Lesson 19: AP FRQ Strategies — The Four-Step ProcessAP

Forty multiple-choice questions test what you know; the six free-response questions test whether you can *communicate* it. AP readers grade thousands of responses against rubrics that award points for specific, nameable moves: correct hypotheses with defined parameters, named procedures, verified conditions, correct mechanics, and conclusions in context that are explicitly linked to the evidence. This lesson teaches the four-step process — State, Plan, Do, Conclude — as the skeleton of every inference response, plus the realities of the Investigative Task, where several topics combine in one unfamiliar scenario.

The four steps, and what each one is worth

• State: the hypotheses (tests) or the parameter of interest (intervals), with the parameter *defined in context*: "p = the true proportion of all students at this school who..." A bare "H0: p = 0.5" without a definition is incomplete.
• Plan: name the procedure precisely ("one-sample t-test for μ"), and verify each condition with evidence — the Random condition tied to the design, the Normal/Large Counts condition *computed* (np0 = 90 ≥ 10), the 10% condition addressed. Naming a procedure with no condition checks leaves the Plan point on the table.
• Do: mechanics — the test statistic or interval, computed correctly, with the p-value or interval reported. Show the formula with numbers substituted; calculator-only answers risk everything on one keystroke error and give the reader nothing to award partial credit to.
• Conclude: a sentence that (1) compares p-value with α (or uses the interval), (2) states the decision about H0, and (3) translates it into the context of the problem, in the direction of the alternative. "Because 0.014 < 0.05, reject H0; there is convincing evidence that the true mean commute time exceeds 30 minutes." Every clause earns its keep.

Worked example 1 (a model response, annotated). *A principal claims at least 80% of parents are satisfied with a new schedule. A random sample of 150 parents finds 108 satisfied. Test at the 5% level.*

• State: H0: p = 0.80; Ha: p < 0.80; p = true proportion of all parents at the school who are satisfied. *(Parameter defined ✓)*
• Plan: One-proportion z-test. Random: random sample stated. Normal: np0 = 150(0.80) = 120 ≥ 10 and n(1−p0) = 30 ≥ 10 ✓. Independent: 150 parents is less than 10% of the parent population. *(All three conditions, with numbers ✓)*
• Do: p̂ = 108/150 = 0.72. SE = √(0.80 × 0.20/150) ≈ 0.0327. z = (0.72 − 0.80)/0.0327 ≈ −2.45. p-value ≈ 0.0072. *(Statistic + p-value ✓)*
• Conclude: Since 0.0072 < 0.05, reject H0. There is convincing evidence that the true proportion of satisfied parents is less than 80%. *(Decision + context + direction ✓)*

That response is short, complete, and scores Essentially correct on every part — length is not the currency; the checklist is.

Worked example 2 (spot the deductions). A student's response to the same problem reads: "H0: p̂ = 0.8. The conditions are fine. z = −2.45, p = 0.0072. Since the p-value is small, we accept that the principal is wrong and the probability the claim is true is 0.72%." Four planted errors, all classics: (1) hypotheses written about p̂ instead of p; (2) "conditions are fine" with no verification; (3) "we accept that..." — tests reject or fail to reject H0, and the conclusion needs a context sentence; (4) the p-value misread as the probability H0 is true. Training your eye on other people's errors is the fastest rubric education there is — the Investigative Task (Question 6) frequently includes exactly this "critique this analysis" genre, alongside multi-part problems that chain design (Unit 3) → probability (Unit 4) → inference (Units 6–9).

Exam logistics that change scores

Section II gives 90 minutes for 6 questions — roughly 13 minutes each for Questions 1–5 and about 25–30 for the Investigative Task, which is worth as much as the other five are weighted to expect depth. Budget accordingly: a flawless Question 1 cannot compensate for a blank Question 6. If stuck on a part, state what you *would* do and move on — later parts often remain accessible, and readers can award method points downstream. Never leave a conclusion blank: a context sentence with your computed numbers is always worth attempting.

Common mistakes (the readers' greatest hits)

• Undefined parameters and hypotheses about statistics.
• Procedures named but conditions asserted ("all conditions met") without verification.
• Conclusions with no linkage — no comparison to α, or a decision floating free of the context ("Reject H0." — full stop).
• "Accept H0", "prove", "the probability H0 is true is..." — three phrases that each cost the conclusion point on sight.
• Answering a different question: a test when an interval was requested, a one-sided alternative when the prompt says "differs".

AP Exam Experience

Question 1. Which conclusion statement would earn full credit for a test with p-value 0.032 at α = 0.05?

• (A) "The probability that H0 is true is 0.032, so we reject it."
• (B) "Since 0.032 < 0.05, reject H0; there is convincing evidence that the true mean differs from 12."
• (C) "We accept the alternative hypothesis."
• (D) "The data are significant."
• (E) "H0 is false with 95% confidence."

Answer: (B). It contains all three required elements: the linkage (p-value compared with α), the decision, and the conclusion in context in the direction of Ha. Every other option contains a flagged phrase — (A) and (E) misinterpret the p-value/confidence, (C) misuses "accept", (D) has neither linkage nor context.

Question 2. In the Plan step of a one-proportion z-test, which item is REQUIRED for full credit?

• (A) A histogram of the sample data
• (B) Verification of conditions, including the computed check np0 ≥ 10 and n(1−p0) ≥ 10
• (C) The sample size formula
• (D) A statement of the power of the test
• (E) A sketch of the population distribution

Answer: (B). Condition verification with numbers is a scored component of the Plan step. (A) and (E) are neither possible nor required for proportion procedures; (C) and (D) belong to other tasks entirely.

Question 3. A student writes: "Because the p-value is 0.21, we accept H0 and conclude the treatment has no effect." The most serious error is:

• (A) The p-value should be compared with 0.05, not reported alone.
• (B) "Accept H0" and claiming no effect — a large p-value means insufficient evidence, not proof of the null.
• (C) The conclusion lacks a test statistic.
• (D) The student should have used a confidence interval.
• (E) Nothing is wrong.

Answer: (B). With p = 0.21 we *fail to reject* H0; the study may simply lack power. Claiming the treatment "has no effect" asserts exactly what the test cannot establish. Choice (A) notices a real omission but it is secondary to the logical error.

Free-response question. A classmate's solution to a significance test reads: "Hypotheses: H0: p̂ = 0.6, Ha: p̂ > 0.6. I did a z-test. Conditions are all satisfied. I got z = 1.85 and p-value = 0.032. Since the p-value is small there is a 96.8% chance the claim is true, so we accept Ha and the proportion is definitely greater than 0.6." (a) Identify FOUR distinct statistical errors in the solution. (b) Rewrite the conclusion as it should have been written (you may keep z = 1.85 and p-value = 0.032, and take α = 0.05).

Scoring guideline.

• Part (a): Essentially correct if any four of these are identified: (i) hypotheses stated in terms of p̂ — hypotheses concern the parameter p; (ii) the procedure is unnamed ("a z-test" — must be a one-proportion z-test for p); (iii) conditions merely asserted, not verified with the problem's numbers; (iv) the p-value interpreted as a probability about the claim ("96.8% chance the claim is true") — the p-value is a probability about data assuming H0; (v) "we accept Ha" — tests reject or fail to reject H0; (vi) "definitely greater" — conclusions are statements of evidence, not certainty. Partially correct for three errors correctly identified. Incorrect for vague complaints ("the work is sloppy") with no statistical content.
• Part (b): Essentially correct for a conclusion in the form: "Since 0.032 < 0.05, reject H0. There is convincing evidence that the true proportion [in the problem's context] is greater than 0.6." — linkage, decision, context and direction all present. Partially correct if the decision is right but the context or the p-value-to-α comparison is missing.

Deeper dive: how the AP Reading actually works — and what it implies

Each June, thousands of teachers and professors gather to grade FRQs. Before scoring begins, leaders finalise the rubric from sample student work, and every reader is trained and monitored against it; a response earns "Essentially correct" (E), "Partially correct" (P) or "Incorrect" (I) *per part*, and part scores combine into the question score. Two consequences follow for strategy. First, partial credit is structural, not charitable: a response with perfect setup and one arithmetic slip still earns most of the value — so always show setup, formulas with substituted values, and named conditions, because those are literally the scored objects. Second, readers grade what is on the page, not what you meant: undefined symbols, calculator dumps without interpretation, and conclusions that trail off mid-thought cannot be rescued by intent. Write as if a tired expert is hunting for specific evidence of understanding — because one is. The Investigative Task adds a third lesson: its parts are chained but *not* gated — an error in part (b) does not poison part (d) if (d) is done correctly with the given (or your stated) value, so never abandon later parts.

Exam-day card: timing and triage

Section I: 40 multiple-choice questions in 90 minutes — about 2 minutes 15 seconds each; flag and move rather than bleed time, because unanswered questions earn the same zero as wrong ones and there is no penalty for guessing. Section II: 90 minutes for 6 free-response questions — plan roughly 12–13 minutes each for Questions 1–5 and reserve 25–30 minutes for the Investigative Task (Question 6), whose weight justifies the investment. Within a question, work the parts in order but never let one part hostage the rest: if a computation stalls, write the setup ("I would compute t = ... with df = ..."), take the method credit available, and continue — later parts often need only the *conclusion* of the stalled part, which you can state as an assumption and use. Final five minutes: sweep for the three cheapest recoverable points in the booklet — defined parameters, a named procedure, a conclusion sentence in context. They are present or absent on nearly every page you wrote, and each takes seconds to add.