DummyVariable

A dummy variable, also known as an indicator variable, is a numerical variable that represents the presence or absence of a qualitative attribute. It takes values 0 or 1 and is used to include categorical data in statistical models and machine learning algorithms.

In data preparation, a categorical variable with k categories is commonly represented by k-1 dummy variables through one-hot encoding, with one category serving as the reference. This approach avoids perfect multicollinearity between the dummy variables and the model intercept. Some modeling strategies may use all k dummies with a constraint, but the standard practice for linear models is to omit the reference category.

In regression and similar models, the coefficient on a dummy variable expresses the expected difference in the dependent variable between the category indicated by 1 and the reference category indicated by 0, holding other variables constant. If multiple dummy variables exist for a categorical feature, each coefficient compares its category to the reference.

Example: for a categorical variable Color with Red, Green, and Blue, two dummies might be Color_Green and Color_Blue. If Red is the reference, the coefficient on Color_Green represents the effect of being Green relative to Red.

Dummy variables can be used in interaction terms to explore whether the effect of a predictor varies by category. They are central to many statistical analyses, but care is needed to manage high-cardinality categories and to avoid misinterpretation of coefficients.