Confidence Interval for Mean Value of Y
Inference for the Mean Response at a Chosen Predictor
The estimand controls the standard error
In simple linear regression, the estimated mean response at x0 is yhat0=b0+b1x0. Its standard error is s times the square root of [1/n + (x0-xbar)^2/Sxx]. A two-sided confidence interval is yhat0 plus or minus t* times that standard error, using n-2 residual degrees of freedom.
For example, with yhat0=42, s=5, n=25, x0=xbar, and t*=2.069, the standard error is 1 and the interval is 42 plus or minus 2.069. Away from xbar, leverage increases uncertainty.
This interval concerns an average response, not a single future Y; the latter includes an additional 1 under the square root.
The usual regression conditions include linear mean structure, independent errors, roughly constant variance, and approximate normality for small-sample t inference. The interval does not establish causality.
Where is the mean-response confidence interval generally narrowest, all else fixed?
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.
At x0=xbar, n=36, residual standard error s=12, fitted mean 80, and t*=2.03. Compute the 95% confidence interval for the mean response.