Inference in Linear Regression
A Unified Workflow for Regression Inference
Estimate, standardize, and interpret
For y=beta0+beta1x+error, least squares produces b0 and b1. Residual standard error s estimates the common error standard deviation. A test of H0: beta1=beta10 uses t=(b1-beta10)/SE(b1) with n-2 degrees of freedom; the matching confidence interval is b1 plus or minus t*SE(b1).
For example, if b1=1.8, SE=0.5, and the null slope is zero, t=3.6. Statistical evidence about slope does not by itself establish causality. Intercept inference is mathematically similar but its practical meaning can be weak when x=0 is outside the data range.
Mean-response intervals use uncertainty in the fitted mean and widen away from xbar. Valid t procedures rely on a linear conditional mean, independent errors, roughly constant variance, and adequate normality for the sampling task.
Which residual pattern most directly challenges the constant-variance condition?
Mark Yourself Question
- Grab a piece of paper and try this problem yourself.
- When you're done, check the "I have answered this question" box below.
- View the solution and report whether you got it right or wrong.
A slope estimate is -1.20, SE=0.40, and t*=2.10. Compute the two-sided confidence interval for beta1.