What Is a Dummy Variable? How to Create Dummy Variables for Regression
What Is a Dummy Variable
A dummy variable is a numeric representation of a categorical variable that allows you to include categories in a regression model or another quantitative analysis. The most common coding uses 0 and 1. For example, gender can be coded as Male = 0 and Female = 1, or the other way around.
The values 0 and 1 are only codes for group membership. They do not mean that the group coded 1 is larger, better, or higher than the group coded 0. When you read the output, you need to know which group is the reference group, meaning the group coded 0. The dummy coefficient shows the difference between the group coded 1 and the reference group, while the other variables in the model are held constant.
For a categorical variable with two groups, one dummy variable is enough. For a variable with k groups, the usual rule is to create k - 1 dummy variables and keep one group as the reference group. If region has three groups, North, Central, and South, you could create KV_Trung and KV_Nam, with North as the reference group.
You can think of a dummy variable as a way to make categorical information readable to the software. SPSS cannot use text such as “Male”, “Female”, “North”, or “South” directly in a regression equation in the way you need. You therefore need to code the categories before running the model. If you are checking unusual values in your file, read the guide to outliers so that you do not confuse a coding error with an outlier. You can also browse the Statistics topic when you need to compare additional analysis procedures.
The Meaning of Dummy Variables in Quantitative Research
Dummy variables help you answer questions about differences between groups. For example, your study may test whether customers in urban areas have a different level of satisfaction from customers in rural areas. In that case, region enters the model as a dummy variable.
In a simple regression equation:
Y = β0 + β1X1 + β2D + ε
D is the dummy variable, with a value of 0 or 1. Y is the dependent variable, X1 is another independent variable, β0 is the constant, and β2 is the estimated difference between the group where D = 1 and the group where D = 0 when X1 remains unchanged.
If β2 is positive and statistically significant, the group where D = 1 has a higher value of Y than the reference group according to the model. If β2 is negative, the group where D = 1 has a lower value of Y than the reference group. If the p-value does not meet the significance level you selected, you do not have enough evidence to conclude that the two groups differ in the population being studied.
You can also use dummy variables to control for sample characteristics. For example, you could include gender, grouped age, marital status, or region in the model to examine the effect of the main variable after controlling for these differences. Adding a dummy variable does not automatically improve the model. Add it when it has a basis in the research question, the theoretical model, or the research design.
Do not confuse a dummy variable with a moderator. A categorical variable may only be a control variable, but it may also act as a moderator if you test whether it changes the effect of X on Y. In that case, the model needs an interaction variable, such as X × D. Including D by itself is not enough.
How Many Dummy Variables Are Acceptable
A dummy variable does not have an “acceptable” cutoff like Cronbach's Alpha, KMO, or factor loading. A dummy variable is appropriate first when its coding accurately represents the groups in the data, each observation belongs to the correct group according to the design, and the reference group is clearly identified.
The table below helps you check common decisions. The sample-size and testing criteria in the table belong to the relevant analysis, not to the dummy variable itself.
| Check | Decision | Source |
|---|---|---|
| Variable with two groups | Use one dummy variable, coded 0 and 1 | Model coding rule, with no separate cutoff in the source list |
| Variable with k groups | Usually use k - 1 dummy variables and keep one reference group | Model coding rule, with no separate cutoff in the source list |
| Sample size for regression | You may refer to n of 50 + 8m, where m is the number of independent variables | (Tabachnick and Fidell, 2013) |
| Dummy variable in PLS-SEM | Code it consistently before entering the data and check the model structure | (Hair et al., 2022) |
| VIF of variables in the model | VIF below 5 is the level commonly reported | (Hair et al., 2019) |
| Meaning of the dummy coefficient | Check the coefficient sign and p-value or confidence interval according to the analysis method | Result interpretation rule, with no separate dummy cutoff in the source list |
The question “how many dummy variables are acceptable” therefore needs to be separated into two parts. If you are asking about the value of the variable, the answer is usually 0 or 1 according to your coding convention. If you are asking what makes a dummy coefficient good, there is no fixed number. Read the coefficient together with the scale of the dependent variable, the standard error, the p-value, and the practical meaning of the difference.
Do not impose a rule that the dummy coefficient must be greater than 0.5 or must be positive. A negative coefficient can be correct when the group coded 1 has a lower Y than the reference group. Your explanation needs to identify the groups being compared, whether the difference is statistically significant, and whether the difference matters in practice.
How to Read a Dummy Variable in the Output
In SPSS, a dummy variable usually appears in the Coefficients table after you run regression through Analyze > Regression > Linear. The B column is the unstandardized coefficient, Std. Error is the standard error, Beta is the standardized coefficient, t is the test statistic, and Sig. is the p-value.
Suppose you are studying satisfaction, with KV_ThanhThi coded as 1 when the respondent lives in an urban area and 0 when the respondent lives in a rural area. Rural is the reference group. The table below is illustrative output, not the result of a specific study.
| Coefficients | B | Std. Error | Beta | t | Sig. |
|---|---|---|---|---|---|
| (Constant) | 2.184 | 0.241 | 9.063 | 0.000 | |
| Service quality | 0.516 | 0.061 | 0.548 | 8.459 | 0.000 |
| KV_ThanhThi | 0.173 | 0.072 | 0.126 | 2.403 | 0.017 |
In the KV_ThanhThi row, B = 0.173 means that, when service quality is held constant, the predicted satisfaction level for the urban group is 0.173 points higher than for the rural group on the Y scale. Because Sig. = 0.017, if the study uses a 5% significance level, the difference is statistically significant according to the model's decision rule.
If you recode urban = 0 and rural = 1, you will interpret the coefficient in the opposite direction. The underlying group difference does not change by itself, but the coefficient sign and the reference group do change. Record the coding in the methods chapter, the variable description table, or the regression table note.
If a variable has three groups, you will see two dummy rows when the first group is selected as the reference group. For example, KV_MienTrung = 1 for Central and 0 for the other groups, while KV_MienNam = 1 for South and 0 for the other groups. The coefficients in the two rows compare Central and South with North, respectively. Do not read these two coefficients as two independent effects on a continuous variable.
When reading the output, also check the Model Summary, ANOVA, and Coefficients tables. Model Summary reports R Square and Adjusted R Square, the ANOVA table helps assess the overall significance of the model, and Coefficients reports the dummy variable and the individual independent variables. If the dummy is an interaction variable, its name may appear as X_D or as a name you created. You cannot interpret it in the same way as the main dummy variable.
What to Do When a Dummy Variable Is Not Acceptable
Start by checking the coding in the data file. Open Variable View to review Values, Missing, and Measure, then open Data View to check the actual values. A variable expected to contain only 0 and 1 but containing 2, 3, 9, or blank cells can distort the model.
Check the Reference Group
Confirm that the group coded 0 is the group you want to use for comparison. If the reference group has very few observations, the coefficient may be unstable and the standard error may be large. Use Analyze > Descriptive Statistics > Frequencies to check the count and percentage for each group before running the regression.
Check for a Group With a Very Low Frequency
A group with very few observations can make the results difficult to interpret. Before combining groups, you need a clear substantive reason. Do not combine two groups only because you want a smaller p-value. If you combine them, record which groups were combined, why they were combined, and how the new coding works.
Check How You Created k - 1 Dummy Variables
If a variable has k groups and you create all k dummy variables while also keeping the constant, the model may encounter perfect multicollinearity. In SPSS, leave one dummy out as the reference group. If you use Transform > Recode into Different Variables, give each dummy a separate name and create a table matching the old codes to the new codes so you can explain the recoding later.
Reconsider the Variable's Role in the Model
If a dummy variable is not statistically significant, do not remove it immediately. It may still need to remain because it was specified in advance as a control variable. Base the decision on the research question and the methodological reasoning, then report the result honestly. Remove a variable only when there is a clear analytical or theoretical reason, not simply to make the results table look better.
Dummy Variables, Categorical Variables, and Moderators
A categorical variable is the broader concept. It describes data that belong to groups such as gender, occupation, or region. A dummy variable is the numeric coding created from a categorical variable so that the software can use it in a model. A categorical variable with two groups usually needs one dummy variable, while a variable with several groups needs several dummy variables.
A dummy variable is also different from a Likert scale. A Likert scale uses ordered response levels, such as 1 to 5, and is often combined into a measure representing a construct. Read what is a scale if you are unsure about the difference between a single variable, a scale, and a dummy variable.
A dummy variable is even less interchangeable with a mediator. A mediator explains the mechanism through which X affects Y, while a dummy variable only describes a group or categorical condition. A dummy variable can still be used in a mediation or moderation model, but that role depends on the hypothesis and the equation structure.
If you include a dummy variable to test whether an effect differs between two groups, create an interaction variable between the independent variable and the dummy. For example, to test whether the effect of service quality differs between urban and rural areas, the model needs Service quality, KV_ThanhThi, and Service quality × KV_ThanhThi. Looking only at the KV_ThanhThi coefficient does not answer the question about whether the effect differs.
Common Errors
The first error is coding 1 and 2 instead of 0 and 1 and then interpreting the variable as a dummy variable. Some software will still run the model, but the coefficient will no longer represent the difference between the two groups directly in the easiest way to read. Recode the variable to 0 and 1, and record the reference group.
The second error is creating a dummy variable for every group while keeping the constant. This is the dummy variable trap. When the dummy variables always add up to 1, they have a perfect linear relationship. The usual solution is to leave out one dummy variable or to run a model without a constant when there is an appropriate methodological reason.
The third error is calling a dummy variable quantitative and comparing the means of codes 0 and 1 as if the distance between the codes had measurement meaning. Codes 0 and 1 only represent groups. The meaning of the difference comes from the group comparison in the model, not from the arithmetic distance between the two codes.
The fourth error is changing the reference group after several runs without updating the results table. The coefficient sign may then be misread. Keep a record of the variable name, coding, reference group, number of observations in each group, and the version of the file used for the analysis.
Finally, do not remove a dummy variable only because Sig. is greater than 0.05. Distinguish between a variable with no evidence of an effect, a variable that needs to remain as a control, and a variable assigned the wrong role. If you need to present several models, report the order in which the variables were entered and the reason in the results chapter.
Frequently asked questions
What is a dummy variable in SPSS?
A dummy variable in SPSS is a variable coded with values such as 0 and 1 so that you can include a categorical variable in a regression or quantitative model. For example, gender can be coded as Male = 0 and Female = 1.
When you run the regression, SPSS reads the dummy coefficient as the difference between the group coded 1 and the group coded 0, while the other variables are held constant.
Should a dummy variable be coded 0 and 1 or 1 and 2?
You should code a dummy variable as 0 and 1. The value 0 usually represents the reference group, while 1 represents the group being compared. This makes the coefficient in the Coefficients table easier to interpret.
If the original data use 1 and 2, create a new variable through Transform > Recode into Different Variables instead of overwriting the original data immediately.
How many dummy variables should you create for a variable with three groups?
A variable with three groups usually requires two dummy variables, with one group kept as the reference group. For example, for North, Central, and South, you could create KV_Trung and KV_Nam, while North is coded 0 on both variables.
Each coefficient compares one group with North. Record the reference group in the results table so that readers do not misinterpret the coefficients.
Should you delete a dummy variable when it is not statistically significant?
Do not delete it immediately. First check the coding, group frequencies, reference group, and role of the dummy in the model. If it is a control variable specified in advance, you can keep it and report that its effect is not statistically significant.
Deleting a variable only because you want the p-value to meet a cutoff can reduce the transparency of the study. Discuss the criteria for retaining or removing the variable with your supervisor before rerunning the model.
Is a dummy variable a moderator?
A dummy variable is not automatically a moderator. It is only a coding method for a categorical variable. It becomes part of a moderation analysis when you create and test an interaction between the dummy variable and the independent variable.
If you are studying a moderating variable, distinguish between the dummy coefficient, the independent-variable coefficient, and the interaction coefficient. You can read more about moderating to follow this structure correctly.
Open the .sav file again, check the frequency table for the categorical variable, record the reference group, and create k - 1 dummy variables before rerunning Analyze > Regression > Linear. If you need to run the full analysis on your actual data and check the interpretation, see DoThesis M4 analysis.