Hypothesis Testing for Individual Coefficients
You'll learn how to test whether each predictor in a multiple regression model actually matters.
The Core Idea
After fitting a multiple regression model, you naturally wonder: "Does this predictor really help predict my response variable, or is its coefficient just random noise?" The answer comes from hypothesis testing.
For each coefficient βᵢ (where i might be 1, 2, 3, etc. for different predictors), we set up:
- H₀: βᵢ = 0 (the null hypothesis: this predictor has no real effect)
- Hₐ: βᵢ ≠ 0 (the alternative: this predictor matters)
If βᵢ = 0 were true, that predictor contributes nothing to explaining your response variable—it might as well not be in the model.
The t-Test Machinery
We use a t-test for each coefficient. The test statistic is:
t = (estimated βᵢ - 0) / SE(βᵢ)
where SE(βᵢ) is the standard error of the coefficient estimate (measuring how much sampling variability we'd expect).
This t-statistic follows a t-distribution with (n - p - 1) degrees of freedom, where n is sample size and p is the number of predictors.
Interpreting the p-value
The p-value tells you: "If this predictor truly had zero effect, what's the probability I'd see a coefficient estimate this extreme just by chance?"
- Small p-value (typically < 0.05): Strong evidence against H₀; the predictor appears significant
- Large p-value: Insufficient evidence; you can't rule out that the coefficient is just noise
Key Takeaway: Testing H₀: βᵢ = 0 with t-tests and examining p-values lets you identify which predictors in your multiple regression model are statistically significant contributors to explaining the response.