Realizing how to include **dummy variables** into a **regression** is the best way to end your introduction into the world of **linear regressions**. Another useful concept you can learn is the **Ordinary Least Squares**. But now, onto **dummy variables**.Ā Apart from the offensive use of the word ādummyā, there is another meaning ā an imitation or a copy that stands as a substitute.

## What Is a Dummy Variable?

In **regression analysis**, a **dummy** is a variable that is used to include categorical data into a **regression model**. In previous tutorials, we have only used numerical data. We did that when we first introduced **linear regressions** and again when we were exploring the **adjusted R-squared**. However, representing *numbers* on a scale makes more sense than representing *categories* like gender or season. Itās time to find out how to include such variables into a **regression** we are working with.

## How to Include Categorical Data Into a Regression

Firstly, make sure that you check the article where we made our first steps into the world of **linear regressions.**Ā We will be using the SAT-GPA example from there. If you donāt have time to read it, here is a brief explanation: *Based on the SAT score of a student, we can predict his GPA.* Now, we can improve our prediction by adding another **regressor** ā *attendance*.

In the picture below, you can see a dataset that includes a variable that measures if a student attended more than 75% of their university lectures.

Keep in mind that this is *categorical* data, so we cannot simply put it in the **regression**.

We will start off by going through the process of using a **dummy** and explain it later.

**Using a Dummy Variable**

The time has come to write some code. We can begin by importing the relevant libraries by writing:

__import numpy as np__

__import pandas as pd__

__import statsmodels.api as sm__

__import matplotlib.pyplot as plt__

__import seaborn as sns__

__sns.set()__

After that, letās load the file ā1.03. Dummies.csvā into the variable **raw_data**. You can download the file from here. If you donāt know how to load it, hereās what you need to type:

__raw_data = pd.read_csv(ā1.03. Dummies.csvā)__

Now, letās simply write

__raw_data __

and see what happens.

As you can tell from the picture above, there is a third column named āAttendanceā. It reflects if a student attended more than 75% of the lessons with two possibilities ā *Yes* and *No*.

**Mapping Values**

What we would usually do in such cases is to map the *Yes/No* values with ** 1s** and

**. In this way, if the student attended more than 75% of the lessons, the**

*0s***dummy**will be equal to

**. Otherwise, it will be a**

*1***.**

*0*So, we will have transformed our *yes/no* question into ** 0s** and

**. Thatās what the**

*1s***dummy**name stands for ā we are imitating the categories with numbers.

**How to Do it**

In *pandas*, thatās done quite intuitively.

Letās create a new variable **data** equal to **raw_data**. This is what we need to run:

__data = raw_data.copy()__

Then, we have to overwrite the series āattendanceā in the data frame. So, this is how the code should look like:

__data[āAttendanceā] = data[āAttendanceā].map({āYesā:1, āNoā : 0})__

This is the proper syntax to map *Yes* to 1 and *No* to 0.

We can write

__data __

and find out if we have done our job.

As you can see in the picture above, we have successfully created a **dummy variable**! The categorical data in the series was replaced or *mapped* to numerical.

**The Descriptive Statistics**

Letās take a look at the descriptive statistics of the variables. We can do that by writing:

__data.describe()__

The **mean** of āAttended more than 75%ā is 0.46, as shown below.

The fact that the **mean** is less than 0.5 gives us the information that there are more ** 0s** than

**. Since the two outcomes are**

*1s***and**

*0***, this implies that 46% of the students have attended more than 75% of the lessons.**

*1*In any case, now we can create a **regression** that explains GPA takingĀ both *SAT scores* and *attendanceĀ *into consideration.

**Creating the Regression**

We can load GPA in the variable **y**, and SAT, and āAttendanceā in the variable **x1**. This is the code we need to run:

__y = data[āGPAā]__

__x1 = data[[āSATā, āAttendanceā]]__

We must use the *statsmodels* method for adding a constant. Then we can fit the **regression** and get the summary as before.

__x = sm.add_constant(x1)__

__results = sm.OLS(y,x).fit()__

__results.summary()__

** The Results**

As you can see in the picture below, our overall model is significant,

the SAT score is significant, and the **dummy variable** is significant.

The **adjusted R-squared** of this model is 0.555, which is a great improvement from what we would get without attendance.

A model without the **dummy variable** would be:

GPA = 0.275 + 0.0017 * the SAT score of a student.

The model, including the **dummy variable** is:

GPA = 0.6439 + 0.0014 * the SAT score of a student + 0.2226 * the **dummy variable**.

**Explaining the Equation**

Now, we said that the **dummy** is ** 0** or

**, so actually we can represent this equation with two others.**

*1*If the student did not attend, the **dummy** would be ** 0**. So, 0.2226 * 0 is 0. The model becomes GPA = 0.6439 + 0.0014 * SAT.

If the student attended, the **dummy** **variable** would be ** 1**, so the model becomes:

GPA = 0.6439 + 0.0014 * SAT + 0.2226.

Letās add the intercept and the **dummy** together.

As you can see in the picture above, we got GPA = 0.8665 + 0.0014 * SAT.

**Plotting the Data**

There will be two equations, which we can call **yhat_no**, and **yhat_yes**. They will represent the two equations we just talked about. Certainly, we can parametrize these equations, but there is no need for such a simple example.

**The 2 Equations**

So, what we observe above are two equations that have the same slope but a different intercept. The students who attended are spread around the upper line.

On average, their GPA is 0.2226 higher than the GPA of students who did not attend.

We can even think about these as two separate **regressions**. We can color the points, which refer to students who attended classes, so the red line, and students who did not attend ā the green line.

You can clearly see the difference now.

Finally, we will put the original **regression line** on the graph.

As you can see, it is steeper and goes somewhat between the two lines of the **dummies**.

To use this model for prediction purposes, we need two pieces of information: an SAT score and whether a person attended more than 75% of their lectures.

**The Process of Adding a Dummy Variable into a Regression**

So, why do we even need **dummy variables**? We can simply use numerical data all the time and we wonāt have any problems, right?

Well, not really. Whenever we have to represent categorical data, such as brands, a **dummy **is what we need.

- We map the categories.
- Then we drop one of them, which serves as the baseline model
- We just deal with it.

As you can see, the process of creating a **regression** even with categorical data is not brain surgery.

***

**Interested in learning more? You can take your skills from good to great with our statistics course!**

**Try statistics course for free**

**Next tutorial: Correlation vs Regression**