respondents can choose multiple answers | Rule for converting survey data into an analysis format |
You also need to distinguish between “acceptable” at two levels. The first level is whether the variable has been coded correctly from a methodological perspective. The second is whether there are enough observations in each group to perform the analysis you intend to use. A variable may theoretically have four groups but still be difficult to analyze if one group contains very few respondents.
There is no single percentage that applies to every nominal variable for the share of observations in each group. If one group has a low frequency, review the purpose of the analysis, combine groups only when there is a clear substantive basis, or choose a more suitable method. Do not combine groups simply to make the results table look balanced.
How to Read a Nominal Scale in the Output
In SPSS, you specify a nominal scale in the Measure column of Variable View. Set Nominal for variables such as gender, region, or access channel. Use the Values column to define labels, such as 1 = Male and 2 = Female. You can use the Missing column for missing codes such as 99, but you must define that code so SPSS does not count it as a real group.
To view the number and percentage of observations in each group, open Analyze > Descriptive Statistics > Frequencies. Move the variable you want to check into Variable(s), keep the frequency table option selected, and click OK. The output usually contains the columns Frequency, Percent, Valid Percent, and Cumulative Percent.
For a nominal variable, focus on Frequency, Percent, and Valid Percent. Cumulative Percent only describes accumulation according to the code order, while the code order of a nominal scale has no ordinal meaning. If there are missing values, Percent is calculated from all rows, whereas Valid Percent is calculated after missing cases are excluded. When writing your thesis, state clearly whether you are using the valid sample or the full number of observations.
The table below is illustrative output, constructed to match the format of SPSS output. The numbers only demonstrate how to read the table and are not results from an actual study.
| GIOITINH | Frequency | Percent | Valid Percent | Cumulative Percent |
|---|---|---|---|---|
| Male | 86 | 42.6 | 43.0 | 43.0 |
| Female | 114 | 56.4 | 57.0 | 100.0 |
| Total Valid | 200 | 99.0 | 100.0 | |
| Missing | 2 | 1.0 | ||
| Total | 202 | 100.0 |
The table shows 202 rows of data, including 2 cases with a missing value for GIOITINH. There are 200 valid observations, consisting of 86 male and 114 female respondents. Because there are missing values, the valid percentages are 43.0% and 57.0%, for a total of 100%. You cannot conclude that the female group is “higher” than the male group in the sense of the measurement scale. You can only state that the female group represents a larger share of the illustrative sample.
To examine the relationship between two nominal variables, use Analyze > Descriptive Statistics > Crosstabs. Place one variable in Row(s) and the other in Column(s), click Statistics to select Chi-square, then click Cells to select frequencies and percentages. Read the Chi-Square Tests table, especially the Pearson Chi-Square row and the Asymptotic Significance (2-sided) column. When reporting the result, also check the expected-frequency condition rather than looking at only one p-value.
How to Handle a Nominal Scale That Is Not Acceptable
If you find a coding error, work from the original data to the output in that order. This gives you a record that explains why the results table changed after the correction.
Check the Original Questionnaire and Codebook
Open the questionnaire and codebook beside the .sav or .csv file. Compare each code with its label and look for cases where 1 means Male in one part of the data but Female in another. If you do not have a codebook, create a table containing the variable name, content, value codes, and missing codes before making corrections.
Correct Labels in Variable View
In SPSS, go to Variable View and check Values, Missing, and Measure. If only the label is wrong while the original numerical code is correct, edit the label in Values instead of changing the data in bulk. Then rerun Analyze > Descriptive Statistics > Frequencies to check the distribution table.
Handle Out-of-Range Codes
If a variable should contain only codes 1 and 2 but includes 0, 3, or 99, determine whether the value is an entry error, an “other” response, or a missing value. Do not automatically delete every affected row. For an invalid code caused by an entry error, compare it with the questionnaire or original data and correct it. If you cannot determine the correct value, set it to missing and record the number of affected cases.
Combine Groups Only With a Substantive Reason
You may combine groups with very few observations if they are conceptually close and the combination fits your hypothesis. For example, some regional groups may be combined by area if the study only needs to compare the North, Central, and South. Keep the original variable, create a new variable with Transform > Recode into Different Variables, and record the reason for combining the groups in the methods chapter.
If the problem comes from wording that led respondents to understand the question differently, the software cannot repair the data. In that situation, acceptable options may include reporting the limitation, excluding the variable from the relevant analysis when there is a clear basis, or collecting additional data under the adjusted design. Changing codes to produce a more attractive distribution increases the risk of difficult questions about the original data.
Distinguishing a Nominal Scale From an Ordinal Scale
A common mistake is to treat every variable entered with numbers as a quantitative variable. Numbers are only a way to store data. You need to look at the meaning of the response options.
A nominal scale only classifies. For example, payment method may include cash, card, and e-wallet. There is no basis for saying that an e-wallet has a “higher” value than cash simply because it is coded as number 3.
An ordinal scale has an order, but the distances between levels are not necessarily equal. Satisfaction levels such as very dissatisfied, dissatisfied, neutral, satisfied, and very satisfied are common examples. If you use several items of this type to measure a construct with a Likert scale, present the scale and the data treatment consistently. See the article on the Likert scale if your file contains agreement levels.
An interval scale has an order and assumes equal distances between levels, while a ratio scale also has an absolute zero point. Age in years, income, and number of purchases are usually handled differently from group codes. Do not turn a nominal code into a measurement simply because SPSS allows you to calculate its mean.
One case that needs attention is the question “How many channels did you use to learn about the brand?” If respondents can select only one channel, it is a nominal variable. If they can select multiple channels, each option should become a binary variable, usually coded 0 = not selected and 1 = selected. If the question asks for the number of channels selected, the total count has a different nature.
Common Errors
Assigning Nominal to Every Variable Entered as a Number
The Measure column should not be selected based on whether the data are numeric or text. Select it based on the meaning of the measurement. Codes 1, 2, and 3 for field of study are still nominal, while years of work experience are a numeric quantity with a clearer order and distance.
Calculating the Mean of Group Codes
The Descriptives table can return the mean of GIOITINH, but that number has no valid interpretation when 1 and 2 are only group labels. Use Frequencies or Crosstabs to describe the groups.
Using Cumulative Percent to Infer Order
SPSS still displays a cumulative column when the variable is nominal. It accumulates values according to the entered code order. It does not prove that group 3 comes after or is higher than group 2.
Combining Groups After Looking at the Results
If you combine groups only because the p-value is not significant, the decision may look like a result-driven adjustment. Base it on theory, the research objective, or the analysis plan stated in advance, and keep the original variable.
Confusing a Nominal Variable With a Dummy Variable
A gender variable coded 1 and 2 is still one categorical variable. When you include it in regression, the software may require dummy variables with one reference group. You need to know which group is the reference and interpret the coefficient relative to that group instead of reading the code as a score.
Frequently asked questions
What is a nominal scale in SPSS?
In SPSS, a nominal scale is a categorical variable set to Measure = Nominal in Variable View. Values such as 1, 2, and 3 are usually group codes, while the meaning is stored in the Value Labels section.
Use Analyze > Descriptive Statistics > Frequencies to check the number and percentage in each group. If you want to examine whether two categorical variables are related, use Crosstabs and an appropriate test.
Can you calculate a mean for a nominal scale?
Technically, SPSS can calculate the mean of numerical codes. In terms of interpretation, you should not do this when the codes are only group labels. The mean of gender codes or region codes does not represent a meaningful characteristic.
Is gender a nominal or ordinal scale?
Gender is usually a nominal scale because the groups do not have an order of greater or lesser value. Coding Male = 1 and Female = 2 does not turn the variable into an ordinal scale.
If the questionnaire includes “Other” or “Prefer not to answer”, define separate labels and decide in advance whether to include these groups in the descriptive results or treat them as missing.
Can you run Cronbach's Alpha for a nominal scale?
You do not run Cronbach's Alpha for a single nominal variable such as gender, field of study, or region. Cronbach's Alpha evaluates the internal consistency of a group of items measuring the same construct, as described in (Cronbach, 1951).
If you have a scale made up of several Likert items, check the structure of that scale according to your research objective. A nominal variable can describe the sample or serve as a grouping variable, but it does not become an item of the scale simply because it is stored in the same file.
Can a nominal scale be used in regression?
Yes, but a categorical variable usually needs to be coded as dummy variables or handled in the way required by the software and model. For a variable with multiple groups, you usually need to identify a reference group for comparison.
State the coding method clearly in the methods chapter. A dummy-variable coefficient represents the difference from the reference group, not a one-unit increase on a nominal scale.
Open Variable View first, check Value Labels, Missing, and Measure, then run Frequencies for each categorical variable in your file. If you need to standardize the questionnaire and code the variables before collecting or entering data, you can use this survey form; the analysis and output reading on your .sav or .csv file can be handled in M4 data analysis by DoThesis.