A polling firm reports: "52% of likely voters support the ballot measure."
That sounds precise. But the firm didn't ask every voter — they asked a random sample of 1,000 people. If a different polling firm drew a different random sample of 1,000 voters the same day, would they also get exactly 52.0%? Almost certainly not. They might get 50%, or 54%, or 51.3%.
So which number is the "real" level of support in the whole population? Here's the uncomfortable truth: we will never know it exactly without surveying everyone. Every sample gives a slightly different answer because of random sampling variability — the idea you studied in Lessons 15–18.
That's the problem confidence intervals solve. Instead of pretending one sample gives the answer, we report a range of plausible values and attach a level of confidence to our method. Today is all about the big idea — what that range means, and (just as important) what it does NOT mean. The wording here is the single most-tested and most-misunderstood idea on the entire AP exam. Let's nail it.
When you compute a statistic like a sample proportion p̂ = 0.52, you have a point estimate — a single number that is your single best guess for the unknown population parameter (here, the true proportion p).
The trouble with a point estimate is that it is almost certainly wrong — not wildly wrong, but off by some amount, because of sampling variability. A point estimate carries no information about how wrong it might be. Is 0.52 nailing the truth, or could the truth easily be 0.46 or 0.58?
An interval estimate fixes this. Instead of a single number, we report a whole range of plausible values for the parameter, built around our point estimate:
estimate ± margin of error
The point estimate sits in the middle. The margin of error is the "give or take" — the amount we extend in each direction to account for sampling variability. A poll reporting "52% with a margin of error of 3 percentage points" is really reporting the interval from 49% to 55%.
Every confidence interval in this course is built the same way:
margin of error = (critical value) × (standard error)
Two ingredients:
z for the procedures in this unit) controls how confident* we want to be. It says how many standard errors wide to make the interval. The more confident we want to be that our net catches the parameter, the wider we cast it, so a higher confidence level uses a larger critical value.For the proportion intervals coming in Lesson 20, the common z* critical values are worth memorizing:
| Confidence level | Critical value z* |
|---|---|
| 90% | 1.645 |
| 95% | 1.960 |
| 99% | 2.576 |
(You'll learn exactly where these come from and how to compute the standard error for a proportion in Lesson 20. Today we stay conceptual.)
Here is the heart of the lesson. Imagine we could repeat our study over and over: draw a random sample, build a 95% confidence interval, then draw a new random sample, build another interval, and so on — thousands of times. Each sample gives a different point estimate, so each interval lands in a slightly different place. The true parameter, meanwhile, sits at one fixed (unknown) value the whole time.
[GRAPH: A vertical dashed line is drawn at the true parameter value p = 0.50, labeled "true proportion p (unknown in real life)." To the right, 25 horizontal confidence intervals are stacked vertically, one per simulated sample, each a short horizontal segment with a dot at its center (the sample's point estimate). Most segments — about 24 of the 25 — cross the dashed line, drawn in blue and labeled "captures p." One or two segments fall entirely to the left or right of the dashed line, drawn in red and labeled "misses p." Caption: "Each random sample produces a different interval. The method captures the true p in about 95% of all possible samples. We never know whether OUR one interval is a blue one or a red one."]
Notice what the picture shows: about 95% of the intervals (the blue ones) capture the true parameter, and about 5% (the red ones) miss it entirely. The "95%" is a property of the method, measured across all the samples we could have taken — not a property of any single interval.
So the correct interpretation of the confidence level is:
Confidence level interpretation: "If we took many random samples and constructed a confidence interval from each one using this method, about 95% of those intervals would capture the true [parameter in context]."
And when we report one specific interval — say a poll's interval of 49% to 55% — the correct interpretation of that interval is:
Single-interval interpretation: "We are 95% confident that the true [parameter in context — e.g., the true proportion of all voters who support the measure] is between 49% and 55%."
That's it. Memorize the shape of both sentences. On the AP exam you will write the single-interval sentence constantly, always filling in the parameter in context (who? what population? what quantity?) and the two endpoints. A sentence that says "we are 95% confident the value is between 49% and 55%" with no mention of what value or which population loses points — context is everything in Practice 4.
This is where students lose the most points. Each of the following is WRONG:
When in doubt, anchor every interpretation to one idea: a confidence interval is a set of plausible values for the unknown parameter, produced by a method that succeeds about 95% of the time.
Because margin of error = critical value × standard error, anything that changes those two pieces changes the width of the interval.
This is the confidence vs. precision trade-off. A super-wide interval ("between 0% and 100%") would be 100% confident but useless — it tells you nothing. A razor-thin interval is wonderfully precise but might rarely capture the truth. Statisticians usually settle on 95% as a reasonable balance, and when they need both high confidence and a tight interval, the only honest fix is to collect a larger sample.
Problem. A nutrition researcher takes a random sample of adults and is 90% confident that the true proportion of U.S. adults who eat breakfast daily is between 0.58 and 0.66. Interpret this interval in context.
Strategy. Use the single-interval template: "We are [C]% confident that the true [parameter in context] is between [low] and [high]."
Solution / Interpretation.
We are 90% confident that the true proportion of all U.S. adults who eat breakfast daily is between 0.58 and 0.66.
Note the three required pieces: the confidence level (90%), the parameter in full context (the true proportion of all U.S. adults who eat breakfast daily — not "of the sample"), and both endpoints.
Problem. For the same study, explain what "90% confident" means.
Strategy. The level is about the method across many samples, not about this one interval. Use the confidence-level template.
Solution / Interpretation.
If the researcher took many random samples of the same size and built a 90% confidence interval from each, about 90% of those intervals would capture the true proportion of U.S. adults who eat breakfast daily.
Common trap avoided. It is tempting to write "there's a 90% chance the true proportion is between 0.58 and 0.66." That's wrong — the true proportion is fixed and this interval either contains it or not. The 90% describes the method's long-run capture rate.
Problem. The researcher recomputes the interval at 99% confidence instead of 90%, using the same data. Will the new interval be wider or narrower? Why?
Strategy. Margin of error = critical value × standard error. Same data ⇒ same standard error. Only the critical value changes.
Solution / Interpretation. The 99% critical value (z = 2.576) is larger than the 90% one (z = 1.645), so the margin of error grows and the interval gets wider. To be more confident of capturing the parameter, we must report a larger range of plausible values. We gained confidence but lost precision.
Problem. A campaign manager wants a 95% confidence interval for the proportion of voters supporting her candidate. Her first poll of 400 voters gives a margin of error of about 5 percentage points, which she considers too wide. A staffer suggests two fixes: (i) keep the sample at 400 but drop to 90% confidence, or (ii) keep 95% confidence but increase the sample to 1,600 voters. For each fix, explain the effect on the margin of error, and state which fix has a downside.
Strategy. Tie each change back to margin of error = critical value × standard error. Confidence level moves the critical value; sample size moves the standard error.
Solution / Interpretation.
Interpretation. Both fixes narrow the interval, but only increasing the sample size keeps the confidence level intact. When you need both high confidence and high precision, the honest answer is almost always "get more data."
Mistake 1: "There's a 95% probability the parameter is in this interval."
Why it's wrong: After the data are collected, the parameter is fixed and the interval is fixed — the true value is either inside or outside, with no probability left to assign. Fix: The 95% is the long-run success rate of the method across many samples, not the probability for one finished interval. Say "we are 95% confident," not "there's a 95% probability."
Mistake 2: Confusing the interval with the data.
Why it's wrong: Writing "95% of the data fall in this interval" treats the interval as describing individual observations. A confidence interval is about the parameter, never the spread of the raw data. Fix: Anchor every interpretation to the unknown parameter.
Mistake 3: Leaving out context.
Why it's wrong: "We are 95% confident the value is between 0.49 and 0.55" gives no parameter and no population, so it earns no credit on the FRQ. Fix: Always name who (the population) and what (the quantity) — e.g., "the true proportion of all registered voters who support the measure."
Mistake 4: Saying "95% of samples fall in this interval."
Why it's wrong: Each sample builds its own interval; other samples' statistics don't have to land inside yours. Fix: The method captures the parameter in about 95% of samples — that's a statement about intervals capturing p, not about samples landing in one interval.
Mistake 5: Getting the trade-off backwards.
Why it's wrong: Students sometimes think higher confidence makes a narrower interval, or that a bigger sample makes it wider. Fix: Higher confidence → bigger critical value → wider. Larger sample → smaller standard error → narrower.
12. (Short answer) A wildlife biologist takes a random sample of trout from a lake and is 95% confident that the true proportion of trout in the lake carrying a certain genetic marker is between 0.12 and 0.18. (a) Interpret this interval in context. (b) Interpret what "95% confident" means in context.
13. (Short answer) A student writes: "We are 95% confident that 95% of the trout fall between 0.12 and 0.18." Identify two things wrong with this statement.
14. (Short answer) A survey of 800 people gives a 95% confidence interval with a margin of error of 3.5 percentage points. The team wants to cut the margin of error roughly in half without lowering the confidence level. Describe what they should do and explain why it works.
15. (Short answer) Explain the confidence-vs-precision trade-off in one or two sentences, and state the only way to improve both at once.
FRQ — Statistical Practice 4: Interpret Results
A consumer-advocacy group wants to estimate the proportion of all online shoppers who abandon their shopping cart before completing a purchase. The group selects a random sample of 600 online shopping sessions and constructs a 95% confidence interval for the true proportion of sessions that end in cart abandonment. The resulting interval is (0.66, 0.74).
(a) Identify the point estimate and the margin of error for this interval. (2 points)
(b) Interpret the confidence interval (0.66, 0.74) in the context of this study. (2 points)
(c) Interpret what it means to be "95% confident" in the context of this study. (2 points)
(d) A team member states: "There is a 95% probability that the true proportion of cart-abandoning sessions is between 0.66 and 0.74." Explain why this interpretation is incorrect. (2 points)
(e) The group decides the interval is too wide and wants a narrower one. Describe one change that would produce a narrower interval without reducing the confidence level, and explain why it works. (2 points)
(a) The point estimate is the center of the interval:
point estimate = (0.66 + 0.74) / 2 = 0.70.
The margin of error is the distance from the center to either endpoint:
margin of error = (0.74 − 0.66) / 2 = 0.04.
So the point estimate is 0.70 and the margin of error is 0.04.
(b) We are 95% confident that the true proportion of all online shopping sessions that end in cart abandonment is between 0.66 and 0.74.
(c) If the group took many random samples of 600 online shopping sessions and constructed a 95% confidence interval from each sample, about 95% of those intervals would capture the true proportion of all sessions that end in cart abandonment.
(d) Once the sample is taken and the interval (0.66, 0.74) is computed, the true proportion is a fixed (though unknown) number, and the interval is a fixed range. The true proportion is therefore either in this particular interval or it is not — there is no longer any randomness to which a probability of 0.95 could apply. The "95%" describes the long-run capture rate of the method across all possible samples, not the probability that this one specific interval contains the parameter.
(e) Increase the sample size (for example, from 600 to a larger number of sessions). A larger sample reduces the standard error, which reduces the margin of error (= critical value × standard error) while the critical value stays the same at the 95% level. This narrows the interval without lowering the confidence level. (Equivalently valid only if it kept confidence fixed — lowering the confidence level would also narrow it but is disallowed here.)
Part (a) — 2 points
Part (b) — 2 points
Part (c) — 2 points
Part (d) — 2 points
Part (e) — 2 points
Where students lose points:
1. B. A point estimate is a single best-guess number for a parameter.
- A describes an interval estimate. C names a different quantity (the give-or-take). D is wrong because the point estimate is almost never exactly the true value — that's why we need intervals.
2. B. Confidence intervals are built as estimate ± margin of error.
- A is wrong because we don't know the parameter (that's what we're estimating). C omits the critical value (margin of error is more than just the standard error). D combines the wrong two pieces.
3. C. Margin of error = critical value × standard error.
- A inverts the operation. B adds instead of multiplies. D uses the point estimate, which is not part of the margin-of-error formula.
4. B. z* = 1.960 for 95% confidence.
- A (1.645) is the 90% value; C (2.576) is the 99% value; D is the confidence level itself, not a critical value.
5. B. Higher confidence → larger critical value → larger margin of error → wider interval.
- A and C reverse or ignore the effect; D is wrong because the direction is fully determined.
6. B. Larger n → smaller standard error → smaller margin of error.
- A reverses it; C ignores the effect; D is wrong because the direction does not depend on the parameter value.
7. C. Correct single-interval interpretation: 95% confident the true proportion is between the endpoints.
- A misreads the interval as a statement about individual students' behavior (data, not parameter). B is the probability trap — no probability applies to a finished interval. D wrongly claims other samples' statistics fall inside this interval.
8. B. Correct confidence-level interpretation: across many samples, about 95% of the resulting intervals capture the parameter.
- A confuses the interval with the data. C claims certainty about the sample proportion equaling the parameter, which never happens. D nonsensically treats the parameter as something that moves.
9. B. The flaw is that a finished interval either contains the fixed parameter or it does not, so no probability applies to this interval.
- A, C, and D are irrelevant to the interpretation error (width, choice of level, and "no probability in statistics" are all beside the point).
10. B. Pollster B's larger sample (n = 2,000) yields a smaller standard error and thus a narrower interval.
- A reverses the size effect; C ignores it; D is wrong because both are unbiased and centered near the same true proportion (the centers differ only by sampling variability, not systematically).
11. B. At the same data, only the critical value changes; 99% uses a larger z* (2.576 vs. 1.645), widening the interval.
- A is wrong because the standard error depends on the data/sample, not the confidence level. C and D describe quantities that don't change with the confidence level.
12. (a) We are 95% confident that the true proportion of all trout in the lake carrying the genetic marker is between 0.12 and 0.18.
(b) If the biologist took many random samples of trout and built a 95% confidence interval from each, about 95% of those intervals would capture the true proportion of all trout in the lake carrying the marker.
13. Two errors: (1) It confuses the interval with the data — the interval is about the parameter (true proportion of trout with the marker), not about where 95% of individual trout fall. (2) It double-counts the 95%, attaching it both to "confident" and to "95% of the trout," which conflates the confidence level with a proportion of observations. The correct statement is simply: "We are 95% confident the true proportion of trout with the marker is between 0.12 and 0.18."
14. They should increase the sample size — specifically, since the margin of error scales with 1/√n, cutting it in half requires about four times the sample (from 800 to about 3,200). A larger n shrinks the standard error, which shrinks the margin of error (= critical value × standard error), and because the confidence level is unchanged the critical value stays the same. (Lowering the confidence level would also narrow it but is ruled out here.)
15. Raising the confidence level makes the interval wider (more confident but less precise); lowering it makes the interval narrower (more precise but less confident) — you generally cannot improve both at once for fixed data. The only way to gain both more confidence and more precision is to collect a larger sample, which shrinks the standard error.
FRQ: Full model response and point-by-point rubric appear in section (f) above. Key scoring reminders: the interval interpretation (b) speaks of one interval ("we are 95% confident…"), the level interpretation (c) speaks of many intervals ("about 95% would capture…"); every interpretation must name the true proportion of all online sessions in context; part (d) requires explaining why (fixed parameter, method's long-run rate), not just relabeling "probability" as "confidence"; part (e) must increase n (not lower confidence) and explain the standard-error mechanism.
StatsIQ · Lesson 19 of 30 · Unit 3: Inference for Categorical Data — Proportions · Phase 4: Inference for Proportions
This lesson is study material for the May 2027 AP Statistics exam (new 5-unit CED). "AP" and "AP Statistics" are trademarks of the College Board, which was not involved in producing and does not endorse this product.
Statistical accuracy reviewed: all critical values (z = 1.645, 1.960, 2.576), interval/midpoint arithmetic, and — most critically — every confidence-level and single-interval interpretation have been independently recomputed and checked against the AP scoring conventions. Reviewed for accuracy by Isaac, retired actuary.*