Which statement correctly interprets the 90% level of an interval for the mean daily water use of a city's households?
Understanding confidence levels
What "95% confident" means, why a computed interval has no 95% probability, how level and sample size move width and reliability, and the interval–test connection.
Key ideas
- Probability versus confidence
Probability describes a future random event. Confidence describes our certainty about something already fixed, like whether this interval captured .
- Level
The expected long-run capture rate. In 100 intervals at 95% we expect about 95 hits, not exactly 95.
- Level and width
Higher level → larger multiplier → wider interval. More reliable, less precise.
- Sample size
Larger → smaller SE → narrower interval. Reliability (the level) does not change; this surprised many students in lecture.
- Interval–test connection
For a two-sided test: null value inside a level-C interval ⇔ p-value greater than ; outside ⇔ p-value less than . An interval is the set of null values the data do not reject at that level.
- Reading simulation plots
To rank levels, count misses (fewest misses = highest level). To rank sample sizes, compare widths among plots at the same level (narrowest = largest ).
Connection rules
Tutoring and quiz gains
A random sample of 25 students records their quiz score gain after tutoring. Use the output to say what a two-sided test of and of would show, and what a 99% interval would look like.
t.test(gain, conf.level = 0.95)
One Sample t-test
data: gain
t = 4.625, df = 24, p-value = 0.0001078
alternative hypothesis: true mean is not equal to 0
95 percent confidence interval:
2.048881 5.351119
sample estimates:
mean of x
3.7 $\mu_0 = 0$: 0 is outside (2.05, 5.35), so the two-sided p-value is below 0.05 (the output confirms 0.0001).
$\mu_0 = 5$: 5 is inside the interval, so the two-sided p-value is above 0.05 (it is about 0.117): not enough evidence the mean gain differs from 5.
99% interval: same center 3.7, larger multiplier, so wider: about (1.46, 5.94). Still excludes 0, so p < 0.01 for .
Interpretation: we are 95% confident the true mean quiz gain for all students like these is between 2.05 and 5.35 points.
0 is implausible at both levels; 5 remains plausible. Raising the level widened the interval around the same center.
Check yourself
A 90% interval for the proportion of a college's students who live off campus is (0.31, 0.39). A two-sided test of with the same data would give:
A two-sided test of gives p = 0.03. Which is true about intervals from the same data?
A simulation draws 100 intervals for the same parameter, all with the same sample size; 11 of them miss the parameter. Which confidence level was most likely used?
Quantity A: the reliability (long-run capture rate) of 95% intervals from samples of 50. Quantity B: the reliability of 95% intervals from samples of 500.
A 95% interval for a proportion is . Using for 95% and for 90%, about what is the margin of error of the 90% interval from the same sample?