Make the Reason for Your Choice Visible
In “Common Errors When Selecting a Procedure,” you practiced auditing the response, parameter, design, and goal before choosing a method. The next skill is to make that reasoning visible in writing. A correct procedure name by itself is not a justification: your reader should be able to see which feature of the question makes the procedure appropriate.
For a mean-inference problem, a useful justification usually takes one or two sentences. Name the specific procedure, identify the parameter or comparison it addresses, and connect the method to the study design. Then mention the relevant conditions using facts supplied in the problem. This is not the same as writing the entire solution: if the question asks for a test or interval, you may need additional work beyond the justification.
The frame is a guide, not a script to repeat mechanically. Include only reasons that apply to the situation. For example, a 10% condition is relevant when sampling without replacement from a finite population; it is not a condition to invent for a randomized experiment. If the problem does not provide enough information to check a condition, say what is missing instead of claiming that the condition is satisfied.
The parameter and design clues come from earlier tutorials such as “Identifying the Parameter in a Mean Problem,” “Paired t Versus Two-Sample t,” and “Matching Procedures to Conditions.” Here, the new task is to compress that reasoning into clear AP-style writing without replacing an explanation with a label.
What a Strong Justification Includes
A strong justification answers three questions: What procedure fits the question? What population quantity does it address? What facts about the data collection and data shape support its use? The wording should be specific enough that it could not be copied unchanged onto a different study with a different design.
Say “one-sample t test,” “paired t interval,” or “unpooled two-sample t test,” as appropriate. Match test versus interval to whether the question asks for evidence or an estimate.
Identify \(\mu\), \(\mu_d\), or \(\mu_1-\mu_2\), and make the context and order clear when comparing means.
Explain whether there is one sample and a fixed benchmark, genuine pairs, or two separate groups. Do not use equal sample sizes or table layout as proof of pairing.
State the relevant random process, independence and 10% condition when applicable, and evidence about the data shape for the chosen t procedure.
Avoid a vague sentence such as “Use t because the data are numerical.” Quantitative data make mean inference a possibility, but they do not distinguish one-sample, paired, and two-sample procedures. Likewise, “the conditions are met” does not tell the reader which conditions you considered or what evidence supports them.
You do not need to repeat every detail in a prompt. Select the facts that justify the chosen method. For a paired t procedure, for instance, describe the genuine pairing and the shape of the pairwise differences; describing the shape of the two original columns separately would not address the data analyzed by that procedure.
Worked Example: A One-Sample t Test Against a Benchmark
A fictional training center randomly selects 16 trainees from 240 eligible trainees. The center records each selected trainee’s completion time, in minutes, for a task. The sample mean is 52 minutes and the sample standard deviation is 8 minutes. A plot shows a roughly symmetric distribution with no apparent outliers. The center wants to know whether the population mean completion time exceeds the 48-minute target. Use \(\alpha=0.05\).
State: Let \(\mu\) be the true mean completion time, in minutes, for the population of trainees represented by the random sample. The hypotheses are \(H_0:\mu=48\) and \(H_a:\mu>48\).
Plan and justify: Use a one-sample t test for \(\mu\), because there is one quantitative sample compared with a fixed 48-minute benchmark, not a second sample. The trainees were randomly selected, \(16\leq0.10(240)=24\) satisfies the 10% condition, and the roughly symmetric times with no apparent outliers support using a t procedure.
That plan is the concise justification in context: it names the procedure and parameter, explains why the benchmark makes this a one-sample problem, and cites evidence for the conditions. If the problem had not described the distribution’s shape, the response should not invent that description.
Do: The standard error and test statistic are
The degrees of freedom are \(16-1=15\). For the right-tailed alternative, the p-value for \(t=2.00\) with 15 degrees of freedom is approximately \(0.0320\), rounded. Since \(0.0320<0.05\), reject \(H_0\).
Conclude: The data provide convincing evidence that the population mean completion time for the trainees represented by the sample exceeds 48 minutes.
Why a shorter answer may be incomplete: “Use a t test because \(n=16\)” names neither the correct kind of t test nor the reason it fits. The benchmark is fixed, so there is one sample; the random selection and distribution description support the conditions.
Tailor the Sentence to the Design
The design clue should change when the design changes. “Two measurements” is not enough: they might be repeated measurements on the same units, deliberately matched observations, or data from independent groups. As “Two-Sample Versus Paired Test on the Same Numbers” explains, the link comes from how the study was carried out.
For paired data, name the difference and its order. Then refer to the sample of differences when describing the shape condition. For two independent groups, state the order of the population means and explain that the groups are unlinked. In both cases, choose a test or interval according to the question’s goal, as practiced in “Interval or Test: What Is the Question Asking.”
Worked Example: A Paired t Interval for Mean Change
A fictional horticulture team randomly selects 12 greenhouse trays from 180 trays. Each tray is measured under both an old and a new watering setting. Define \(d=\text{old-setting water use}-\text{new-setting water use}\), measured in liters per week, so positive differences indicate water savings with the new setting. The differences are roughly symmetric with no apparent outliers. The team wants to estimate the population’s mean water savings.
Identify and justify: Use a paired t interval for \(\mu_d\), the population mean savings per tray, because each tray was measured under both settings and the question asks for an estimate of the mean difference. The trays were randomly selected, \(12\leq0.10(180)=18\) satisfies the 10% condition, and the differences are roughly symmetric with no apparent outliers.
Why this is a complete selection justification: The sentence defines the difference in a stated order, connects the repeated measurements to pairing, chooses an interval because the goal is estimation, and checks the relevant sampling and shape facts. The analysis uses one difference per tray; it does not treat the old-setting and new-setting measurements as two independent samples.
What the justification does not claim: The method estimates an average difference, not the savings for every tray. The sentence also does not claim a treatment effect for all greenhouses unless the sampling method supports generalizing to that population.
Worked Example: A Two-Sample t Test for Independent Groups
In a fictional experiment, 24 volunteers are randomly assigned to use one of two versions of a study app, with 12 volunteers assigned to each version. After two weeks, each volunteer reports one quantitative measure: average focused-work minutes per day. Each volunteer uses only one version, and the groups contain different people with no matching. Plots show roughly symmetric distributions with no apparent outliers in either group. The researchers ask whether the population mean focused-work times differ between the two app versions.
Identify and justify: Use an unpooled two-sample t test for \(\mu_1-\mu_2\), the difference in the population mean focused-work minutes per day for app versions 1 and 2, because the quantitative responses come from two separate, unpaired groups and the question asks whether the means differ. Random assignment supports comparing the treatments, each volunteer contributes one response, and the roughly symmetric distributions with no apparent outliers in both groups support the t procedure.
The first sentence states the method, parameter order, and design clue; the second cites relevant facts about the study and data. A 10% condition is not needed here to justify sampling from a finite population: the volunteers were randomly assigned to treatments, not randomly sampled from a stated finite population. The experiment’s volunteer recruitment also means that any conclusion about cause and effect from assignment should not automatically be treated as a generalization to all students.
Why the alternatives do not fit: There is no fixed benchmark, so this is not a one-sample test. The volunteers are not measured under both versions and are not deliberately matched, so there are no pairwise differences for a paired t test. Equal group sizes do not create pairs.
Common Mistakes and AP Exam Tips
- Giving only the name: “Use a paired t test” does not explain why. Add the design fact: the same units were measured twice or distinct units were deliberately matched.
- Using the wrong target: For paired data, identify \(\mu_d\) and define \(d\). For two independent groups, identify \(\mu_1-\mu_2\) and state the group order. A method name without its target can leave the comparison ambiguous.
- Mixing up a benchmark and a group: A fixed value such as a target time is not a sample of observations. Explain that one population mean is being compared with the benchmark.
- Writing “random” without saying what was randomized: Random sampling and random assignment serve different purposes. Say which occurred; do not claim random sampling if the prompt describes only treatment assignment.
- Checking the wrong shape: For paired t inference, discuss the differences. For a two-sample t procedure, discuss each group separately. For one-sample t inference, discuss the individual measurements.
- Claiming conditions without evidence: “All conditions are met” is not supported if the problem gives no information about random selection, assignment, or shape. State what is known and identify what information would be needed.
- Writing a conclusion instead of a justification: “The data provide convincing evidence…” belongs in the conclusion after the test result. A procedure justification explains why the selected method fits before interpreting the result.
- Adding irrelevant details to sound complete: Keep the justification focused. Mention the relevant parameter, design, and conditions; do not restate every number if it does not support the choice.
A useful final check is to ask whether your justification would still make sense if the sample sizes changed. If your reason for calling data paired is “both groups have 12 observations,” the reasoning is not design-based. If the reason is “each selected tray was measured under both settings,” the pairing is clear regardless of sample size.
Check Your Understanding
For each situation, write a one- or two-sentence procedure justification. Include the parameter or comparison, the design clue, and relevant condition evidence that is given.
- A random sample of 20 bottles is drawn from 300 bottles. Their fill volumes are quantitative, the distribution is roughly symmetric with no apparent outliers, and the question asks whether the population mean differs from a fixed 500-milliliter target. What procedure fits, and what facts should the justification mention?
- The same 15 cyclists are timed on two course surfaces. The question asks for an estimate of the population mean difference in times. What should \(d\) represent, and which data shape should be described?
- Two unlinked groups of 14 randomly selected community gardens each are compared on a quantitative measure of weekly vegetable yield. What procedure fits an estimation question, and why does equal sample size not imply pairing?
- A prompt says that a test uses a random sample but provides no plot, summary of shape, or statement about outliers. What can your justification say about random selection, and what should it avoid claiming about shape?
- In a randomized experiment, 30 participants are assigned to one of two treatments, with each participant receiving only one treatment. Which design fact supports a two-sample t procedure rather than a paired t procedure?