Confidence and Prediction Intervals in Multiple Regression
You'll learn how to construct intervals that quantify uncertainty for both average responses and individual predictions in multiple regression.
In simple linear regression, you learned about inference for predictions. Now in multiple regression, we distinguish between two types of intervals based on what we're trying to estimate:
Confidence Interval for Mean Response
When you want to estimate the average outcome for all observations with specific predictor values, you construct a confidence interval for the mean response. This answers: "What is the average Y for everyone with these X values?"
For example, if predicting house prices based on square footage and number of bedrooms, a confidence interval for mean response tells you the average price of all houses with 2000 sq ft and 3 bedrooms.
This interval is narrower because averaging reduces variability.
Prediction Interval for Individual Observations
When you want to predict a single new observation, you need a prediction interval. This answers: "What will Y be for this one specific case with these X values?"
Using the same example, a prediction interval estimates the price of one particular house with 2000 sq ft and 3 bedrooms.
This interval is wider because individual observations vary more than averages. It accounts for both:
- Uncertainty in estimating the mean response
- Natural variability of individual observations around that mean
The Key Difference
Think of shooting arrows at a target. A confidence interval estimates where the bullseye is (the true mean). A prediction interval estimates where your next arrow will land (an individual value that scatters around the bullseye).
Key Takeaway: Confidence intervals estimate mean responses and are narrower; prediction intervals estimate individual observations and are wider because they account for additional individual-level variability.