Turn a Long Prompt Into a Set of Answerable Tasks
A multi-part free-response question may move from identifying a method to calculating a result and interpreting it. The challenge is not only knowing statistics; it is noticing exactly what each part asks and making each requested piece of reasoning visible. A correct calculation can still leave credit unearned if the response omits a condition, skips an interpretation, or answers a related question instead of the one asked.
Use the habits from “Writing a Complete Inference Response” and “Selecting Evidence for a Written Conclusion,” but apply them part by part. First identify each task. Then match the task to evidence, calculation, or explanation. Finally, check that the response addresses every part, including any small sub-questions embedded in a sentence.
A Part-by-Part Planning Routine
Before doing arithmetic, scan the entire prompt once. Mark the part labels and circle command words such as identify, calculate, interpret, compare, justify, and conclude. These words signal different kinds of work. “Calculate” calls for a result and enough setup to make it clear what was calculated. “Interpret” calls for a contextual meaning, often with units. “Justify” asks you to connect a choice or conclusion to relevant evidence or conditions.
Next, make a compact part map. It can be a few notes in the margin: what the part asks, the information it uses, and what you need to write. Notice whether a later part uses a result from an earlier part. If it does, label that connection, but do not assume every part depends on the one before it. Some questions deliberately test separate skills using the same scenario.
| Prompt signal | Planning question | What a complete response usually makes explicit |
|---|---|---|
| Identify or name | What object, method, variable, or feature is requested? | The requested name plus a brief reason when the prompt asks for one. |
| Calculate or determine | What quantity is wanted, and which inputs define it? | The setup, numerical result, and appropriate rounding. |
| Interpret or describe | What does the result mean in this setting? | The quantity’s meaning, context, and units when relevant. |
| Justify or explain | What evidence supports the choice or claim? | A direct link between the evidence and the requested claim. |
| Conclude | What answer follows, and how far can it reach? | A direct contextual conclusion with appropriate scope. |
A useful response plan is one part, one visible answer. That does not mean every part needs a long paragraph. It means the reader should not have to hunt through unrelated work to find the requested answer. Use the part label, a short sentence, or a clearly separated calculation to make the match obvious.
Notice all parts, shared information, and any later request that depends on an earlier result.
Mark whether it asks you to select, calculate, interpret, justify, compare, or conclude.
Choose the relevant variable, procedure, graph, statistic, or contextual fact. Do not include evidence that does not help answer that part.
Show the setup and reasoning the task calls for, then check that every requested item has an explicit answer.
For an inference part, the four-part structure from “Writing a Complete Inference Response”—State, Plan, Do, Conclude—still applies. It need not be repeated in full for unrelated parts. Instead, place those elements where they answer the relevant request. A part asking only for a conclusion does not require you to copy the entire calculation again, but it does require a conclusion in context that agrees with the result.
Worked Example: Map an Inference Question Before Solving It
Scenario. A fictional community center takes a random sample of 120 of its 1,500 adult members. In the sample, 78 members say they would use a proposed evening class. The question has three parts: (a) define the population parameter and state hypotheses to test whether more than 60% of adult members would use the class; (b) check conditions and carry out the test at significance level \(0.05\); and (c) state a conclusion in context.
Map the parts. Part (a) asks for a parameter and hypotheses. Part (b) asks for both a procedure and its conditions, then the calculation. Part (c) asks for the decision and a contextual conclusion. Because part (c) uses the test result, keep that result available, but answer each part under its own label.
(a) State. Let \(p\) be the proportion of all adult members of this community center who would use the proposed evening class. The hypotheses are \(H_0:p=0.60\) and \(H_a:p>0.60\).
(b) Plan and check conditions. Use a one-proportion \(z\) test. The problem states that the sample is random. The 10% condition holds because \(120\leq0.10(1500)=150\). Under the null hypothesis, the Large Counts condition holds: \(np_0=120(0.60)=72\geq10\) and \(n(1-p_0)=120(0.40)=48\geq10\).
(b) Do. The sample proportion is \(\hat p=78/120=0.65\). The test statistic is
For the upper-tail alternative, the p-value is approximately \(0.1318\), rounded to four decimal places.
(c) Conclude. Since \(0.1318>0.05\), fail to reject \(H_0\). The sample does not provide convincing evidence that more than 60% of the community center’s adult members would use the proposed evening class.
Why this earns credit efficiently. The parameter and hypotheses answer part (a); the named method and explicit conditions answer the planning request in part (b); and the statistic, p-value, decision, and contextual conclusion complete the remaining requests. Repeating the definition of \(p\) in every part would add length without adding a needed answer.
Let the Prompt Determine Which Evidence You Use
When parts share a setting, they may ask about different features of the same information. A regression question might ask for a description of the association in one part, a slope interpretation in another, and a limitation in a third. Do not reuse one sentence as if it answered all three. Match each statistic or graph to the claim it actually supports, as emphasized in “Selecting Evidence for a Written Conclusion.”
The same discipline applies when a prompt supplies calculator output. The output can provide numbers, but your part map should still identify what each number is for and what explanation remains your responsibility. “Using Calculator Technology Effectively” discusses why reporting a command or display alone is not a complete statistical answer.
Worked Example: Match Regression Output to Each Part
Scenario. A fictional class records the number of minutes, \(x\), each student spends reviewing instructional videos and the student’s quiz score, \(y\), in points. For these students, a least-squares regression output gives \(\hat y=42+1.8x\), \(r=0.84\), \(r^2=0.7056\), and \(s=5.2\) points. The observed review times range from 5 to 25 minutes. A residual plot shows points scattered around zero without a clear pattern. The question asks: (a) interpret the slope; (b) explain \(r^2\) in context; (c) predict the score for a student who reviewed for 20 minutes; and (d) explain whether the model supports a prediction for 30 minutes.
Map the parts. Part (a) asks for the slope’s meaning and units. Part (b) asks what proportion of variation is accounted for in context. Part (c) asks for a fitted value. Part (d) asks about the scope of a prediction, so compare 30 minutes with the observed range. The residual-plot description supports a comment about fit, but it does not make an out-of-range prediction reliable.
(a) Interpret the slope. For each additional minute spent reviewing videos, the model predicts that quiz score will increase by 1.8 points, on average, for students represented by these data. The units are quiz-score points per minute of review.
(b) Interpret \(r^2\). About \(70.56\%\) of the variation in quiz scores among these students is accounted for by the linear relationship between quiz score and video-review time. This interpretation is about variation in the response variable, quiz score; it does not say that video review caused 70.56% of scores.
(c) Calculate the prediction. Substitute \(x=20\) into the fitted line:
The predicted quiz score for a student who reviewed for 20 minutes is 78 points. Since 20 minutes lies within the observed range of 5 to 25 minutes, this is an interpolation.
(d) Assess the 30-minute prediction. Thirty minutes is beyond the largest observed review time of 25 minutes, so using the line there would be extrapolation. The model’s output does not, by itself, justify trusting that prediction. Although the stated residual plot shows no clear pattern within the observed data, it cannot establish that the linear relationship continues beyond the observed range.
Why the parts need separate answers. The slope, \(r^2\), fitted value, and observed range answer different questions. A response that gives only the regression equation does not interpret the slope or \(r^2\), and a good fit within the observed range does not remove the extrapolation concern.
Show the Model and the Event in Probability Parts
A multipart probability question often builds from identifying a model to finding a probability and then interpreting it. Keep the setup visible: define the random variable, identify the event in the wording, and show how the chosen calculation matches that event. “Approaching Probability and Random Variable Question Sets” and “Using Calculator Technology Effectively” cover those tools; here the emphasis is on ensuring each part receives its own answer.
Worked Example: Keep a Binomial Question in Order
Scenario. Suppose each visitor to a fictional museum kiosk independently chooses an audio guide with probability \(0.25\). For a group of 8 visitors, the question asks: (a) define a random variable and explain why a binomial model applies; (b) find the probability that at least 3 visitors choose an audio guide; and (c) give the expected number of visitors in such a group who choose one.
Map the parts. Part (a) is a definition and model justification. Part (b) is an event-probability calculation, with “at least” requiring careful attention to the endpoint. Part (c) asks for the model’s expected count, not the probability from part (b).
(a) Define and justify. Let \(X\) be the number of the 8 visitors who choose an audio guide. A binomial model applies because there is a fixed number of trials, \(n=8\); each visitor has two relevant outcomes, chooses or does not choose; the probability of choosing is \(0.25\) for each visitor; and the choices are stated to be independent. Thus \(X\) has a binomial distribution with \(n=8\) and \(p=0.25\).
(b) Calculate the probability. “At least 3” means \(P(X\geq3)\). Use the complement:
Under the stated model, the probability that at least 3 of the 8 visitors choose an audio guide is approximately \(0.3215\), or \(32.15\%\).
(c) Find the expected count. For this binomial model, the expected number is \(np=8(0.25)=2\) visitors. This is a long-run average count under the model, not a claim that exactly 2 visitors will choose an audio guide in every group of 8.
Why the part map matters. Part (b) asks about the chance of an event, while part (c) asks for an expected count. Giving \(0.3215\) as the answer to both would confuse a probability with a number of visitors.
Common Mistakes and AP Exam Tips
- Answering the general topic instead of the exact part. A prompt about regression does not mean every part asks for a general description of the relationship. Underline the verb and identify the requested quantity or claim before writing.
- Skipping a small sub-request. “Name the procedure and check conditions” has two requirements. Writing only the procedure leaves the conditions unanswered. Split combined instructions into separate items on your part map.
- Giving a number without its meaning. If a part says “interpret” or “conclude,” a calculator value by itself is not enough. State what it means in the setting and use appropriate units or statistical language.
- Using correct evidence for the wrong claim. For example, \(r^2\) describes the proportion of response variation accounted for by a linear model; it is not the slope, a causal effect, or a prediction error. Name the evidence that directly addresses the part.
- Contradicting an earlier result. If a conclusion depends on a test in a preceding part, carry forward the reported decision and direction accurately. Do not change “fail to reject” into “prove there is no effect.”
- Writing everything in one crowded paragraph. Dense writing makes it difficult to see whether every part was answered. Use part labels, short paragraphs, and a displayed calculation when helpful; organization is a way to make correct reasoning visible.
- Adding unsupported claims. Extra statements can introduce errors. Keep explanations focused, and do not claim causation or broader generalization unless the study design supports it, as discussed in “Making Conclusions Consistent with Study Design.”
A final audit should mirror the prompt. Read part (a), then point to the sentence or calculation that answers it. Repeat for every subpart. Check that each requested condition, unit, interpretation, justification, or conclusion is present. This last pass is especially valuable when one part feels easy: short parts still have specific scoring requirements.
Check Your Understanding
For each situation, decide how you would divide the work before writing a full response.
- A question asks you to name a parameter, check conditions, calculate a test statistic and p-value, and state a conclusion. List the four distinct tasks in the order you would address them.
- A regression prompt asks you to interpret the slope and explain \(r^2\). What does each part require, and why would one interpretation not substitute for the other?
- A binomial question asks for the probability of “at least 4” successes and the expected number of successes. Identify the two different quantities requested and one setup detail you should make visible.
- A later part asks whether a prediction at an explanatory-variable value is trustworthy. What comparison should you make before accepting the predicted value?
- What should your final audit check when a prompt has several lettered parts, including one that asks for a justification?