Skip to content

Free-response practice

FRQ 2

10 points2 original prompts4 scored examples

FRQ 2 is the second of four free-response questions on the AP Statistics exam. The free-response section gives you 90 minutes for all four questions and counts for 50% of your exam score. If you split your time evenly, that leaves about 22 minutes for this question. This guide scores each sample answer out of 10 points, with points assigned to each part, so you can see exactly where credit is won and lost. Treat these scorecards as a study tool, not as the official College Board scoring method. The rubric below uses a typical split of 3, 3, 2 and 2 points, but part values change from question to question. In the ferry question, for example, (a) is worth 2 points, (b) 3, (c) 2 and (d) 3.

On this page · 6 sections

The question targets two skill categories. Practice 3 is using probability and simulation. That means building a probability model, finding a probability with a Normal or binomial distribution, working with a sampling distribution, or designing and reading a simulation. Practice 4 is statistical argumentation. That means using a probability you found to decide whether a claim is believable, explaining a result in context, and describing what an error would mean. Expect the question to open with one or two calculations and finish with a part that asks "Does this give convincing evidence…?" or "Explain…", which leans on your earlier answer.

Readers score your communication as much as your arithmetic. A bare calculator command such as binomcdf(20, 0.0766, 2) does not earn full credit, because it does not tell the reader what the numbers mean. Name the distribution, give its parameters, write the event with the correct boundary, and then give the answer. Every conclusion has to use the context of the problem. If an early part goes wrong, a later part can still earn credit when you use your earlier answer correctly, so never leave the argument part blank.

The rubric, in plain terms

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

Practice 3: set up the probability model3 pts

Name the model and its parameters before you calculate. That means Normal with a mean and SD, binomial with n and p plus why it fits, or x̄ with SD σ/√n and a reason it is Normal. For a simulation, assign digits that match p, say how many digits make one trial, allow repeats, and say what you count. A calculator command alone does not name a model.

Practice 3: write the event and find the probability3 pts

Write the event in symbols with the right boundary, such as P(X ≥ 3) = 1 − P(X ≤ 2), or include every simulated trial at 40 or fewer. Then reach the correct value with work shown: a z-score, labeled inputs, or a count divided by the number of trials. Readers accept a correct calculation that uses your own earlier value.

Practice 4: argue from the probability2 pts

1 point: compare the probability to α or 0.05. Readers accept your own value from an earlier part. 1 point: say the probability was found assuming the claim or the old value is true, then conclude about the parameter in context. Use the language of evidence, never proof.

Practice 4: explain a result in context2 pts

Explain a statistical idea in the problem's words. For a Type I or Type II error, give both halves: what is actually true and what you decide. For a change in sample size, give the direction and the mechanism. Less variability in the statistic is a mechanism; "more power" only renames the effect.

Practice prompts

Prompt 1

Ferry crossing times

A ferry company runs a short commuter route across a harbor. Under normal conditions, the time for one crossing is approximately Normally distributed with mean 42 minutes and standard deviation 3.5 minutes. Assume that crossing times are independent of each other.

  1. (a)Find the probability that a randomly selected crossing takes more than 47 minutes. Show your work. [2 pts]
  2. (b)A commuter takes 20 randomly selected crossings in one month. Find the probability that at least 3 of them take more than 47 minutes. Show your work. [3 pts]
  3. (c)Find the probability that the mean time for 9 randomly selected crossings is greater than 44 minutes. Show your work. [2 pts]
  4. (d)After a new boarding procedure begins, a random sample of 9 crossings has a mean time of 45.1 minutes. Assume the standard deviation is still 3.5 minutes. Does this result give convincing evidence that the mean crossing time is now greater than 42 minutes? Justify your answer using an appropriate probability. [3 pts]
Prompt 2

Tomato seed germination

A seed company claims that 90% of its tomato seeds germinate (sprout). A garden club suspects the true rate is lower. The club plants a random sample of 50 of the company's seeds, and 40 of them germinate.

To judge the claim, a club member runs a simulation. Each trial plants 50 seeds that each germinate with probability 0.90, and records how many germinate. The results of 200 trials are shown below.

Number germinating: 39 or fewer | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 Number of trials: 2 | 3 | 7 | 13 | 22 | 30 | 37 | 36 | 28 | 15 | 6 | 1

  1. (a)Describe how the club member could use a table of random digits to carry out one trial of this simulation. [2 pts]
  2. (b)Use the simulation results to estimate the probability that 40 or fewer of 50 seeds germinate if the company's claim is true. [2 pts]
  3. (c)Using a significance level of α = 0.05, do the club's results give convincing evidence that the true germination rate is less than 90%? Justify your answer using your result from part (b). [2 pts]
  4. (d)Describe a Type II error in the context of this question. [2 pts]
  5. (e)The club is planning a second study with 200 seeds. Suppose the true germination rate is actually 85%. Explain how using 200 seeds instead of 50 would change the probability of a Type II error. [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.

Ferry crossing times · high-scoring response10/10

(a) Let T = time for one crossing. T ~ N(mean 42, SD 3.5). z = (47 − 42)/3.5 = 1.43 P(T > 47) = normalcdf(lower 47, upper 1000, mean 42, SD 3.5) = 0.0766 So about a 7.7% chance that one crossing takes more than 47 minutes.

(b) Let X = number of the 20 crossings that take more than 47 min. Each crossing is either over 47 or not, the crossings are independent, and p is the same each time, so X is binomial with n = 20 and p = 0.0766. P(X ≥ 3) = 1 − P(X ≤ 2) = 1 − binomcdf(n = 20, p = 0.0766, x = 2) = 1 − 0.8058 = 0.1942

(c) The population is Normal, so the sampling distribution of x̄ for n = 9 is Normal with mean 42 and SD = 3.5/√9 = 1.167. z = (44 − 42)/1.167 = 1.71 P(x̄ > 44) = 0.0432

(d) If the mean crossing time were still 42 minutes, x̄ for 9 crossings would be N(42, 1.167). z = (45.1 − 42)/1.167 = 2.66 P(x̄ ≥ 45.1) = 0.0039 So if nothing had changed, there is only about a 0.4% chance of getting a sample mean of 45.1 minutes or more. Because 0.0039 is far below 0.05, this result would be very unusual under the old mean. This is convincing evidence that the mean crossing time with the new boarding procedure is greater than 42 minutes.

How it scored

2/2a — Names the Normal model with its mean and SD, shows the z-score and labeled inputs, and gives the correct right-tail probability of 0.0766.
3/3b — Identifies a binomial count with n = 20 and p = 0.0766 and explains why it fits. Writes the complement with the right boundary, P(X ≥ 3) = 1 − P(X ≤ 2), and gets 0.1942.
2/2c — Uses the sampling distribution of x̄ with SD σ/√n = 1.167 and gives the reason it is Normal. The answer of 0.0432 is correct.
3/3d — Calculates the probability assuming the mean is still 42, compares 0.0039 to 0.05, and concludes about the mean crossing time in context. It speaks of evidence, not proof.
Ferry crossing times · mid-scoring response2/10

(a) z = (47 − 42)/3.5 = 1.43. normalcdf(47, 1000, 42, 3.5) = 0.0766

(b) This is binomial, n = 20, p = .0766. P(X ≥ 3) = 1 − binomcdf(20, .0766, 3) = 1 − .938 = .062

(c) P(x̄ > 44) = normalcdf(44, 1000, 42, 3.5) = 0.284

(d) normalcdf(45.1, 1000, 42, 3.5) = 0.188. Since 0.188 > 0.05 it is not convincing evidence that the mean crossing time increased.

How it scored

1/2a — Earns the answer point because the z-score shows the work behind the correct 0.0766. Loses the model point because it never names the Normal distribution and never writes the event P(T > 47). A z-score and calculator syntax do not replace naming the model.
0/3b — Names binomial with n and p but never says why the binomial model fits, such as independent crossings with the same p each time. The boundary is also wrong: 1 − P(X ≤ 3) gives P(X ≥ 4), not P(X ≥ 3), so the answer of 0.062 is wrong.
0/2c — The question is about the mean of 9 crossings, so the model is the sampling distribution of x̄: Normal with mean 42 and SD 3.5/√9 ≈ 1.167. The inputs normalcdf(44, 1000, 42, 3.5) use the distribution of a single crossing instead, so the model is wrong. Having the right center and tail direction is not enough for the setup point. As in (a), it never names a distribution or writes the event P(x̄ > 44), and the answer of 0.284 is wrong.
1/3d — Earns the comparison point: it compares its own probability to 0.05, and its conclusion follows correctly from 0.188. Loses the second argue point because it never says the probability was found assuming the mean is still 42. Part (d) is scored on its own merits, and using σ = 3.5 instead of σ/√9 makes the probability wrong here too, so the calculation point is also lost.
Tomato seed germination · high-scoring response9/10

(a) Let digits 0–8 stand for a seed that germinates and 9 stand for a seed that does not. That gives a 9/10 = 0.90 chance. Read 50 digits from the table in a row, and allow repeats because each seed is independent. Count how many of the 50 digits are 0–8. That count is the number of seeds that germinated in one trial.

(b) The trials with 40 or fewer are the 2 trials at 39 or fewer plus the 3 trials at 40, so 5 trials. P(40 or fewer) ≈ 5/200 = 0.025

(c) Yes. If the company's 90% claim were true, 40 or fewer germinating out of 50 happened only about 2.5% of the time in the simulation. Since 0.025 < 0.05, the club's result is unusual if the claim is true. So there is convincing evidence that the true germination rate of the company's tomato seeds is less than 90%.

(d) A Type II error would happen if the true germination rate really is less than 90%, but the club does not find convincing evidence of that. The club would fail to reject the company's 90% claim even though the seeds actually germinate less often.

(e) With 200 seeds the probability of a Type II error would be smaller, because a bigger sample gives the test more power.

How it scored

2/2a — Assigns digits that match p = 0.90, says how many digits make one trial, allows repeats with a reason, and says what to count.
2/2b — Counts every trial in the event 40 or fewer, not just the trials at 40, and divides by 200 to get 0.025.
2/2c — Links the estimate to the assumption that the claim is true, compares it to α = 0.05, and concludes about the true germination rate in context.
2/2d — Correctly describes failing to reject a false claim, in the language of seeds and germination.
1/2e — The direction is right, but "more power" just renames the effect. The reason is missing: a larger sample makes the sample proportion vary less, so 85% is easier to tell apart from 90%.
Tomato seed germination · mid-scoring response5/10

(a) Use digits 1–9 for a seed that germinates and 0 for one that doesn't. Pick 50 different digits from the table and count the 1–9s.

(b) 3/200 = 0.015

(c) Yes, 0.015 is less than 0.05 so this proves the company is lying about the 90%.

(d) A Type II error is saying the germination rate is less than 90% when it actually is 90%.

(e) With more seeds there is less variability in the proportion of seeds that germinate, so a true rate of 85% would be easier to tell apart from 90%. That means the chance of a Type II error goes down.

How it scored

1/2a — The digit assignment matches 0.90, which earns one point. Asking for 50 different digits cannot work with only ten digits, and banning repeats breaks the independence between seeds.
1/2b — Earns the calculation point for dividing its own count by the 200 trials. Loses the event point because it counts only the 3 trials at exactly 40. The event is 40 or fewer, so the 2 trials at 39 or fewer are missing, and the answer never reaches 0.025.
1/2c — Earns the comparison point, because readers accept the student's value of 0.015 from (b) and it is correctly compared to 0.05. Loses the second point for three reasons. It never says 0.015 was found assuming the 90% claim is true, it says "proves", and it concludes about the company lying instead of the true germination rate.
0/2d — This describes a Type I error, rejecting a claim that is true. A Type II error is failing to reject a claim that is false.
2/2e — Gives the right direction and the reason for it: less variability in the sample proportion makes a real difference easier to detect.

Where points die

Swapping Type I and Type II errors, especially describing "concluding there is an effect when there isn't" when the question asks for a Type II error. Anchor both errors to the null hypothesis. Type I means you reject a null that is actually true, a false alarm, and its probability is α. Type II means you fail to reject a null that is actually false, a missed effect. Write the answer as two halves, the truth and your decision, both in the problem's words.

Treating a small p-value as proof, or as a sign that the effect is large or important. A small p-value only says the result would be rare if the null were true. Say the result "gives convincing evidence," never that it "proves" anything. A very large study can make a tiny difference significant, such as a 0.5-point gain on a 100-point test, so statistical significance is not the same as practical importance.

Using the population standard deviation σ when the question is about the mean of several observations. The mean of n independent values varies less than one value does. Its standard deviation is σ/√n. For 9 crossings with σ = 3.5, that is 3.5/3 ≈ 1.17. If the question says "the mean of" or "the average of," switch to the sampling distribution before you calculate.

Getting the boundary wrong on "at least" or "or fewer" with a discrete count, for example computing 1 − P(X ≤ 3) for P(X ≥ 3), or counting only the trials at exactly 40 in a simulation. Write the event in symbols before you touch the calculator. For counts, P(X ≥ 3) = 1 − P(X ≤ 2) because the value 3 belongs in the event. In a simulation table, add every bar that fits the event, including any "or fewer" group at the edge.

Using a Normal or binomial model without saying why it fits, such as calling x̄ Normal with no reason or skipping the binomial conditions. Give the reason in one line before you calculate. For x̄, say the population is Normal, or that the sample is large enough for the Central Limit Theorem to apply. For a binomial count, check that there is a fixed number of trials, two outcomes, the same p each time, and independent trials. In the ferry question, x̄ for 9 crossings is Normal only because single crossing times are Normal. If the population were strongly skewed, a sample of 9 would be too small for the Central Limit Theorem, and a Normal model for x̄ would not be justified.

Writing only calculator syntax, such as normalcdf(47, 1000, 42, 3.5), with no labels or named distribution. Readers need to see the model and the event. Name the distribution and its parameters, write the event such as P(T > 47), then show either a z-score or labeled inputs.

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.