Skip to content

Free-response practice

FRQ 4

10 points2 original prompts4 scored examples

FRQ 4 is the last of the four questions in Section II, the free-response section, which gives you 90 minutes for all four questions and counts for 50% of your AP score. Each question is worth 10 points, so FRQ 4 alone is worth about 12.5% of the exam, and an even split of your time leaves about 22 minutes for it.

On this page · 6 sections

This question mixes three of the course's skill practices in one scenario. Practice 2 is Data Analysis, where you describe, compare and summarize data from a table, a graph or a list of summary statistics. Practice 3 is Using Probability and Simulation, where you calculate probabilities from a model, a two-way table or a simulation. Practice 4 is Statistical Argumentation, where you use the numbers you found to back a claim and say what the data can and cannot prove.

This question type is new for the May 2027 exam, so no scoring guide from a past exam exists for it yet. The rubric below is a ScoreGrind model rubric, built on how AP Statistics readers score other multi-part questions. Expect the parts to build on each other: part (a) will probably ask you to read the data, the middle parts to find a probability, and the final part to argue for a conclusion. Readers will most likely score each part against a fixed set of points, as they do on the other free-response questions. A correct number with no work, or a correct conclusion with no context, usually earns only part of the credit. You get a graphing calculator, a formula sheet and tables, but calculator syntax such as normalcdf(25, 1000, 21, 3) is not enough on its own, so name the distribution, its parameters and the boundary you used.

The rubric, in plain terms

10 points, earned line by line. Readers score what is on the page, nothing else.

Part (a) — Describe or summarize the data in context2 pts

One point for correctly identifying the key feature the part asks for, such as shape with an outlier check, or the conditional proportions for each group. One point for stating center and variability, or for comparing the groups, using the real variable names and units from the stem.

Part (b) — Calculate and interpret a value from the data2 pts

One point for the requested calculation with the setup shown, such as a z-score formula with numbers substituted or a conditional probability written as a fraction from the table. One point for an interpretation in context: what a z-score measures in standard deviations, or which group a conditional probability is out of, plus the follow-up the part asks for (the outlier fence check, or how the reversed conditional differs).

Part (c) — Probability from a model or table2 pts

One point for identifying the right probability model or the right comparison, with its parameters or the two probabilities written out. One point for the correct value or a correct conclusion that is clearly supported by the work shown.

Part (d) — Multi-step or simulation-based probability2 pts

One point for a correct method, such as using the complement for 'at least one' or counting simulated results at or beyond the observed value. One point for the correct value together with an interpretation that states the assumption behind it, such as 'if the claimed rate were true'.

Part (e) — Argument and scope of conclusion2 pts

One point for a conclusion that is directly linked to the simulated probability and reads it correctly: how often a result this extreme happens if the claim is true, so a small value means the result is unusual under the claim. One point for correctly stating what the data cannot show, with a reason from the context. For cause and effect, saying the study is observational with no random assignment earns the point. For generalizing, saying the data were not randomly selected from the larger group earns the point. Naming a specific confounder or source of bias makes the answer stronger but is not required.

Practice prompts

Prompt 1

Bus route travel times

A city transit agency recorded the travel time, in minutes, for 40 randomly selected weekday trips on bus Route 12, with these summary statistics: minimum 18, Q1 22, median 24, Q3 27, maximum 41, mean 25.3, standard deviation 4.6. The agency also runs a new Express route, and it claims Express travel times are approximately normally distributed with mean 21 minutes and standard deviation 3 minutes.

  1. (a)Describe the distribution of Route 12 travel times. Include shape, center, variability, and whether any trips are outliers by the 1.5 × IQR rule. Show your fence calculations. [2 pts]
  2. (b)One Route 12 trip took 33 minutes. Calculate and interpret the z-score for this trip. Is this trip an outlier by the 1.5 × IQR rule? [2 pts]
  3. (c)Assume the agency's claim about the Express route is true. Find the probability that a randomly selected Express trip takes more than 25 minutes, and show your work. [2 pts]
  4. (d)A rider takes the Express route on 5 days, and you may assume the trips are independent and the agency's claim is true. Find the probability that at least one of the 5 trips takes more than 25 minutes, and interpret it in context. [2 pts]
  5. (e)Over 20 workdays, the rider recorded 6 Express trips that took more than 25 minutes. These 20 trips were all the rider's own 8 a.m. commutes, and they were not randomly selected from all Express trips. A computer simulated 1,000 sets of 20 trips, assuming each trip has the probability from part (c) of exceeding 25 minutes, and only 7 of the 1,000 simulated sets had 6 or more such trips. Does the rider have convincing evidence that the rider's 8 a.m. Express trips take more than 25 minutes more often than the agency's claim implies? Justify your answer using the simulation. Can this result be generalized to all Express trips? Explain. [2 pts]
Prompt 2

Commute mode and lateness

A researcher surveyed a random sample of 250 adult workers in the town of Lakeview, and each worker reported their main way of getting to work and whether they arrive late at least once a week.

Car: 120 workers, 30 late at least once a week. Transit (bus or train): 80 workers, 32 late at least once a week. Bike or walk: 50 workers, 8 late at least once a week. Total: 250 workers, 70 late at least once a week.

A separate town report, based on earlier records, claims that 28% of all Lakeview workers are late at least once a week.

  1. (a)For each commute mode, calculate the proportion of workers who are late at least once a week. Describe the association between commute mode and lateness for this sample. [2 pts]
  2. (b)Calculate the probability that a randomly selected worker from this sample uses transit, given that the worker is late at least once a week. Explain in context how this differs from the probability that a worker is late, given that the worker uses transit. [2 pts]
  3. (c)For a worker chosen at random from this sample, are the events 'uses transit' and 'late at least once a week' independent? Justify your answer using probabilities. [2 pts]
  4. (d)A student simulated 200 random samples of 80 workers, assuming each worker has the town report's 0.28 chance of being late at least once a week. In 3 of the 200 samples, the proportion late was 0.40 or higher. Use the simulation to estimate the probability of getting a sample proportion of 0.40 or higher when the true rate is 0.28, and interpret it in context. [2 pts]
  5. (e)Based on the simulation, is there convincing evidence that Lakeview transit commuters are late at least once a week at a rate higher than 0.28? Can the researcher conclude that taking transit causes workers to be late? Explain both answers. [2 pts]

Scored examples

Two responses per prompt — one near the top of the rubric, one in the middle — with every point accounted for. The mid response earns its score through the mistakes students actually make.

Bus route travel times · high-scoring response10/10

(a) The Route 12 times are skewed right, because the mean (25.3) is above the median (24) and Q3 − median = 3 is bigger than median − Q1 = 2, with a long stretch from 27 up to 41. IQR = 27 − 22 = 5, so 1.5(5) = 7.5, which makes the upper fence 27 + 7.5 = 34.5 and the lower fence 22 − 7.5 = 14.5. The 41-minute trip is above 34.5, so it is a high outlier, and no trip is below 14.5 since the minimum is 18. The median trip took 24 minutes, and the middle 50% of trips took between 22 and 27 minutes.

(b) z = (33 − 25.3)/4.6 ≈ 1.67, so the 33-minute trip is about 1.67 standard deviations longer than the mean Route 12 trip time. It is not an outlier, because 33 is below the upper fence of 34.5.

(c) Let X = Express time, where X ~ N(21, 3). For P(X > 25), z = (25 − 21)/3 = 4/3 ≈ 1.333 and P(Z > 1.333) ≈ 0.091, so there is about a 9.1% chance a trip takes more than 25 minutes.

(d) P(a trip is 25 min or less) = 1 − 0.091 = 0.909, so P(none of 5 is over 25) = 0.909^5 ≈ 0.62 and P(at least one over 25) = 1 − 0.62 = 0.38. If the agency's claim is true and the trips are independent, there is about a 38% chance that at least one of the 5 Express trips takes more than 25 minutes.

(e) Yes. If the agency's claim were true, the simulation shows that getting 6 or more long trips out of 20 happened only 7 times in 1,000, about 0.007, which is very unlikely to happen by chance alone. Since the rider did get 6, this is convincing evidence that the rider's 8 a.m. Express trips go over 25 minutes more often than the claim implies. But the result cannot be generalized to all Express trips, because the 20 trips were not a random sample of Express trips. They were all one rider's 8 a.m. trips, and rush-hour traffic could make those trips slower than Express trips at other times of day.

How it scored

2/2a — Shape is justified with two pieces of evidence, both fences are computed, the outlier is named, and center and spread are given in minutes for Route 12 trips.
2/2b — The z-score setup is shown and interpreted as standard deviations above the mean, and the outlier check uses the fence from (a) correctly.
2/2c — Names the normal model with its mean and SD, shows the boundary of 25, and gets about 0.091 from the unrounded z of 4/3.
2/2d — Uses the complement and independence correctly, the answer 0.38 matches the model, and the interpretation states both assumptions: the claim is true and the trips are independent.
2/2e — Links the conclusion to the 0.007 estimate under the condition 'if the claim were true', and keeps the claim to the rider's 8 a.m. trips, which is what the sample can support. The scope point is earned because the student says the trips were not randomly selected and gives a specific reason from the context, the 8 a.m. rush hour.
Bus route travel times · mid-scoring response5/10

(a) The distribution is skewed right because the mean is bigger than the median. The center is about 24 minutes and the IQR is 5 minutes, and the whiskers go from 18 to 41 so there are no outliers.

(b) z = (33 − 25.3)/4.6 = 1.67, so this trip is in the top 1.67% of trips. Upper fence is 27 + 1.5(5) = 34.5, and 33 < 34.5 so it is not an outlier.

(c) normalcdf(25, 1000, 21, 3) = 0.0912, using a normal with mean 21 and SD 3, so P(X > 25) = 0.0912.

(d) Each trip has a 0.0912 chance, so for 5 trips it is 5 × 0.0912 = 0.456.

(e) Yes, the simulation says only 7 out of 1000, which means there is a 0.7% chance that the agency is right, so the Express route is late more often. It only counts for this rider's trips though, not all Express trips, because the trips weren't a random sample.

How it scored

1/2a — Right skew is identified and center and spread are correct. The outlier point is lost because the student assumed whiskers always reach the minimum and maximum, so the 41-minute trip was never checked against the fence.
1/2b — The calculation point is earned for the z-score, with the setup shown. The interpretation point is lost even though the fence check is correct, because a z-score counts standard deviations above the mean and is not a percent, so 'top 1.67%' is wrong.
2/2c — The calculator command is backed up by naming the normal model, its parameters and the boundary, so full credit.
0/2d — Adding the probabilities five times is not valid, because the same method with 20 trips would give a probability above 1, so the complement method was needed.
1/2e — The scope point is earned because the student says the trips were not a random sample, so the result does not extend to all Express trips. The evidence point is lost because 0.007 is read as the chance the claim is true; it is the chance of 6 or more long trips if the claim is true.
Commute mode and lateness · high-scoring response10/10

(a) Car: 30/120 = 0.25, transit: 32/80 = 0.40, and bike/walk: 8/50 = 0.16. There is an association, because transit commuters in the sample were late the most often (40%), car commuters were in the middle (25%), and bikers and walkers were late the least (16%). If there were no association, these three proportions would be about equal.

(b) P(transit | late) = 32/70 ≈ 0.457, so among the workers who are late at least once a week, about 45.7% use transit. P(late | transit) = 32/80 = 0.40 means that among transit users, 40% are late. The first one starts with the late group and the second starts with the transit group, so the denominators are different (70 vs 80).

(c) Not independent, because P(late) = 70/250 = 0.28 but P(late | transit) = 0.40. Knowing a worker uses transit changes the probability that they are late, so the events are not independent.

(d) The estimated probability is 3/200 = 0.015. If transit commuters really were late at the 0.28 rate, there would be about a 1.5% chance of getting a sample of 80 with 40% or more late, just by chance.

(e) Yes. 0.015 is very small, so the observed 0.40 would rarely happen if the true rate were 0.28, and this is convincing evidence that transit commuters are late more often than 0.28. But we cannot say transit causes lateness, because this was a survey and nobody was assigned to a commute mode, so other things could be different about transit riders.

How it scored

2/2a — All three conditional proportions are correct, and the comparison names the groups and the order of the rates.
2/2b — Correct value with the fraction shown. The explanation clearly says which group each probability starts from.
2/2c — Compares P(late) with P(late | transit) using real numbers and draws the right conclusion.
2/2d — Correct estimate, and the interpretation states the 'if the rate were 0.28' condition and the 'at least 0.40' direction.
2/2e — The evidence conclusion is tied to 0.015 under the 0.28 assumption. Saying this was a survey with no one assigned to a commute mode is the correct reason cause and effect cannot be concluded. Naming a confounder such as distance to work would make it stronger, but it is not required.
Commute mode and lateness · mid-scoring response3/10

(a) Car 25%, transit 40%, bike/walk 16%. Transit riders are late the most and bike/walk the least, so commute mode and lateness are associated.

(b) 32/80 = 0.40, which is the chance a transit rider is late. The other one is the same thing reversed so it is also 0.40.

(c) They are not independent because 40% of transit riders are late, which is a lot more than other people.

(d) 3/200 = 0.015, so there is a 1.5% chance that transit riders are not late more often than everyone else.

(e) Yes, transit makes people late, so Lakeview should run more buses so workers get to work on time.

How it scored

2/2a — All three proportions are correct, and the association is described by comparing the groups.
0/2b — The requested P(transit | late) = 32/70 was never calculated, so the calculation point is lost. The interpretation point is also lost, because the student claims the two conditional probabilities are the same when they have different denominators (70 vs 80).
0/2c — Neither point is earned. Only one probability is given, so the comparison is not set up, and 'a lot more than other people' does not support the conclusion. It needed P(late) = 0.28 next to P(late | transit) = 0.40.
1/2d — The estimate of 0.015 is correct. The interpretation gives the chance the claim is true, which the simulation cannot measure. It should give the chance of 0.40 or more if the rate is 0.28.
0/2e — The answer never refers to the simulation result. It also draws a cause-and-effect conclusion from an observational survey, which is exactly the part that should have been ruled out.

Where points die

Saying there are no outliers because the whiskers run from the minimum to the maximum. On a modified box plot, whiskers stop at the most extreme values inside the fences. When the question asks for the 1.5 × IQR rule, compute Q1 − 1.5 × IQR and Q3 + 1.5 × IQR and compare the minimum and maximum to them. Do not use 1 × IQR or a mean ± SD rule unless the question allows it.

Reading a z-score as a percent, or judging how unusual a value is by its raw distance from the mean or median. A z-score counts standard deviations, so a z of 1.67 means 1.67 SDs above the mean, not the top 1.67%. In a normal model that would sit near the 95th percentile, but the Route 12 data are skewed right, so that percentile reading does not apply to them. A raw gap of 7.7 minutes above the mean means nothing until you divide it by the spread.

Flipping a conditional probability, so P(transit | late) is treated as P(late | transit). The word after 'given' sets your denominator. 'Given late' means divide by the late total (70). 'Given transit' means divide by the transit total (80). Write both fractions out before you interpret either one.

Finding 'at least one' by adding the single-trial probability over and over. Adding fails because the events overlap, and it can push past 1. Find the chance of none, using independence, and subtract it from 1: 1 − (0.909)^5 ≈ 0.38.

Interpreting a simulated probability or p-value as the chance the claim is true. The simulation assumes the claim is true and asks how often a result this extreme happens. Say 'if the rate were 0.28, about 1.5% of samples would give 0.40 or more.' A small value is evidence against the claim, not its probability.

Concluding cause and effect from a survey or other observational study. Only random assignment supports a causal claim. Say the study is observational and that no one was randomly assigned to a commute mode, so cause and effect cannot be concluded. Naming a plausible confounder, such as distance to work, makes the explanation stronger.

Score your own response

0 / 10

Be the harshest reader in the room: award a point only when your response does what the line says, on the page — not what you meant to write.