Properties of Residuals
What You'll Learn
You'll discover two fundamental mathematical properties that residuals always have in simple linear regression fitted by least squares.
The Two Essential Properties
When you fit a simple linear regression line using the least squares method, the residuals (the vertical distances between actual values and predictions) aren't just random leftovers—they have special mathematical properties baked in.
Property 1: Residuals Sum to Zero
The sum of all residuals equals exactly zero. Mathematically: Σeᵢ = 0
This happens because the least squares line is positioned so that the positive errors (points above the line) perfectly balance out the negative errors (points below the line). It's like a seesaw perfectly balanced at its center.
Why it matters: This confirms your line goes "through the middle" of your data cloud. If residuals summed to something large and positive, your line would systematically underpredict. If large and negative, it would systematically overpredict.
Property 2: Residuals Are Uncorrelated with X
The residuals have zero correlation with your predictor variable X. There's no linear relationship between the X values and how far off your predictions are.
Why it matters: This means your model has extracted all the linear information from X. Whatever pattern the residuals show isn't a simple linear function of X that you missed.
The Big Picture
These aren't assumptions you need to check—they're mathematical guarantees of the least squares method. They happen automatically when you compute β₀ and β₁ using the formulas you've learned. Think of them as "receipts" proving the least squares calculation did its job correctly.
Key Takeaway: In simple linear regression, residuals automatically sum to zero and have zero correlation with X—these are built-in properties of the least squares solution, not assumptions to verify.