Course contentsShow
← Mathematics
Lesson 3068 of 3,12160. Statistics — ModelingFree lesson

Assumptions of Simple Linear Regression

Identify the key assumptions: linearity, independence, normality, and equal variance of errors.

Assumptions of Simple Linear Regression

You'll learn: The four key assumptions that must hold for simple linear regression to produce valid, reliable results.

The Foundation of Trustworthy Models

When you fit a regression line using the least squares method, you're relying on certain conditions being true about your data. Violating these assumptions doesn't stop you from calculating the line, but it means your predictions, confidence intervals, and hypothesis tests may be unreliable or misleading.

The Four Core Assumptions

1. Linearity

The true relationship between your predictor variable and response variable must be linear. If the actual relationship is curved or more complex, forcing a straight line through it produces poor predictions.

2. Independence

Each observation's error (the distance from the regression line) must be independent of all others. Think of flipping coins: one flip doesn't influence the next. If errors are correlated—like in time series data where today's error affects tomorrow's—standard regression formulas break down.

3. Normality

The errors (residuals) should follow a normal distribution centered at zero. While regression is somewhat robust to this assumption, extreme violations affect hypothesis tests and confidence intervals. This doesn't mean your data must be normal—just the errors around the line.

4. Equal Variance (Homoscedasticity)

The spread of errors should remain constant across all values of the predictor. Imagine a horizontal "band" of equal width around your regression line. If the spread fans out (getting wider as x increases), you have heteroscedasticity, which distorts standard errors and makes inference unreliable.

Why These Matter

These assumptions ensure that your standard errors, p-values, and confidence intervals are valid. Violating them doesn't necessarily invalidate the regression line itself, but it means you can't trust the uncertainty measures around it.

Key Takeaway: Simple linear regression rests on four pillars—linearity, independence, normality of errors, and equal variance—which together guarantee reliable inference and prediction.