Σ
SigmaPrep
AP Statistics worksheetsBlog
← All articles
AP StatisticsAP Exams

AP Statistics Unit 3 Review: The Most-Tested Inference for Proportions Questions

Σ
Sigma Prep
Math Instructor · 10+ Years Experience
November 2026·13 min read
Key takeaways
  • Unit weight: about a fifth of the exam. 86 of the 406 official multiple-choice questions on today’s course (21.2%), and about 195 of the 1,040 official free-response points we counted (18.7%).
  • Top types: chi-square expected counts and statistics, the one-proportion z-interval, power, and Type I and Type II errors.
  • Practice: Unit 3 is 15.3% of the practice questions we tagged but 21.2% of the official ones. Chi-square calculations, power and the one-proportion interval get well under their share.

Unit 3 is about a fifth of AP Statistics, and it carries the exam’s inference question. College Board’s Course and Exam Description (the CED) gives it 15–25% of the multiple choice. On the six official practice exams from 2012–2019 it was 51 of 215 questions (23.7%), 5 to 11 per exam. In the 24 sample questions of the 2026 CED, the only official questions in the new format, it is 5.

The unit is new as a block: it gathers the proportion half of the old inference units (confidence intervals, significance tests, errors and power for one and two proportions) and the chi-square tests for two-way tables. Chi-square goodness of fit is no longer in the course.

The exam changes in May 2027. Section I becomes 42 multiple-choice questions with four choices instead of five, and some come in sets: two or three questions on one shared stimulus (6 of the 24 CED sample questions are in 2 sets). A study with a confidence interval or a two-way table with a chi-square test is a natural set. Section II becomes 4 free-response questions of 10 points each, scored point by point. At the 2012–2019 rate, about 10 of the 42 multiple-choice questions would be Unit 3; the CED’s 15–25% range puts it at about 6 to 10.

On the free response, Unit 3 earned about 195 of the 1,040 points on the 124 official questions we counted, and 34 of those 124 questions had a Unit 3 part. In the four 2026 sample questions it is 8 of the 40 points, all in sample Question 3.

Unit 3 on the free response: Questions 3 and 4

From 2027 the free-response questions come in a fixed order: Question 1 assesses Practices 1 and 2 (the investigative question and how data are collected), Question 2 Practices 3 and 4 (analyzing data and drawing conclusions), Question 3 inference, and Question 4 a mix of Practices 2, 3 and 4. Question 3 is a full significance test or confidence interval, from the conditions to the conclusion.

Of the 21 official free-response questions that fit the new Question 3, 14 had a Unit 3 part; of the 22 that fit Question 4, 15 did. Inference for proportions was the setting of 12 operational free-response questions from 2012 to 2026, 5 of them since 2019, most recently in 2025, and a chi-square test of 3, most recently in 2017. The 2026 sample Question 3 is a two-proportion test on a randomized experiment.

The 10 most-tested Unit 3 question types

Ranked by the number of official multiple-choice questions on today’s course (13 College Board sources, 1997–2026), with how many of the 13 sources had one, how many were on the six 2012–2019 practice exams, and how many of the 84 operational free-response questions since 2012 had a part on it.

1. Chi-square expected counts, degrees of freedom, statistic and P-value

The arithmetic of a chi-square test for a two-way table: an expected count, the degrees of freedom, one cell’s contribution, the statistic or its P-value.

Topic 3.159 official questions on 8 of 13 sources5 on the 2012–2019 practice exams3 FRQs since 2012, none since 2019
Try it
A library selected a random sample of 315 adult cardholders and recorded each one's age group and preferred kind of book. The table below shows the results. If there is no association between the variables, what is the expected count of cardholders in the 18 to 34 and Fiction cell?
{"rowVar":"Age group","colVar":"Preferred kind","rows":["18 to 34","35 to 54","55 and over"],"cols":["Fiction","Nonfiction","Graphic"],"counts":[[61,52,37],[41,41,8],[35,33,7]]}Age groupFictionNonfictionGraphicTotal18 to 3461523715035 to 54414189055 and over3533775Total13712652315
  1. 39.1439.1439.14
  2. 50.0050.0050.00
  3. 60.0060.0060.00
  4. 65.2465.2465.24
Show solutionHide solution
Solution

(D) 65.2465.2465.24. If there is no association, every age group has the same split of preferred kinds as the whole table. The 18 to 34 row has 150 cardholders and the Fiction column 137 of all 315, so the expected count is 150×137315=65.24\frac{150 \times 137}{315} = 65.24315150×137​=65.24.

Common mistake
multiplying by the wrong total. 90×137315=39.14\frac{90 \times 137}{315} = 39.1431590×137​=39.14 (A) uses the 35 to 54 row. An expected count is always (its own row total)(its own column total)table total\frac{(\text{its own row total})(\text{its own column total})}{\text{table total}}table total(its own row total)(its own column total)​, and it does not have to be a whole number.

Practice: Chi-Square Goodness of Fit worksheet · AP Statistics Unit 3 worksheet

2. Construct a one-proportion z-interval

p^±z∗p^(1−p^)n\hat{p} \pm z^* \sqrt{\frac{\hat{p}(1 - \hat{p})}{n}}p^​±z∗np^​(1−p^​)​​, with the right z∗z^*z∗ for the confidence level.

Topic 3.39 official questions on 6 of 13 sources8 on the 2012–2019 practice exams3 FRQs since 2012, 1 of them since 2019
Try it
To estimate the proportion of all smartphone owners in the city who use a daily screen-time limit, a researcher selected a random sample of 520 smartphone owners. In the sample, 155 of the smartphone owners use a daily screen-time limit. (The city has about 900,000 smartphone owners.) Which of the following is a 90% confidence interval for the proportion of all smartphone owners in the city who use a daily screen-time limit?
  1. (0.246,0.350)(0.246, 0.350)(0.246,0.350)
  2. (0.251,0.345)(0.251, 0.345)(0.251,0.345)
  3. (0.265,0.331)(0.265, 0.331)(0.265,0.331)
  4. (0.278,0.318)(0.278, 0.318)(0.278,0.318)
Show solutionHide solution
Solution

(C) (0.265,0.331)(0.265, 0.331)(0.265,0.331). The sample proportion is p^=155520=0.298\hat{p} = \frac{155}{520} = 0.298p^​=520155​=0.298, and 90% confidence uses z∗=1.645z^* = 1.645z∗=1.645. The standard error is 0.298(0.702)520=0.0201\sqrt{\frac{0.298(0.702)}{520}} = 0.02015200.298(0.702)​​=0.0201, so the margin of error is 1.645(0.0201)=0.0331.645(0.0201) = 0.0331.645(0.0201)=0.033 and the interval is 0.298±0.0330.298 \pm 0.0330.298±0.033, from 0.265 to 0.331.

Common mistake
stopping at the standard error. 0.298±0.0200.298 \pm 0.0200.298±0.020 gives (D), an interval with no confidence level at all. The margin of error is z∗z^*z∗ times the standard error; (A) is the same mistake in the other direction, the 99% z∗=2.576z^* = 2.576z∗=2.576 with a 90% question.

Practice: Confidence Intervals worksheet · AP Statistics Unit 3 worksheet

3. Power: what it is and what increases it

Power is the probability that a test rejects H0H_0H0​ when a particular alternative is true. A larger sample, a larger α\alphaα and a true value farther from H0H_0H0​ all increase it.

Topic 3.88 official questions on 8 of 13 sources4 on the 2012–2019 practice exams1 FRQ since 2012
Try it
Before the museum opened a new wing, 30% of first-time visitors came back within a year. The museum's director will select a random sample of 300 first-time visitors to the science museum and test H0:p=0.30H_0: p = 0.30H0​:p=0.30 against Ha:p>0.30H_a: p > 0.30Ha​:p>0.30 at the α=0.05\alpha = 0.05α=0.05 level, where ppp is the proportion of all first-time visitors to the science museum who come back within a year. Suppose the true value of ppp is 0.40. Which of the following is the power of the test against this alternative?
  1. The probability that the test rejects H0H_0H0​ if the true proportion is 0.40
  2. The probability that the true proportion is 0.40
  3. The probability that the test fails to reject H0H_0H0​ if the true proportion is 0.40
  4. The probability that the test rejects H0H_0H0​ if the true proportion is 0.30
Show solutionHide solution
Solution

(A) The probability that the test rejects H0H_0H0​ if the true proportion is 0.40. Power is a probability of rejecting H0H_0H0​, and it is always computed for one particular alternative. Here the alternative is a true proportion of 0.40, which is in the direction of Ha:p>0.30H_a: p > 0.30Ha​:p>0.30, so the power is the probability that the test rejects H0H_0H0​ when p=0.40p = 0.40p=0.40.

Common mistake
rejecting when the null value is true. (D), the probability of rejecting H0H_0H0​ when p=0.30p = 0.30p=0.30, is α\alphaα, the probability of a Type I error. (C) is the probability of a Type II error, which is one minus the power.

Practice: Hypothesis Tests worksheet · AP Statistics Unit 3 worksheet

4. Type I and Type II errors in context

Name the error that matches a decision, describe it in context, and give its consequence.

Topic 3.84 official questions on 4 of 13 sources, one of them a 2026 CED sample2 on the 2012–2019 practice exams4 FRQs since 2012, 2 of them since 2019
Try it
A state survey reported that 45% of households recycle glass. The county's recycling coordinator will test H0:p=0.45H_0: p = 0.45H0​:p=0.45 against Ha:p≠0.45H_a: p \neq 0.45Ha​:p=0.45, where ppp is the proportion of all households in the county that recycle glass. Which of the following describes a Type II error in this context?
  1. Concluding that the proportion of all households in the county that recycle glass is different from 45% when in fact it is different from 45%
  2. Not concluding that the proportion of the sampled households that recycle glass is different from 45% when in fact it is different from 45%
  3. Not concluding that the proportion of all households in the county that recycle glass is different from 45% when in fact it is different from 45%
  4. Concluding that the proportion of all households in the county that recycle glass is different from 45% when in fact it is 45%
Show solutionHide solution
Solution

(C) Not concluding that the proportion of all households in the county that recycle glass is different from 45% when in fact it is different from 45%. A Type II error fails to reject H0H_0H0​ when H0H_0H0​ is false. Here that means not concluding that the proportion of all households in the county that recycle glass differs from 45% when in fact it does differ from 45%.

Common mistake
writing the error about the sample. (B) says “the proportion of the sampled households”; a test concludes about the population proportion, never about the sample, which is known exactly. (D) is the Type I error.

Practice: Hypothesis Tests worksheet · AP Statistics Unit 3 worksheet

5. Sample size for a margin of error

Solve z∗p∗(1−p∗)n≤Ez^* \sqrt{\frac{p^*(1 - p^*)}{n}} \le Ez∗np∗(1−p∗)​​≤E for nnn, with a prior estimate p∗p^*p∗ or the conservative p∗=0.5p^* = 0.5p∗=0.5, and round up.

Topic 3.34 official questions on 4 of 13 sources2 on the 2012–2019 practice exams1 FRQ since 2012
Try it
The county's recycling coordinator plans to estimate the proportion of all households in the county that recycle glass with a 95% confidence interval. A pilot study suggests the proportion is about 0.45. What is the smallest sample size that will give a margin of error of at most 0.02?
  1. 494949
  2. 167516751675
  3. 237723772377
  4. 240124012401
Show solutionHide solution
Solution

(C) 237723772377. With z∗=1.960z^* = 1.960z∗=1.960 and the pilot estimate p∗=0.45p^* = 0.45p∗=0.45, solve 1.9600.45(0.55)n≤0.021.960\sqrt{\frac{0.45(0.55)}{n}} \le 0.021.960n0.45(0.55)​​≤0.02: n≥(1.9600.02)2(0.45)(0.55)=2376.90n \ge \left(\frac{1.960}{0.02}\right)^2 (0.45)(0.55) = 2376.90n≥(0.021.960​)2(0.45)(0.55)=2376.90. A sample size is a whole number at least that large, so n=2377n = 2377n=2377.

Common mistake
using p∗=0.5p^* = 0.5p∗=0.5 when there is an estimate. 2401 (D) is the conservative size, which is never wrong for the margin but is larger than needed here. Always round up: 2376 would give a margin just over 0.02.

Practice: Confidence Intervals worksheet · AP Statistics Unit 3 worksheet

6. What changes the width of a confidence interval

A larger sample narrows the interval by a factor of n\sqrt{n}n​; a higher confidence level widens it.

Topic 3.44 official questions on 4 of 13 sources1 on the 2012–2019 practice exams1 FRQ since 2012
Try it
To estimate the proportion of all employees of the hospital who carpool to work, the hospital's transportation office selected a random sample of 420 employees of the hospital; 102 of them carpool to work. The margin of error of a 99% confidence interval for the proportion of all employees of the hospital who carpool to work is 0.054. Suppose a random sample of 1,680 were used instead, with about the same sample proportion. How would the margin of error of the 99% interval change?
  1. About the same, because the confidence level is the same.
  2. About half as large, because the margin of error is proportional to 1n\frac{1}{\sqrt{n}}n​1​.
  3. About one-fourth as large, because the sample is four times as large.
  4. About twice as large, because a larger sample gives more variable results.
Show solutionHide solution
Solution

(B) About half as large, because the margin of error is proportional to 1n\frac{1}{\sqrt{n}}n​1​. The margin of error is z∗p^(1−p^)nz^*\sqrt{\frac{\hat{p}(1 - \hat{p})}{n}}z∗np^​(1−p^​)​​, so with the same confidence level and about the same p^\hat{p}p^​ it changes with 1n\frac{1}{\sqrt{n}}n​1​. Going from 420 to 1,680 multiplies nnn by 4 and the margin by 14=12\frac{1}{\sqrt{4}} = \frac{1}{2}4​1​=21​: about 0.027 instead of 0.054.

Common mistake
dividing by the factor nnn grew by, not its square root: one-fourth as large (C). Four times the sample halves the margin of error; to cut it to a quarter, the sample must be 16 times as large.

Practice: Confidence Intervals worksheet · AP Statistics Unit 3 worksheet

7. One-sided and two-sided P-values

Turn a one-sided P-value into a two-sided one, or back, for the same data.

Topic 3.64 official questions on 4 of 13 sources3 on the 2012–2019 practice examsno FRQ since 2012
Try it
The museum's director selected a random sample of 380 first-time visitors to the science museum to test a claim about ppp, the proportion of all first-time visitors to the science museum who come back within a year; the sample proportion was p^=0.3658\hat{p} = 0.3658p^​=0.3658. For H0:p=0.30H_0: p = 0.30H0​:p=0.30 against Ha:p>0.30H_a: p > 0.30Ha​:p>0.30, the P-value was 0.0026. What would the P-value have been for Ha:p≠0.30H_a: p \neq 0.30Ha​:p=0.30, with the same data?
  1. 0.0013, because a two-sided P-value is half the one-sided P-value.
  2. 0.0026, because the data are the same, so the P-value does not change.
  3. 0.9974, because a two-sided test also counts the area on the other side of the observed value.
  4. 0.0052, because a two-sided P-value counts both tails, so it is twice the one-sided P-value in the direction of the data.
Show solutionHide solution
Solution

(D) 0.0052, because a two-sided P-value counts both tails, so it is twice the one-sided P-value in the direction of the data. The sample proportion 0.3658 is above 0.30, in the direction of the one-sided alternative, so 0.0026 is the area of one tail. The two-sided P-value adds the matching tail on the other side: 2(0.0026)=0.00522(0.0026) = 0.00522(0.0026)=0.0052.

Common mistake
halving instead of doubling: 0.0013 (A). A two-sided alternative counts results as extreme in either direction, so its P-value is larger. (C), 1−0.00261 - 0.00261−0.0026, is the P-value only when the data fall on the side opposite a one-sided HaH_aHa​.

Practice: Hypothesis Tests worksheet · AP Statistics Unit 3 worksheet

8. Conditions for two-proportion inference

Random samples or random assignment, independence (the 10% condition for samples only), and large counts in both groups.

Topic 3.103 official questions on 3 of 13 sources, one of them a 2026 CED sample1 on the 2012–2019 practice exams4 FRQs since 2012, none since 2019
Try it
A plant scientist tested whether coating bean seeds with a new compound changes how many seeds sprout. 250 bean seeds were randomly assigned to the coating (100) or no coating (150). Of the 100 coated seeds, 85 sprouted; of the 150 uncoated seeds, 101 sprouted. Which of the following correctly justifies treating the responses as independent for a two-sample zzz procedure?
  1. The bean seeds were randomly assigned to the two treatments, so the responses in the two groups can be treated as independent; no 10% condition is needed because the bean seeds were not sampled from a population.
  2. The bean seeds must be less than 10% of a larger population, so the 10% condition must be checked for each group.
  3. The groups are independent because each group has at least 10 successes and 10 failures.
  4. The groups are independent because the two groups have different sizes.
Show solutionHide solution
Solution

(A) The bean seeds were randomly assigned to the two treatments, so the responses in the two groups can be treated as independent; no 10% condition is needed because the bean seeds were not sampled from a population. In an experiment, the random assignment of the treatments is what lets the two groups be treated as independent. The 10% condition is about sampling without replacement from a population, and the seeds were not sampled from one, so it does not apply.

Common mistake
checking the 10% condition for an experiment (B). It belongs to random samples. Large counts (C) is a separate condition: it makes the sampling distribution approximately normal, not the groups independent.

Practice: Confidence Intervals worksheet · AP Statistics Unit 3 worksheet

9. Carry out a one-proportion z-test

The test statistic z=p^−p0p0(1−p0)/nz = \frac{\hat{p} - p_0}{\sqrt{p_0(1 - p_0)/n}}z=p0​(1−p0​)/n​p^​−p0​​ and its P-value, one-sided or two-sided.

Topic 3.73 official questions on 3 of 13 sources3 on the 2012–2019 practice exams2 FRQs since 2012, both since 2019
Try it
A state survey reported that 40% of households recycle glass. The county's recycling coordinator wants to know whether the proportion of all households in the county that recycle glass is different from 40%. The county's recycling coordinator selected a random sample of 370 households in the county; 159 of them recycle glass. The county's recycling coordinator will test H0:p=0.40H_0: p = 0.40H0​:p=0.40 against Ha:p≠0.40H_a: p \neq 0.40Ha​:p=0.40, where ppp is the proportion of all households in the county that recycle glass. The figure below shows the sampling distribution of p^\hat{p}p^​ when H0H_0H0​ is true, with the observed sample proportion marked and the area for the P-value shaded. Which of the following gives the test statistic and the shaded area?
{"mu":0.4,"sigma":0.025468581598680533,"obs":0.4297297297297297,"alt":"different"}observed0.32360.34910.37450.40000.42550.45090.4764Sample proportion (p̂) if p = 0.40
  1. z=1.17z = 1.17z=1.17 and P-value =0.1215= 0.1215=0.1215
  2. t=1.17t = 1.17t=1.17 and P-value =0.2438= 0.2438=0.2438
  3. z=1.17z = 1.17z=1.17 and P-value =0.2431= 0.2431=0.2431
  4. z=1.17z = 1.17z=1.17 and P-value =0.7569= 0.7569=0.7569
Show solutionHide solution
Solution

(C) z=1.17z = 1.17z=1.17 and P-value =0.2431= 0.2431=0.2431. The sample proportion is p^=159370=0.4297\hat{p} = \frac{159}{370} = 0.4297p^​=370159​=0.4297, and the test’s standard error uses the null value: 0.40(0.60)370=0.0255\sqrt{\frac{0.40(0.60)}{370}} = 0.02553700.40(0.60)​​=0.0255. So z=0.4297−0.400.0255=1.17z = \frac{0.4297 - 0.40}{0.0255} = 1.17z=0.02550.4297−0.40​=1.17. The alternative is two-sided, so the P-value is the area beyond 1.17 in both tails: 0.2431.

Common mistake
shading one tail for a two-sided test: 0.1215 (A). Read HaH_aHa​ before you compute. And a test for a proportion uses zzz, never ttt (B).

Practice: Hypothesis Tests worksheet · AP Statistics Unit 3 worksheet

10. Bias and variability of estimators

An unbiased estimator is centered at the parameter; among unbiased ones, the one with less variability is better.

Topic 3.13 official questions on 3 of 13 sources2 on the 2012–2019 practice exams1 FRQ since 2012
Try it
A wildlife biologist is comparing three methods, A, B and C, of estimating the mean length of the trout in a lake. To compare them, a simulation took 30 random samples from a population in which the mean length of all trout in the lake is 32.3 centimeters, and used each method on each sample. The dotplots below show the 30 estimates from each method; the dashed line marks 32.3. Based on the dotplots, which method is the best choice for estimating the mean length of all trout in the lake?
{"groups":[{"name":"Estimator A","values":[36,36.4,35.8,36.4,35,35.6,36,36.6,36.6,35.2,35.8,37.2,36.2,35,36,37,35.2,35.6,36.2,35,35,35.8,36.2,36,35.4,36.6,36.4,34.8,36,35.8]},{"name":"Estimator B","values":[33,32.4,32.4,31.8,30.4,33,31.4,31.8,30.8,31.6,32.6,32.4,34.4,32.8,33,32,32.4,33,33.6,29.6,31.4,32.2,31.6,32.2,32.6,34.4,32.2,33.2,31.8,33.2]},{"name":"Estimator C","values":[30.2,31.2,32.2,29.8,36.4,32.6,31.2,35,36.8,27.2,33.6,33.2,33.8,36.8,31,29.8,29.6,33.2,35.2,32.4,32.6,29,32.8,32.8,32.6,35,30.2,30.8,33.6,30.8]}],"unit":0.2,"axis":{"lo":27,"hi":38,"step":1,"dp":0},"observed":32.3}Estimator AEstimator BEstimator C272829303132333435363738Estimate (centimeters)
  1. Estimator B, because it is unbiased and has less variability than Estimator C
  2. Estimator C, because it is unbiased
  3. Estimator A, because its values are closest together, so it is unbiased
  4. Estimator A, because it has the least variability
Show solutionHide solution
Solution

(A) Estimator B, because it is unbiased and has less variability than Estimator C. Estimators B and C are centered at 32.3, so they appear unbiased; Estimator A is centered above 32.3, so it is biased. Of the two unbiased estimators, B has the smaller spread, so its estimates tend to be closer to the true mean length.

Common mistake
calling the tightest dotplot unbiased (C, D). Low variability and no bias are different things: Estimator A is precise but consistently too high.

Practice: Sampling Distributions worksheet · AP Statistics Unit 3 worksheet

On the free response: what multiple choice rarely asks

  • Hypotheses for a two-proportion test. No official multiple-choice question, but 4 free-response questions since 2012, 2 of them since 2019. Define p1p_1p1​ and p2p_2p2​ in context, as population (or true treatment) proportions.
  • The conclusion of a two-proportion test. No official multiple-choice question, 4 free-response questions, 2 of them since 2019. Compare the P-value with α\alphaα, decide, and state the conclusion about the population proportions in terms of HaH_aHa​, never “accept H0H_0H0​”.
  • Chi-square hypotheses and conditions. 1 official multiple-choice question, 4 free-response questions, 2 of them since 2019. Every expected count must be at least 5.

Rising and fading in Unit 3

We compare the 42 operational free-response questions of 2012–2018 with the 42 of 2019–2026 (both 7 exams; 2020 was never released).

  • Unit 3 as a whole is fading on the free response: 24.5% of the free-response points in 2012–2018, 14.3% in 2019–2026 (14 questions with a Unit 3 part, then 10).
  • Rising: one-proportion tests and choosing a chi-square test. Carrying out a one-proportion z-test went from no question to 2, and choosing between homogeneity and independence from none to 2.
  • Steady: two-proportion tests and errors. The two-proportion z-test was in 2 questions in each period, and Type I and Type II errors in 2 and 2.
  • Fading: chi-square arithmetic and two-proportion conditions. Chi-square calculations went from 3 questions to none, and the conditions for two proportions from 4 to none.

Never asked yet: a confidence interval for a difference of proportions used to judge a claim

Topic 3.11, using or interpreting a confidence interval for p1−p2p_1 - p_2p1​−p2​, was in the course before 2026, but no official question has asked about it, multiple choice or free response. It is the same reasoning as a one-proportion interval and a claim (is 0 in the interval?), so practice it with type 2 above.

What students over- and under-practice in Unit 3

We tagged 1,130 multiple-choice questions from teacher tests, review packets, AP Classroom exports and prep books (reposted official questions removed) and compared each type’s share of that practice with its share of the 406 official questions. Unit 3 is 15.3% of the practice questions against 21.2% of the official ones.

Unit 3 question types: share of 1,130 practice multiple-choice questions against share of the 406 official ones
Question typePractice vs official shareRatio
Interpret a P-value1.59% vs 0.25%6.5× over
Conditions for one-proportion inference0.71% vs 0.25%2.9× over
Sampling distribution of p^\hat{p}p^​1.24% vs 0.74%1.7× over
Type I and Type II errors1.42% vs 0.99%1.4× over
Chi-square calculations0.27% vs 2.22%0.1× under
One-sided vs two-sided P-values0.27% vs 0.99%0.3× under
One-proportion z-interval0.80% vs 2.22%0.4× under
Sample size for a margin of error0.35% vs 0.99%0.4× under
Power0.80% vs 1.97%0.4× under

Official counts per type are small (1 to 9 questions), so read the ratios as directions.

  • Interpreting a P-value is drilled at six times its multiple-choice share. It does matter on the free response (in 2026, for one), but the multiple choice asks for the computation more often than the sentence.
  • The two most-asked types get a third of their share or less. Chi-square calculations and the one-proportion interval each have 9 official questions; practice them as calculations, not only as definitions.
  • Power is under-practiced although 8 of the 13 sources asked about it.

How we counted

Every number here comes from public official AP Statistics material: 406 multiple-choice questions on today’s course from 13 College Board sources (the 1997, 2002 and 2007 released exams, the 2008, 2012–2015, 2018 and 2019 practice exams, and the sample questions in the 2005, 2020 and 2026 course descriptions), and 124 free-response questions (the 2012–2026 exams except 2020, the practice exams and the four 2026 samples; 1,040 points). Free-response points before 2027 are split from each question’s holistic score in proportion to its scoring sections. Each question is tagged to its CED topic and question type. Every example problem is our own.

Practice Unit 3

Our free AP Statistics Unit 3 worksheet has 20 original multiple-choice questions with four choices, including two sets of questions on one shared stimulus (a one-proportion confidence interval and a two-way table for a chi-square test), weighted by how often the official exams ask each type. It comes with an answer key, worked solutions and a printable PDF.

For free response, the AP Statistics Unit 3 FRQ set has 2 questions in the style of Questions 3 and 4 of the 2027 exam, including a full two-proportion test like Question 3, each scored out of 10 points, with a scoring guide and a PDF. Both make good practice before the Unit 3 Progress Check in AP Classroom.

The formulas this unit uses, each with a worked example and the calculator command, are on our free AP Statistics formula sheet.

Unit 3 builds on Unit 2’s normal distributions and sampling distributions: see the AP Statistics Unit 2 review and the Unit 2 worksheet.

← Previous unitUnit 2: Probability, Random Variables, and Probability DistributionsNext unit →Unit 4: Inference for Quantitative Data: Means

Common Questions

The CED gives Unit 3 15–25% of the multiple choice. On the six official practice exams from 2012–2019 it was 51 of 215 multiple-choice questions (23.7%), 5 to 11 per exam. On the free response it earned about 195 of the 1,040 points on the 124 official questions we counted (18.7%), and 34 of those questions had a Unit 3 part.

Mostly Questions 3 and 4. Question 3 is a full inference question, a significance test or a confidence interval; of the 21 official free-response questions that fit it, 14 had a Unit 3 part. Of the 22 that fit Question 4, 15 had a Unit 3 part, often a Type I or Type II error, power or a P-value.

1-PropZInt and 2-PropZInt for intervals, 1-PropZTest and 2-PropZTest for tests, and the χ²-Test for a two-way table (it also gives the expected counts). On the free response, name the procedure, state the hypotheses and check the conditions in writing; a calculator result alone does not earn those points.

Start with the most-tested question types above and the common mistake for each, then take a timed 20-question practice test with an answer key (about 2 minutes a question, as on the exam) and the unit’s free-response questions with a scoring guide. Read the worked solution of every question you missed, then redo those types.

Practice this unit: free AP Statistics Unit 3 worksheet

Exam-style multiple choice in the 2027 format, with an answer key and a printable PDF. No sign-up needed.

Get the Unit 3 worksheet →

Keep reading

AP Statistics

AP Statistics Unit 1 Review: The Most-Tested Data and Study Design Questions

AP Statistics Unit 1 review for the 2027 exam: the data and study design questions the official exams ask most, with examples and a practice test.

AP Statistics

AP Statistics Unit 2 Review: The Most-Tested Probability and Distributions Questions

AP Statistics Unit 2 review for the 2027 exam: the probability and distribution questions the official exams ask most, with examples and a practice test.

AP Statistics

AP Statistics Unit 4 Review: The Most-Tested Inference for Means Questions

AP Statistics Unit 4 review for the 2027 exam: the inference for means questions the official exams ask most, with examples and a practice test.

Σ
SigmaPrep

Sigma Prep (sigmaprep.io) makes free AP Statistics practice: exam-style multiple choice and free response for all five units, with answer keys and scoring guides, plus a formula sheet.

AP Statistics

Practice tests by unitFormula sheetUnit reviewsStatistics worksheets

Sigma Prep

Math worksheetsAboutContact

Legal

TermsPrivacy
© 2026 Sigma Prep. All rights reserved.AP® is a trademark registered by the College Board, which is not affiliated with, and does not endorse, this site.