Skip to content

AP Statistics · technique

How to earn full credit for your work

On this page · 5 sections

Showing work is what turns a correct number into points on Section II, the free-response half of the AP Statistics exam. Section II has four questions in 90 minutes, each worth 10 points, and it counts for 50% of your score. The multiple-choice half only checks your final letter, so everything in this guide is about the free-response half.

Readers score each part against a short list of required elements. A typical list asks for the right method, the setup with numbers, the correct value, and an interpretation in context. You earn each element separately, so a wrong final number can still collect most of a part's points if the reader can follow your method. The reverse is also true. A correct answer with no visible method usually earns little or no credit for that part, because readers score the method, not just the final number.

Statistics readers care about communication more than graders in most math courses. A calculator command like normalcdf(2.32, 1000, 0, 1) is calculator syntax, not statistics. The reader wants to see which distribution you used, its mean and standard deviation, and the event you found the probability of. Every conclusion must also name the real variable and population from the problem, such as "the proportion of all city residents who ride the bus", rather than "the proportion" or "p".

The steps

  1. 1

    1. Read the whole question before writing

    Spend the first minute reading every part, because later parts often reuse an earlier answer. Underline the context words you will need again: the population, the variable, and the units. For example, mark "adult residents of Linden" and "rides public transit at least once a week". Then underline the exact verb. "Identify" needs a name, "calculate" needs a number with work, "interpret" needs a sentence in context, and "justify" needs a reason tied to a number.

  2. 2

    2. Name the method and define your symbols

    Write the name of the procedure in words, such as "one-sample z-test for a population proportion" or "two-sample t-interval for a difference in means". Then define every symbol you will use, in context. Write "p = the true proportion of all adult Linden residents who ride transit weekly", not just "p = proportion". For a test, state both hypotheses with the parameter symbol, never the sample statistic. H0: p = 0.40 is correct, but H0: p̂ = 0.40 loses the point, because hypotheses are claims about the population.

  3. 3

    3. Check every condition with the actual numbers

    Naming a condition earns nothing on its own, so show why each one is met. For the random condition, point to the sentence in the stem that says the sample or the treatment assignment was random. For the 10% condition, show that the sample is less than 10% of the population, such as 250 < 0.10(80,000) = 8,000. You need the 10% check only when you sample without replacement from a population, so skip it for a randomized experiment. For large counts in a one-sample test for a proportion, plug in the null value: np0 = 250(0.40) = 100 and n(1 − p0) = 150, both at least 10. For a t-procedure with a small sample, say what a graph of the data shows, such as no strong skew and no outliers.

  4. 4

    4. Show the setup, then the number

    Write the formula with the numbers already substituted, such as z = (0.472 − 0.40) / √(0.40 × 0.60 / 250). You do not need to show the arithmetic after that. If you use a calculator function, label each input in plain statistical notation instead of writing the command alone. For example, write "Normal distribution, mean 0.40, SD 0.031, P(p̂ ≥ 0.472)". Keep three or four decimal places in the middle of the work, so rounding does not change your final answer.

  5. 5

    5. Link the number to the decision

    A conclusion has three pieces, and readers look for all of them. First, compare the p-value to α out loud, such as "Because 0.010 < 0.05, we reject H0". Second, state what that means about the parameter, using the alternative hypothesis. Third, put it in context, naming the population and variable. Use the language of evidence. Say "convincing evidence" or "not convincing evidence", never "proves", and never "accept H0". A large p-value means you fail to reject H0, which is not the same as showing H0 is true.

  6. 6

    6. Answer the exact question, then stop

    Reread the question and make sure your last sentence answers it directly. If it asked "Does the interval support the claim?", say yes or no and point to whether the claimed value is inside the interval. Do not add a second method "just in case". If you write two answers that disagree, readers will not pick the better one for you, so cross out the one you do not want scored. Leave no part blank. A later part that uses your wrong earlier answer correctly can still earn its points.

  7. 7

    7. Use the same habits on the non-inference questions

    Only one of the four Section II questions is a full inference question. The other three mix topics such as exploring data, collecting data, probability and regression, and they reward the same habits. Name the tool you used, show the setup before the number, and finish with a sentence in context that uses the units. The worked section below shows what full-credit work looks like for a probability calculation, a distribution description, and a regression interpretation.

Worked through

THE QUESTION (invented for practice). Linden's transit office says that 40% of the city's 80,000 adult residents ride public transit at least once a week. A planner thinks the true share is higher. She takes a random sample of 250 adult residents and finds that 118 ride transit weekly. Do the data give convincing evidence, at α = 0.05, that more than 40% of Linden adults ride transit weekly?

STEP 1: STATE. Let p = the true proportion of all adult Linden residents who ride public transit at least once a week. H0: p = 0.40 Ha: p > 0.40 The alternative is "greater than" because the planner suspects the share is higher. It is not "not equal", since she only asked about an increase.

STEP 2: PLAN. Method: a one-sample z-test for a population proportion. Conditions: - Random: the stem says the planner took a random sample of adults. - 10%: 250 is less than 10% of all adult residents, because 0.10(80,000) = 8,000. - Large counts: np0 = 250(0.40) = 100 and n(1 − p0) = 250(0.60) = 150. Both are at least 10, so the sampling distribution of p̂ is approximately Normal. Note that this check uses p0 = 0.40, not p̂. A test assumes H0 is true, so the counts come from the null value.

STEP 3: DO. The sample proportion is p̂ = 118/250 = 0.472. The standard deviation of p̂ under H0 is √(0.40 × 0.60 / 250) ≈ 0.0310. z = (0.472 − 0.40) / 0.0310 ≈ 2.32 p-value = P(Z ≥ 2.32) ≈ 0.010 In words: if the true share really were 40%, a random sample of 250 would give a p̂ of 0.472 or higher only about 1% of the time.

STEP 4: CONCLUDE. Because the p-value of 0.010 is less than α = 0.05, we reject H0. There is convincing evidence that more than 40% of all adult Linden residents ride public transit at least once a week.

IF THE QUESTION HAD ASKED FOR AN INTERVAL. A 95% one-sample z-interval uses p̂ in the standard error: 0.472 ± 1.96√(0.472 × 0.528 / 250) ≈ 0.472 ± 0.062. A full-credit interpretation reads: "We are 95% confident that the interval from 0.410 to 0.534 captures the true proportion of all adult Linden residents who ride public transit at least once a week."

WHY EACH LINE IS THERE. The definition of p names the population and the variable, so the reader knows what the hypotheses are about. Each condition shows a number or points to the stem, which proves you checked it rather than recited it. The z formula has the numbers substituted, so the reader can follow your method even if you mistype into the calculator. The sentence after the p-value shows you know what a p-value means, which some parts ask for directly. The conclusion compares to α, states a decision, and uses context, which covers all three pieces readers look for.

WHAT A WEAKER ANSWER LOOKS LIKE. "1-PropZTest: z = 2.32, p = 0.010. Reject H0, so the claim is wrong." This answer has the right numbers, but it is missing most of the elements readers look for. It never defines p or states hypotheses, and it skips the conditions entirely. It also says the claim is "wrong", which is proof language, and it never names Linden adults or transit riding.

FULL-CREDIT WORK ON THE OTHER QUESTION TYPES.

(a) Probability. Suppose 20 Linden adults are chosen at random and the true share of weekly riders is 0.40. Find the probability that at least 10 of them ride weekly. Full-credit work: "Let X = the number of the 20 randomly chosen adults who ride transit weekly. X is binomial with n = 20 and p = 0.40. P(X ≥ 10) = 1 − P(X ≤ 9) ≈ 0.245." The reader sees the distribution, X defined in context, both parameters, and the event, all before the number.

(b) Describing a distribution. A full answer covers shape, center, variability and outliers, with values and context. For example: "The commute times of the 40 sampled residents are skewed right, with a median of 24 minutes and an IQR of 15 minutes (Q1 = 17, Q3 = 32). The 95-minute commute is an outlier, because it is above Q3 + 1.5(IQR) = 54.5 minutes." When you compare two groups, use comparison words: "The median commute for bus riders (31 minutes) is greater than the median for drivers (22 minutes), and bus riders' commutes vary more (IQR of 18 minutes versus 9 minutes)." Listing the two medians side by side without "greater than" does not make a comparison.

(c) Regression. Suppose the least-squares line is predicted rides = 6.8 − 0.35(miles from downtown). A full slope interpretation reads: "For each additional mile a resident lives from downtown, the predicted number of weekly transit rides decreases by 0.35, on average." A residual is actual minus predicted. A resident 4 miles out has predicted rides = 6.8 − 0.35(4) = 5.4, so if she actually takes 7.5 rides, her residual is 7.5 − 5.4 = 2.1. In context: "This resident takes 2.1 more weekly rides than the line predicts for someone who lives 4 miles from downtown." Leaving out "predicted" or "on average" makes the slope sound like a fixed rule for every person, which it is not.

Common mistakes

  • Writing only calculator syntax, such as 1-PropZTest or normalcdf(0.472, 1, 0.40, 0.031), with no named distribution, labeled mean and standard deviation, or event. Readers do not accept calculator commands as statistical work.

  • Defining the parameter without context, or writing hypotheses about the sample statistic. H0: p̂ = 0.40 is wrong, because hypotheses describe the population parameter p, and p must be defined as the proportion of a named population.

  • Listing conditions by name ("random, 10%, large counts") without showing why each holds. Each one needs a number or a reference to the stem. In a one-sample z-test for a proportion, the large counts check uses p0, not p̂. In a two-sample test for p1 − p2, it uses the pooled proportion p̂c with each sample size.

  • Ending with a conclusion that has no context, such as "Reject H0", or using proof language such as "this proves the mean increased" or "we accept H0". Name the population and variable and use the words "convincing evidence".

  • Interpreting a p-value or confidence level backwards, such as saying a p-value of 0.010 means a 1% chance H0 is true. A p-value is the probability of a result at least this extreme, assuming H0 is true. A 95% confidence level does not mean a 95% chance that one interval holds the parameter. It means that if you took many random samples and built an interval from each, about 95% of those intervals would capture the true value.

  • Leaving a part blank because an earlier part went wrong, or writing two different answers to one part. Readers can credit a later part that correctly uses your wrong earlier value, but they will not choose between two answers you leave uncrossed.

Checklist

  • I named the procedure in words, such as "one-sample z-test for a population proportion", not just a calculator menu item.
  • I defined every parameter in context, naming the population and the variable, and my hypotheses use the parameter symbol.
  • Each condition I checked shows a number or points to a sentence in the stem.
  • For a one-sample test on a proportion, my large counts check uses the null value p0.
  • Every calculation shows the formula with numbers substituted, or a named distribution with its labeled mean, standard deviation and event.
  • My conclusion compares the p-value to α, states a decision about H0, and answers the question in context with the language of evidence.
  • My interval interpretation says "we are __% confident" the interval captures the true parameter of the named population.
  • For probability, I named the distribution, defined X with n and p in context, and wrote the event, such as P(X ≥ 10), before the number.
  • When describing or comparing distributions, I gave shape, center, variability and outliers with values, and used comparison words such as "greater than".
  • My slope and residual interpretations use context, units, and the words "predicted" or "on average".
  • I never wrote "proves", "accept H0" or a context-free "the proportion" in a final answer.
  • Every part has an answer, and no part has two competing answers left uncrossed.

More AP Statistics technique