One-Sided Confidence Bounds
What you'll learn: How to construct confidence intervals that bound a parameter from only one direction when you care about "at least" or "at most" questions.
The Core Idea
So far, you've learned about two-sided confidence intervals that give both lower and upper bounds (e.g., "we're 95% confident the mean is between 10 and 15"). But sometimes you only care about one direction:
- "Is the battery life at least 8 hours?" (lower bound only)
- "Is the error rate at most 5%?" (upper bound only)
- "Can the bridge support at least 50 tons?" (lower bound only)
When only one direction matters for your inference, you can construct a one-sided confidence bound that provides either an upper or lower limit—not both.
Why Use One-Sided Bounds?
Think of it like setting safety limits. If you're testing whether a new drug lowers blood pressure, you might only care that it doesn't raise it above a certain threshold (upper bound). The lower it goes doesn't concern you for safety purposes.
One-sided bounds use the same confidence level (like 95%) but allocate all the uncertainty to one tail instead of splitting it between two tails. This makes the bound slightly tighter in the direction you care about.
Construction
Instead of using critical values that split α (like 0.025 in each tail for 95%), you place the entire α (0.05) in one tail:
-
Lower bound only: Mean - t*(SE)
(confident the parameter is at least this value) -
Upper bound only: Mean + t*(SE)
(confident the parameter is at most this value)
Key Takeaway: One-sided confidence bounds provide directional certainty when you only need to establish a minimum or maximum value, making your inference more focused and slightly more precise in that direction.