Statistical Significance of Individual Coefficients
What You'll Learn
You'll learn how to use t-tests and p-values to determine whether each predictor in your multiple regression model truly has a meaningful relationship with the outcome—or if its coefficient could just be random noise.
The Core Idea
In multiple linear regression, you estimate a coefficient for each predictor. But just because you get a number doesn't mean that predictor actually matters. The coefficient might be non-zero purely by chance in your sample data.
Statistical significance testing answers the question: "Is this predictor's coefficient reliably different from zero, or could I have gotten this result just from random variation?"
The t-Test for Individual Coefficients
For each coefficient, we perform a t-test with these hypotheses:
- Null hypothesis (H₀): The true coefficient equals zero (the predictor has no effect)
- Alternative hypothesis (H₁): The true coefficient does not equal zero (the predictor matters)
The t-statistic measures how many standard errors the estimated coefficient is away from zero:
t = (estimated coefficient) / (standard error of coefficient)
The p-Value: Your Decision Guide
The p-value tells you the probability of observing a coefficient this extreme (or more) if the null hypothesis were actually true.
- Small p-value (typically < 0.05): Strong evidence against the null hypothesis → the predictor is statistically significant
- Large p-value (≥ 0.05): Insufficient evidence → cannot conclude the predictor matters
Think of it like a courtroom: a small p-value means "beyond reasonable doubt" that the predictor has a real effect.
When to Use This
After fitting a multiple regression, check the p-value for each coefficient. Predictors with high p-values might be candidates for removal, though remember: statistical significance isn't the same as practical importance!
Key Takeaway: Use t-tests and p-values to determine which predictors in your model have statistically significant relationships with the outcome, helping you distinguish real effects from random noise.