AP Statistics | 2026–27 Course Revision, Digital Exam and Preparation

AP Statistics 2026–27 | Revised Five-Unit Course, Fully Digital 2027 Exam and Statistical Reasoning Guide

AP Statistics changed substantially for the 2026–27 school year. College Board revised the course to align more closely with current introductory college statistics courses and with the statistical problem-solving process emphasized by the American Statistical Association. The revision reorganises the course into five units, changes the prerequisite expectation, removes several legacy topics, restructures the exam, and moves the May 2027 AP Statistics Exam to fully digital testing in Bluebook.

This means older AP Statistics material can still teach useful statistics, but it should not automatically be treated as a current-format simulation. The current course and exam must be read through the 2026–27 framework.

The revised course has five units:

1. Exploring One-Variable Data and Collecting Data;
2. Probability, Random Variables, and Probability Distributions;
3. Inference for Categorical Data: Proportions;
4. Inference for Quantitative Data: Means;
5. Regression Analysis.

College Board also defines four statistical practices: Formulate Questions, Collect Data, Analyze Data, and Interpret Results. The entire course should be taught as a cycle rather than as disconnected formula units.

1. The prerequisite changed

College Board removed the previous second-year algebra prerequisite. The revised course is intended to be accessible to students who have successfully completed a first-year algebra course.

This does not mean algebra is unimportant. Students still need enough algebra to:

• rearrange formulas;
• interpret linear relationships;
• work with proportions;
• understand standardized values;
• use calculator/statistical output correctly.

The shift reflects the conceptual emphasis of statistics rather than a lowering of mathematical rigor.

2. The course is now five units, not nine

Legacy AP Statistics materials may still refer to Units 1–9. For the 2026–27 course, College Board consolidated the course into five units.

Current unit titles and approximate multiple-choice weightings are:

• Unit 1: Exploring One-Variable Data and Collecting Data — 20%–30%;
• Unit 2: Probability, Random Variables, and Probability Distributions — 15%–25%;
• Unit 3: Inference for Categorical Data: Proportions — 15%–25%;
• Unit 4: Inference for Quantitative Data: Means — 10%–20%;
• Unit 5: Regression Analysis — 10%–20%.

3. Several legacy topics were removed

College Board states that the revised course removed topics including:

• analyzing departures from linearity;
• combining random variables as a standalone topic;
• the geometric distribution;
• chi-square goodness-of-fit;
• the former Unit 9 inference for slopes.

This matters for students using old textbooks or practice questions. The removed material can still be interesting statistics, but it should not displace current required content.

4. Statistical practice begins with a question

A statistical investigation should begin with an investigative question, not with a formula.

Examples:

• Do students who sleep at least eight hours tend to report lower stress?
• Is the proportion of customers choosing option A different from 50%?
• Is the mean battery life above a claimed standard?

The question determines what data should be collected and which analysis is appropriate.

5. Practice 1: Formulate Questions

A valid investigative question should involve variability and a population or process.

Weak:

“What is Jamie’s height?”

Stronger:

“What is the distribution of heights among students at this school?”

The second question requires data and statistical reasoning.

6. Practice 2: Collect Data

Students should distinguish:

• observational study;
• sample survey;
• experiment;
• random sampling;
• random assignment;
• bias;
• confounding.

Sampling supports generalisation to a population. Random assignment supports causal inference. Those are different jobs.

7. Sample versus experiment

If 100 randomly selected students are surveyed about exercise and grades, the study can support population generalisation if the sample is representative. It cannot establish that exercise caused higher grades.

If students are randomly assigned to different exercise programs, causal reasoning becomes stronger, but population generalisation still depends on how the participants were recruited.

8. Original data-collection task

A school wants to estimate the proportion of students who use public transport.

Bad method: survey only students waiting at the bus stop.

Better method: randomly sample students from the school roster.

The first method creates selection bias because transport users are overrepresented.

9. Unit 1: one-variable distributions

Students should describe distributions by:

• shape;
• center;
• variability;
• unusual values.

A summary statistic without distribution context can mislead.

10. Mean versus median

Data:

[ 12,13,13,14,48. ]

Mean:

[ 100/5=20. ]

Median = 13.

The distribution is right-skewed because of the extreme value 48. Median better describes a typical central value.

11. Standard deviation is a distance concept

Standard deviation measures typical distance from the mean, in the same units as the data.

It should not be described merely as “spread” without reference to the mean.

12. Standardized values

A z-score is:

[ z=rac{x-mu}{sigma}. ]

If (z=2), the observation is two standard deviations above the mean.

z-scores allow comparisons across different scales.

13. Percentiles are positional

A student at the 80th percentile scored at least as high as about 80% of the comparison group.

It does not mean the student scored 80% of the questions correctly.

14. Unit 2: probability should model uncertainty

Probability is not merely fraction calculation. Students should understand:

• sample spaces;
• conditional probability;
• independence;
• simulation;
• random variables;
• expected value;
• probability distributions.

15. Conditional probability

If:

[ P(Acap B)=0.12,quad P(B)=0.30, ]

then:

[ P(A|B)=0.12/0.30=0.40. ]

The denominator changes because the condition restricts the sample space.

16. Independence should be checked mathematically

Events (A) and (B) are independent when:

[ P(A|B)=P(A) ]

or equivalently:

[ P(Acap B)=P(A)P(B). ]

Independence is a relationship, not a visual impression.

17. Simulation is a statistical tool

Simulation can approximate probabilities and sampling distributions when analytic calculation is difficult or when the goal is to understand variability.

A valid simulation must reflect the original model’s probabilities and repeated structure.

18. Random variables

A random variable maps outcomes to numerical values.

Expected value:

[ E(X)=sum xP(X=x). ]

It is a long-run average, not necessarily a possible single outcome.

19. Original expected-value task

A game pays:

• $10 with probability 0.2;
• $0 with probability 0.8.

Expected payout:

[ 10(0.2)+0(0.8)=2. ]

Expected value is $2, even though $2 is never actually paid.

20. Binomial distribution

A binomial model requires:

• fixed number of trials;
• independent trials;
• two outcome categories;
• constant success probability.

If these conditions fail, the binomial model may be invalid.

21. Original binomial task

Let:

[ Xsim B(10,0.3). ]

Probability of exactly 4 successes:

[ P(X=4)=inom{10}{4}(0.3)^4(0.7)^6. ]

22. Sampling distributions connect sample to population

Students should distinguish:

• population distribution;
• sample distribution;
• sampling distribution.

A sampling distribution describes how a statistic varies across repeated samples.

23. Standard error is about statistic variability

For a sample proportion:

[ SE(hat p)=sqrt{rac{p(1-p)}n} ]

in the theoretical sampling model.

Larger (n) reduces sampling variability.

24. Unit 3: inference for proportions

Students should understand:

• confidence intervals for one proportion;
• confidence intervals for difference of proportions;
• significance tests for one proportion;
• significance tests for two proportions;
• chi-square inference for categorical data where retained by the revised framework.

25. Confidence intervals estimate parameters

A confidence interval is built from a statistic plus/minus a margin of error.

The correct interpretation is about the procedure and parameter, not the probability that one fixed interval “contains the parameter” after it has been calculated.

26. Original confidence-interval interpretation

A 95% confidence interval for a population proportion is:

[ (0.42,0.50). ]

Reasonable interpretation:

“We are 95% confident that the true population proportion lies between 0.42 and 0.50.”

Do not say “95% of individuals lie in this interval.”

27. Margin of error

Margin of error decreases as sample size increases.

To cut a margin of error roughly in half, sample size typically must be multiplied by about four.

This square-root relationship is conceptually important.

28. Hypothesis tests begin with claims

A significance test should identify:

• null hypothesis;
• alternative hypothesis;
• test statistic;
• p-value;
• significance level;
• contextual conclusion.

29. p-values are conditional probabilities

A p-value is the probability, assuming the null hypothesis is true, of obtaining a result as extreme as or more extreme than the observed result.

It is not the probability that the null hypothesis is true.

30. Original hypothesis-test interpretation

If (p=0.03) and (alpha=0.05), reject (H_0).

Then return to context:

“The data provide statistically significant evidence for the alternative claim.”

Do not write only “0.03 < 0.05”.

31. Type I and Type II errors

Type I error: reject a true null hypothesis.

Type II error: fail to reject a false null hypothesis.

Students should interpret errors in the context of the specific claim.

32. Power

Power is the probability of correctly rejecting a false null hypothesis.

Power tends to increase with:

• larger sample size;
• larger effect size;
• larger significance level;
• lower variability.

33. Unit 4: inference for means

Students should work with:

• one-sample t procedures;
• paired data;
• two-sample t procedures;
• conditions and interpretation.

34. Why t rather than z?

When population standard deviation is unknown, inference for means commonly uses the t distribution and sample standard deviation.

The t distribution accounts for additional uncertainty.

35. Paired versus independent samples

Paired data arise when observations are naturally linked, such as:

• before/after measurements;
• matched individuals;
• two measurements on the same subject.

Analyze the differences, not two independent groups.

36. Original paired-data task

Students take a test before and after a study program. For each student define:

[ d= ext{after}- ext{before}. ]

Inference is performed on the mean difference (mu_d).

The pairing reduces irrelevant between-person variation.

37. Unit 5: regression analysis

The revised course gives regression its own unit.

Students should understand:

• scatterplots;
• direction/form/strength;
• correlation;
• least-squares regression;
• residuals;
• coefficient of determination;
• inference for regression where retained in the revised framework.

38. Correlation is standardized association

Correlation (r) measures strength and direction of a linear relationship.

It is unitless and lies between -1 and 1.

Correlation is not causation.

39. Least-squares regression line

A line:

[ hat y=a+bx ]

predicts response (y) from explanatory variable (x).

The slope (b) should be interpreted in units:

“For each 1-unit increase in x, predicted y changes by about b units.”

40. Residuals

Residual:

[ e=y-hat y. ]

A positive residual means the observed value is above the predicted value.

Residual plots help assess whether a linear model is appropriate.

41. (r^2) interpretation

(r^2) is the proportion of variability in the response variable explained by the linear relationship with the explanatory variable.

It is not “the percentage of points on the line.”

42. Extrapolation caution

Predictions outside the observed x-range can be unreliable because the relationship may change.

A regression equation does not guarantee the trend continues indefinitely.

43. Causation requires design

An observational association can be strong and still noncausal.

Randomized experiments provide stronger causal evidence because random assignment helps balance confounding variables.

44. Revised statistical practices

The four practices should appear throughout all units:

1. Formulate Questions;
2. Collect Data;
3. Analyze Data;
4. Interpret Results.

A student who can calculate but cannot ask or interpret is not AP Statistics-ready.

45. The 2027 exam is fully digital

College Board states that the May 2027 AP Statistics Exam is fully digital. Students complete both multiple-choice and free-response questions in Bluebook. Paper FRQ booklets are no longer used.

Students still receive scratch paper and a printed reference-information booklet, while the same reference information is also available in Bluebook.

46. 2027 exam date

The AP Statistics Exam is scheduled for Tuesday, May 11, 2027, Session 2.

47. Multiple-choice structure

42 questions in 1 hour 30 minutes.

50% of exam score.

Questions can be individual or appear in sets based on a shared prompt.

48. Free-response structure

4 questions in 1 hour 30 minutes.

50% of exam score.

Current question roles:

• Question 1: multi-focus on Practices 1 and 2;
• Question 2: multi-focus on Practices 3 and 4;
• Question 3: inference—hypothesis test or confidence interval;
• Question 4: multi-focus on Practices 2, 3 and 4.

49. The old six-FRQ format is obsolete for 2027

Legacy exams had 6 FRQs. The current 2027 exam has 4 FRQs, each worth more points.

Old FRQs remain useful for content practice, but final pacing simulations must use the current 4-question structure.

50. Multiple-choice answer choices changed

The revised AP Statistics exam has four answer choices per MCQ rather than five.

This makes mathematical elimination slightly more valuable, but students should not rely on guessing.

51. Two MCQ sets are new

College Board says the revised exam includes two 3-question multiple-choice sets:

• one focused on probability, random variables and distributions;
• one focused on regression analysis.

Students should practise carrying information across related questions.

52. Calculator policy changed for 2027

College Board’s 2027 calculator policy expects graphing calculators with statistical capabilities. Nongraphing calculators are not allowed for AP Statistics in 2027, even if they have statistics functions.

Bluebook also provides the built-in Desmos graphing calculator.

53. Calculator output must still be interpreted

A calculator can produce:

• mean;
• standard deviation;
• regression coefficients;
• probabilities;
• test statistics;
• p-values;
• confidence intervals.

The exam still requires students to know what those numbers mean and when the procedure is valid.

54. Fully digital FRQ writing needs practice

Students should practise entering statistical reasoning with a conventional keyboard.

College Board provides digital-response guidance and updated Bluebook symbols to reduce notation friction.

Strong responses remain verbal and contextual, not symbol-heavy.

55. Digital FRQ should be concise but complete

A strong inference response can follow:

1. state parameter;
2. state hypotheses or confidence target;
3. check conditions;
4. name procedure;
5. report statistic/p-value or interval;
6. interpret in context.

56. Do not type calculator commands as reasoning

“1-PropZTest gave p=.031” is incomplete.

The student should identify what is being tested and what the p-value implies.

57. Current AP Statistics timing

MCQ:

[ 90/42approx2.14 ext{ minutes per question}. ]

FRQ:

[ 90/4=22.5 ext{ minutes per question on average}. ]

Question roles differ, so timing should be flexible.

58. Final MCQ pacing plan

Possible checkpoints:

• Q14 around 28–30 minutes;
• Q28 around 58–60 minutes;
• Q42 by 85 minutes;
• final 5 minutes for flagged items.

59. FRQ pacing plan

A possible plan:

• Q1: 20 minutes;
• Q2: 20 minutes;
• Q3 inference: 25 minutes;
• Q4 multi-focus: 20 minutes;
• 5 minutes review.

60. Old textbooks can still be useful

College Board says textbooks from the last 10 years can adequately address the revised course material. Teachers and students should map old chapters to the current five-unit framework and skip removed topics when final exam time is limited.

61. Legacy-topic labeling

If using an old question on geometric distribution or inference for slopes, label it:

legacy enrichment—not current 2026–27 required AP Statistics content.

62. Error ledger

63. First-invalid-step checking

If a student chooses a two-sample procedure for paired data, every later calculation may be technically correct but invalid.

Repair the procedure choice first.

64. Fresh-transfer rule

After a confidence-interval error, change the parameter.

After a sampling-bias error, change the survey context.

After a regression error, change the variables.

Statistical reasoning is transferable only if the student can recognize the structure in a new study.

65. Four-week AP Statistics recovery

Week 1: Unit 1 data/design.

Week 2: Unit 2 probability/distributions.

Week 3: Units 3–5 inference/regression.

Week 4: fully digital MCQ + typed FRQ, error repair and fresh transfer.

66. Eight-week plan

Weeks 1–2: data/design.

Weeks 3–4: probability/random variables.

Weeks 5–6: inference for proportions/means.

Week 7: regression and mixed practice.

Week 8: fully digital exam simulation.

67. Final practice should use the revised format

Use:

• 42 MCQ;
• 4 answer choices;
• 90 minutes;
• 4 FRQ;
• 90 minutes;
• fully digital Bluebook responses;
• graphing calculator;
• current reference sheet.

68. Final learner release checklist

• five-unit framework understood;
• study design reasoning stable;
• probability/model conditions explicit;
• confidence intervals interpreted correctly;
• hypothesis tests contextualized;
• regression claims bounded;
• graphing calculator policy current;
• digital FRQ typing practised;
• legacy 6-FRQ format not used for final pacing.

69. Official current-source control

Use AP Statistics Course, AP Statistics Revisions, and AP Statistics Exam for the current 2026–27 framework and May 2027 exam.

70. The durable AP Statistics model

question → data collection → analysis → inference → interpretation → current digital format → first-error repair → fresh transfer.

The revised course is smaller in unit count but broader in statistical reasoning.

AP Statistics Mastery Lab | Study Design, Inference, Digital FRQ Writing and 2027 Exam Execution

The final AP Statistics stage should test whether the student can complete the statistical problem-solving cycle independently: ask a valid question, decide how data should be collected, select an appropriate analysis, and interpret the result with the correct scope. The May 2027 exam makes this especially important because the revised course and fully digital exam reward connected reasoning more explicitly than the legacy structure.

1. One investigation, four practices

Suppose a school wants to know whether a new tutoring program improves algebra performance.

Formulate Questions: Does participation in the program increase average algebra scores?

Collect Data: Decide whether the study will be observational or experimental and how students are selected/assigned.

Analyze Data: Compare groups using appropriate numerical/graphical/inferential methods.

Interpret Results: State what the evidence supports and what cannot be concluded.

This four-practice loop should become the default statistical operating system.

2. Planted error: sample versus population

A student surveys 50 volunteers from one AP class and concludes that “60% of all teenagers prefer online homework.”

The first invalid step is scope. The sample may not represent all teenagers.

Repair by asking:

• Who was sampled?
• How were they selected?
• What population can reasonably be generalized to?

3. Planted error: observational study versus experiment

A study finds that students who exercise more have higher GPA and concludes exercise causes higher GPA.

The study is observational. Confounding variables may explain the association.

Causal conclusions require stronger design, typically random assignment to treatments.

4. Planted error: random sampling versus random assignment

Random sampling helps generalize to a population.

Random assignment helps compare treatments and support causal inference.

Students often use the words interchangeably; the exam expects the distinction.

5. Original design task

A company wants to test whether a new webpage increases purchases.

A strong experiment randomly assigns visitors to old/new webpage versions and compares purchase rates.

The random assignment supports causal interpretation because treatment assignment is controlled.

6. One-variable description should be complete

For a distribution, address:

• shape;
• center;
• spread;
• unusual features.

Do not list numerical summaries without describing what they reveal.

7. Planted error: mean in a skewed distribution

A right-skewed income distribution has mean $82,000 and median $54,000. A student calls $82,000 “the typical income.”

The mean is pulled upward by high values. Median is often more representative of center in a strongly skewed distribution.

8. Standard deviation interpretation

If a dataset has mean 50 and standard deviation 8, standard deviation describes typical distance from the mean in the same units as the original variable.

It is not “the range” and it is not a percentage unless the variable itself is a percentage.

9. Z-score comparison

Student A scores 84 on a test with mean 70, SD 7:

[ z_A=(84-70)/7=2. ]

Student B scores 92 on a test with mean 80, SD 8:

[ z_B=12/8=1.5. ]

A performed better relative to the comparison distribution.

10. Probability trees should represent conditional structure

A tree diagram is not decorative. Each second-stage probability is conditional on the first branch.

If sampling is without replacement, the second-stage probabilities change.

11. Original conditional-probability task

A disease has prevalence 2%. A screening test is positive for 90% of affected individuals and 5% of unaffected individuals.

Students should build:

• affected/not affected;
• positive/negative conditional branches;
• joint probabilities;
• conditional probability given a positive result.

The base rate matters.

12. Independence should be demonstrated

If:

[ P(A)=0.4,quad P(B)=0.5,quad P(Acap B)=0.2, ]

then:

[ P(A)P(B)=0.20=P(Acap B). ]

The events are independent.

13. Random-variable expected value should stay contextual

If a game has expected value (-$1.20), that means the long-run average net outcome per play is a loss of $1.20, not that every player loses exactly $1.20.

14. Sampling distributions are where inference begins

The sampling distribution describes how a statistic varies over repeated samples. Students should distinguish variability of individual observations from variability of a statistic such as (hat p) or (ar x).

15. Larger samples reduce sampling variability

Standard error generally shrinks at a square-root rate as sample size grows.

Quadrupling sample size approximately halves many standard errors, all else equal.

16. Planted error: confidence interval probability

A student writes:

“There is a 95% probability the true proportion lies in this calculated interval.”

After the interval is calculated, the population parameter is fixed. The 95% refers to the long-run success rate of the confidence-interval procedure.

17. Confidence interval response template

1. identify parameter;
2. check conditions;
3. name procedure;
4. calculate interval;
5. interpret in context.

The exam rewards the whole inferential argument, not just calculator endpoints.

18. Planted error: p-value

A student says:

“p = 0.02 means there is a 2% chance the null hypothesis is true.”

Incorrect.

The p-value is computed assuming the null hypothesis is true. It measures how surprising the observed statistic or more extreme values would be under that assumption.

19. Significance does not measure practical importance

A tiny effect can be statistically significant in a large sample.

A meaningful effect can fail to reach significance in a small noisy sample.

Students should separate statistical evidence from practical magnitude.

20. Conditions are part of inference

Do not perform a procedure first and invent conditions afterward.

Check:

• randomness/independence;
• sample-size conditions;
• distribution conditions;
• 10% condition where relevant;
• paired versus independent design.

21. Paired-data planted error

Before/after data from the same individuals are analyzed as two independent samples.

The pairing has been lost.

Correct approach: define differences for each individual and analyze the mean difference.

22. Regression slope interpretation

If:

[ hat y=12+3.4x, ]

then for each additional 1 unit of (x), predicted (y) increases by 3.4 units on average, within the range and context of the model.

23. Intercept interpretation can be meaningless

If (x=0) is outside the observed or meaningful range, the intercept may have no useful contextual interpretation.

Students should not force an interpretation merely because a coefficient exists.

24. Residual meaning

If observed (y=70) and predicted (hat y=64):

[ e=70-64=6. ]

The model underpredicts that observation by 6 units.

25. Residual plots diagnose model form

A clear curved pattern in residuals suggests a linear model misses structure.

Random scatter around zero supports—but does not prove—the reasonableness of a linear model.

26. Planted error: correlation causation

A high (r) does not imply one variable causes the other.

Study design controls causal interpretation, not the numerical size of (r).

27. AP Statistics free-response writing should be statistical prose

Use full sentences where interpretation matters.

Good:

“Because the p-value is less than 0.05, there is statistically significant evidence that the population proportion differs from 0.50.”

Weak:

“Reject H0 because .03 < .05.”

28. Typed FRQ should be concise

Fully digital does not mean longer answers. Students should type the minimum complete statistical argument.

A strong response names:

• parameter;
• procedure;
• evidence;
• conclusion.

29. Practice statistical notation digitally

Students should rehearse how they will type:

• (hat p);
• (ar x);
• (mu);
• hypotheses;
• intervals;
• inequalities.

College Board’s digital response guidance is the appropriate current reference.

30. Calculator output needs labels

If a graphing calculator returns:

z = 2.14, p = 0.0324

the student should know which test produced those quantities and how to use them in a conclusion.

31. Current calculator-release drill

Students should be able to use an approved graphing calculator or built-in Desmos to obtain statistical results, but also to:

• enter lists correctly;
• identify one- versus two-variable statistics;
• run regression;
• compute distributions/probabilities;
• check whether the selected procedure matches the problem.

32. 2027 graphing-calculator rule should be explicit

For AP Statistics in 2027, College Board says nongraphing calculators are not allowed. Use an approved graphing calculator with statistical capabilities and/or the built-in Desmos graphing calculator in Bluebook.

33. Old calculator habits may now be noncompliant

A student who previously used an approved scientific/nongraphing statistics calculator should recheck the 2027 calculator list rather than assume legacy permission continues.

34. Fully digital test-day practice

A simulation should include:

• Bluebook-style screen reading;
• typing FRQ responses;
• scratch paper for planning/calculation;
• printed/reference information availability;
• graphing calculator use;
• time management across 90+90 minutes.

35. MCQ sets require information continuity

The revised exam contains shared-prompt question sets. Students should avoid re-reading the entire prompt from zero for each question. Annotate variables, study design and relevant conditions once, then update as needed.

36. Probability set preparation

One shared-prompt set focuses on probability/random variables/distributions.

Practice carrying:

• distribution parameters;
• event definitions;
• conditional information;
• expectation/variability relationships

across several connected questions.

37. Regression set preparation

The other shared-prompt set focuses on regression analysis.

Practice carrying:

• variable roles;
• slope/intercept;
• correlation;
• residuals;
• model limitations

across connected items.

38. FRQ 1 preparation

Question 1 emphasizes formulating questions and collecting data.

Practice:

• investigative questions;
• sampling methods;
• experimental design;
• bias/confounding;
• scope of inference.

39. FRQ 2 preparation

Question 2 emphasizes analyzing data and interpreting results.

Practice:

• graphical summaries;
• numerical summaries;
• probability/model output;
• clear contextual explanation.

40. FRQ 3 preparation

Question 3 is an inference question: confidence interval or hypothesis test.

Students should have a compact inference workflow that includes conditions and context.

41. FRQ 4 preparation

Question 4 is multi-focus and can integrate multiple content areas.

Do not expect one chapter. Expect statistical reasoning across design, analysis and interpretation.

42. Legacy 6-FRQ habits should be retired

The final 2027 simulation must use 4 FRQs, not 6. Old FRQs remain valuable practice units but should not define final pacing.

43. Legacy topic filter

Before spending review time on an old topic, ask:

Is this still required in the 2026–27 framework?

If not, label it enrichment.

44. Final mixed investigation

Design a complete statistical investigation:

“Do students who use spaced practice score higher on a common test than students who use one long review session?”

Ask the student to:

1. state the investigative question;
2. design sampling/assignment;
3. identify response/explanatory variables;
4. choose summaries;
5. choose inference;
6. state possible limitations;
7. interpret a hypothetical p-value/interval.

45. Delayed transfer

After feedback, wait several days and change the context. If the student can still identify design, inference and interpretation without prompts, the reasoning is more likely to be durable.

46. Final full-exam pacing

MCQ: 42 questions / 90 minutes.

FRQ: 4 questions / 90 minutes.

Build a personal pacing plan and use it repeatedly.

The fully digital interface should be part of every final simulation.

47. Final first-error ledger

48. Final release standard

The revised AP Statistics student is ready when they can move from question to data to analysis to interpretation, select inference from study design, use a current graphing calculator appropriately, type complete FRQ reasoning in Bluebook, and transfer corrections to unfamiliar studies.

Release principle: statistical problem solving first, calculation second, current digital format throughout.

AP Statistics Final Standards Closure | Four-Practice Reasoning, Digital Evidence and Delayed Transfer

The revised AP Statistics course is ready for release when the student can complete the entire statistical problem-solving cycle without being cued by a chapter heading. The student should be able to decide what question is being asked, what data design can answer it, what analysis is appropriate, and what conclusion the evidence can support.

1. Start with parameter language

Before any inference problem, identify the population parameter:

• population proportion (p);
• difference of proportions (p_1-p_2);
• population mean (mu);
• mean difference (mu_d);
• difference of means (mu_1-mu_2).

A correct procedure applied to the wrong parameter is still the wrong analysis.

2. Original parameter task

A school randomly samples 200 students and asks whether they support a new schedule. The target is the proportion of all school students who support the schedule.

Parameter:

[ p= ext{true school-wide support proportion}. ]

The sample proportion (hat p) is the statistic used to estimate (p).

3. Procedure choice should follow design

A compact decision sequence is:

1. categorical or quantitative response?
2. one sample, paired sample or two independent groups?
3. estimate or test?
4. which conditions are required?

Only then choose the calculator procedure.

4. Planted error: two-sample instead of paired

Twenty patients are measured before and after an intervention. A student runs a two-sample t test.

The first invalid decision is treating the observations as independent groups. The data are paired within individuals.

Correct route: analyze the before–after differences.

5. Planted error: causal language

A random sample finds that people who drink more water exercise more often. The student writes, “Drinking water causes exercise.”

Random sampling may help generalize association to a population. Without random assignment, causal inference remains unsupported.

6. Scope-of-inference matrix

7. Conditions should be written for the actual procedure

Do not use one generic “random, normal, independent” paragraph for every problem.

A proportion procedure may need success/failure count conditions. A mean procedure may require a roughly normal sampling distribution or sufficiently large sample. Independence conditions should match the sampling design.

8. Original one-proportion conditions task

Sample size (n=100), sample proportion (hat p=0.62).

Observed successes = 62, failures = 38.

Both are sufficiently large for common normal-approximation inference conditions.

Students should show the counts rather than simply write “normal.”

9. Confidence level versus confidence interval

Increasing confidence level widens the interval when everything else is fixed.

Increasing sample size narrows the interval when everything else is fixed.

Students should explain the tradeoff: more confidence requires a wider net.

10. Margin-of-error design reasoning

If a desired margin of error is cut in half, the needed sample size is roughly multiplied by four because standard error scales with (1/sqrt n).

This is a useful planning relationship even when exact formula details differ by parameter.

11. Hypothesis-test effect of sample size

With larger samples, smaller departures from the null can produce small p-values.

Therefore statistical significance should not be confused with practical significance.

12. Planted error: fail to reject means prove null

A student writes:

“Because (p=0.18>0.05), the null hypothesis is true.”

Incorrect. The evidence was not strong enough to reject the null at the stated significance level. Failure to reject is not proof of truth.

13. Type I error in context

If:

[ H_0: ext{the new treatment has no effect}, ]

a Type I error means concluding there is an effect when in fact there is none.

The student should express the error using the actual context, not only “reject a true null.”

14. Type II error in context

A Type II error means failing to detect a real effect.

Increasing sample size usually increases power and reduces Type II error probability, all else equal.

15. Regression inference should remain conditional on the model

Before interpreting regression, inspect:

• scatterplot form;
• outliers/influential points;
• residual pattern;
• context;
• scope of data.

A high (r^2) cannot rescue a fundamentally inappropriate linear model.

16. Planted error: extrapolation

A regression model was built from ages 12–18 and is used to predict a value at age 80.

The arithmetic may be correct. The inference is unsupported because it is far outside the observed x-range.

17. Original residual task

Observed score = 84. Predicted score = 79.

[ e=84-79=5. ]

The model underpredicted by 5 points.

Students should know sign and meaning, not merely the formula.

18. Digital FRQ evidence rule

A fully digital response should still make the statistical logic recoverable. A useful compact pattern is:

parameter → conditions → procedure → evidence → contextual conclusion.

Typing faster is not a substitute for complete reasoning.

19. Digital notation should stay readable

Students can write “p-hat,” “x-bar,” or use provided symbols according to College Board digital guidance, but the surrounding sentence should remove ambiguity.

For example:

“The sample proportion, (hat p=0.54), estimates the population proportion (p).”

20. Original digital FRQ mini-response

Prompt: A 95% confidence interval for (p_1-p_2) is ((0.03,0.17)).

Strong interpretation:

“We are 95% confident that the true proportion for population 1 exceeds the true proportion for population 2 by between 0.03 and 0.17.”

Because zero is not in the interval, equality of proportions is not plausible at the corresponding two-sided significance level.

21. Fully digital exam execution

Final practice should use:

• 42 MCQ in 90 minutes;
• 4 answer choices;
• shared-prompt question sets;
• 4 typed FRQ in 90 minutes;
• current graphing-calculator policy;
• scratch paper for planning.

22. Shared-prompt annotation

For MCQ sets, write down once:

• population/sample;
• variables;
• study design;
• distribution/model;
• given parameters/statistics.

This prevents inconsistent interpretation across questions using the same prompt.

23. Final practice investigation

Use a new scenario, such as:

“A city wants to know whether a new bus route reduces commute time.”

Require the learner to:

1. define parameter;
2. design a study;
3. state whether causal inference is possible;
4. choose a graph/summary;
5. choose an inference procedure;
6. interpret a hypothetical confidence interval or p-value;
7. state one limitation.

24. Delayed-transfer audit

Several days after feedback, change the population, variable type or design. The learner should still select the right procedure and scope without being told what earlier problem is being tested.

25. Final standards checklist

• five-unit revised framework used;
• legacy removed topics not over-prioritized;
• study design controls scope of inference;
• parameters named before procedures;
• conditions matched to procedure;
• confidence/p-value language calibrated;
• regression claims bounded;
• graphing calculator policy current;
• fully digital FRQ writing practised;
• delayed fresh transfer passed.

26. Release principle

Ask the right question, collect defensible data, analyze with the right model, interpret within the evidence boundary, and communicate the reasoning digitally.

That closes the revised AP Statistics standard for 2027.