Course contentsShow
← Data Science
Lesson 605 of 2,14516. Multiple Linear RegressionFree lesson

Units and Scaling of Coefficients

How coefficient magnitude depends on predictor units and why standardization can aid interpretation.

Units and Scaling of Coefficients

You'll learn: How the units of your predictors directly control the magnitude of your coefficients, and why this matters for interpretation.

Why Coefficient Size Depends on Units

In multiple linear regression, each coefficient tells you how much Y changes per one-unit increase in its predictor. The catch? That "one unit" depends entirely on how you measured the variable.

Imagine predicting house price from size. If you measure size in square feet, you might get a coefficient of 150(priceincreases150 (price increases 150 per square foot). But if you measure the same houses in square meters, your coefficient jumps to roughly $1,615 — because one square meter is about 10.76 square feet, so the "per unit" impact is proportionally larger.

Same relationship, dramatically different numbers.

The Problem with Comparing Raw Coefficients

You cannot compare unstandardized coefficients across predictors with different units:

  • Age (in years): β = 500
  • Income (in dollars): β = 0.02

Which matters more? You can't tell! Age's coefficient looks bigger only because "one year" is a different-sized jump than "one dollar."

Why Standardization Helps

When you standardize predictors (convert to z-scores with mean 0 and standard deviation 1), all variables share the same scale. Now coefficients represent the change in Y per one standard deviation increase in X — a fair, apples-to-apples comparison.

This doesn't change predictions, but it makes interpretation clearer when asking "which predictor has more impact?"

Key Takeaway: Coefficient magnitude reflects predictor units, not just importance. Standardization puts coefficients on a common scale, enabling meaningful comparison across predictors measured in different units.