Categorical Variables in Regression
What You'll Learn
You'll discover why categorical variables need special treatment before entering a regression model and what numerical encoding means.
The Problem with Categories
Up until now, you've worked with numerical predictors like age, income, or temperature—values that naturally exist on a scale. But what happens when your predictor is categorical, like color (red, blue, green), country (USA, Canada, Mexico), or customer type (bronze, silver, gold)?
Regression models are mathematical machines that multiply coefficients by predictor values. Try this equation in your head:
Y = β₀ + β₁ × "red" + ε
It doesn't work! You can't multiply a number by the word "red." The algorithm literally cannot compute with text labels or categories.
Why Numerical Encoding is Necessary
Think of it like a recipe that requires measurements in cups. If someone tells you "add some flour," you can't follow the recipe—you need a number. Similarly, regression needs numbers to crunch.
Categorical variables carry meaningful information (being a "premium" customer likely affects purchase amount), but that information is stored in labels, not quantities. We must translate these labels into a numerical format the model can use—this process is called encoding.
The encoding must preserve the categorical nature of the data without implying false numerical relationships. For example, coding colors as red=1, blue=2, green=3 would wrongly suggest that green is "greater than" blue, or that blue is "twice" red.
Key Takeaway: Categorical variables cannot be directly used in regression because mathematical operations require numerical inputs; we must encode categories as numbers in a way that preserves their distinct, non-ordered nature.