A Decision Flowchart for Mean Inference

From the Question to the Procedure

In “Interval or Test: What Is the Question Asking,” you learned to distinguish estimating a population parameter from evaluating evidence about a claim. Now we can put that decision together with the study design. The aim is to move in order from the response variable to the specific procedure, rather than choosing a test or interval from a keyword alone.

For mean inference, the response must be quantitative: it should be a numerical measurement or count for which a mean makes sense. As covered in “Identifying the Parameter in a Mean Problem,” identify the population parameter and its units. Then determine how the observations are organized: one sample, paired observations, or two independent samples. Finally, use the question’s goal to choose an interval or a test.

Decision rule: Quantitative response and one population mean lead to one-sample t inference. Two linked measurements lead to paired t inference on the differences. Two unlinked groups lead to two-sample t inference for the difference in population means. For each design, an estimation question calls for an interval and a claim-testing question calls for a test.

The Mean-Inference Flowchart

Follow these questions in order. The design determines which population parameter and t procedure fit; the wording of the task determines whether to use the interval or test version of that procedure.

Is the response quantitative, and is the question about a mean?
If yes, continue with mean inference. If the response is categorical or the question targets a proportion, this is not a mean-inference problem.

What is the design-based target?
One sample compared with a fixed value targets one population mean, \(\mu\). Two measurements on the same units, or deliberately matched units, target the mean of the defined pairwise differences, \(\mu_d\). Two unlinked groups target the difference between population means, \(\mu_1-\mu_2\).

Choose the procedure that matches the design.
Use a one-sample t procedure for one sample and \(\mu\); a paired t procedure for one sample of differences and \(\mu_d\); or an unpooled two-sample t procedure for two independent groups and \(\mu_1-\mu_2\).

What does the question ask you to do?
Estimate a parameter or give a plausible range: choose a confidence interval. Assess evidence for a claim: choose a hypothesis test. If the prompt asks for both, answer both.

Check the conditions for that procedure.
Use the study design to assess randomness and independence, check the 10% condition when sampling without replacement, and assess whether the data support t inference. For paired t, assess the differences; for two-sample t, assess each group.

The names “one-sample,” “paired,” and “two-sample” refer to the data used in the inference. In a paired analysis, the two original measurement columns become one sample of differences. The paired t procedure is therefore a one-sample t procedure applied to those differences, not a two-sample procedure on the original columns. This is the design distinction emphasized in “Paired t Versus Two-Sample t.”

Parameter and procedure map:

One sample of quantitative observations: \(\mu\), the true population mean; one-sample t interval or test.
Two linked measurements per pair: \(\mu_d\), the true mean of the differences defined in a stated order; paired t interval or test.
Two independent groups: \(\mu_1-\mu_2\), the difference in true population means; unpooled two-sample t interval or test.

What the Flowchart Does—and Does Not—Decide

The flowchart selects a candidate procedure; it does not replace the conditions check. A random sample or randomized experiment supports the random-design condition, but the source of randomness must be described accurately. When sampling without replacement, check that the sample is no more than 10% of the population. For small samples, assess whether the relevant data are reasonably compatible with t inference: inspect the individual observations for one-sample t, the differences for paired t, and each group separately for two-sample t. These are the considerations developed in “Conditions for a Two-Sample t Test” and “Spotting Paired Designs in Word Problems.”

For a paired procedure, the relevant sample size is the number of pairs, and the relevant distribution is the distribution of the differences. For a two-sample procedure, neither group’s observations should be treated as paired unless the study design creates a genuine one-to-one link. As “Spotting Independent Samples in Word Problems” explains, a table with two columns does not by itself establish pairing.

After identifying the procedure, carry out the analysis using the methods for that procedure. A test requires hypotheses about the population parameter; an interval estimates that parameter. In either case, state the parameter in context and check the appropriate conditions. The examples below show how the flowchart guides those choices.

Worked Examples

Worked Example: One Sample and a Claim About a Mean

A recreation center takes a random sample of 16 members from a population of 500 members. Their mean weekly exercise time is 55 minutes, with a sample standard deviation of 8 minutes. The center wants to know whether there is convincing evidence that the population mean weekly exercise time exceeds 50 minutes. Assume a graph of the sample shows no strong skewness or outliers.

State: The response, weekly exercise time in minutes, is quantitative. Let \(\mu\) be the true mean weekly exercise time for members of this recreation center. The question asks about evidence for an increase above a fixed value, so it calls for a test:

H_0:\mu=50 \qquad H_a:\mu>50

Plan: There is one random sample, and the target is one population mean compared with a fixed value. The flowchart selects a one-sample t test. The sample is random. The 10% condition is met because \(16/500=0.032\), or 3.2%, which is less than 10%. The sample size is small, so the graph matters; the stated absence of strong skewness or outliers supports using t inference.

Do: The standard error is \(s/\sqrt{n}=8/\sqrt{16}=2\) minutes. The test statistic is \(t=(55-50)/2=2.50\), with \(16-1=15\) degrees of freedom. For the right-tailed alternative, the p-value is approximately \(0.01225\), rounded.

t=\frac{\bar{x}-\mu_0}{s/\sqrt{n}} =\frac{55-50}{8/\sqrt{16}} =2.50

Conclude: At \(\alpha=0.05\), \(0.01225<0.05\), so reject \(H_0\). The data provide convincing evidence that the true mean weekly exercise time for members of this recreation center exceeds 50 minutes. The procedure choice follows from one sample and one population mean; the testing goal determines that the test, rather than an interval, is needed.

Worked Example: Repeated Measurements and a Mean Difference

A random sample of 8 adult learners uses a reading app for one week. For each person, the response is the change in minutes spent reading per day, defined as after using the app minus before using the app. The observed differences are \(3, 5, 4, 2, 6, 4, 3,\) and \(5\) minutes. The question asks for a 95% confidence interval for the true mean change. The differences show no strong skewness or outliers.

Identify and plan: Reading time is quantitative. Each learner has two linked measurements, so the target is \(\mu_d\), the true mean after-minus-before change in daily reading time for the population represented by the sample. The question asks for a range of plausible values, so use a paired t interval. The 8 learners were selected randomly. If the population contains at least 80 adult learners, \(8/80=0.10\), so the sample is no more than 10% of that population; a larger population also satisfies the condition. The differences show no strong skewness or outliers, supporting t inference for this small sample.

Calculate: The mean difference is \(\bar{x}_d=32/8=4\) minutes. The sum of squared deviations from 4 is \(12\), so the sample standard deviation of the differences is \(s_d=\sqrt{12/7}\approx1.3093\) minutes. With \(7\) degrees of freedom, the 95% critical value is \(t^*\approx2.3646\). The standard error is \(1.3093/\sqrt{8}\approx0.4629\) minutes, and the margin of error is \(2.3646(0.4629)\approx1.0944\) minutes.

\bar{x}_d\pm t^*\frac{s_d}{\sqrt{n}} =4\pm2.3646\left(\frac{\sqrt{12/7}}{\sqrt{8}}\right) =4\pm1.0944 =(2.9056,\ 5.0944)

Interpret: We are 95% confident that the true mean change in daily reading time for the population represented by these adult learners is between about 2.91 and 5.09 minutes. The after-minus-before order makes positive values represent an increase. Using two independent samples would discard the link between each learner’s measurements.

Worked Example: Two Independent Groups and a Mean Difference

A community garden randomly selects 10 plots using a new watering schedule and 10 different plots using the usual schedule. The response is weekly water use in liters. The sample mean and standard deviation are 18 liters and 3 liters for the new schedule, and 15 liters and 3.5 liters for the usual schedule. The question asks for a 95% confidence interval for the difference in population mean water use, defined as new schedule minus usual schedule. Assume plots were sampled from populations of at least 100 plots per schedule, and each group’s data show no strong skewness or outliers.

Identify and plan: Water use is quantitative. The plots in the two groups are different, and the design describes no matching, so these are two independent samples. Let \(\mu_1\) be the true mean weekly water use for plots using the new schedule and \(\mu_2\) the true mean for plots using the usual schedule. The target is \(\mu_1-\mu_2\), in liters. Because the question asks for an estimate, choose an unpooled two-sample t interval.

Check conditions: Both groups were randomly selected. Each sample has 10 plots out of at least 100, meeting the 10% condition. The stated group distributions support t inference for these small samples. The samples are independent by the design: distinct plots are in the two groups, with no described pairing.

Calculate: The observed difference is \(18-15=3\) liters. The estimated standard error is \(\sqrt{3^2/10+3.5^2/10}=\sqrt{2.125}\approx1.4577\) liters. The Welch degrees of freedom are approximately \(17.6\), and the 95% critical value is approximately \(2.106\). The margin of error is \(2.106(1.4577)\approx3.070\) liters.

(\bar{x}_1-\bar{x}_2)\pm t^* \sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}} =3\pm2.106\sqrt{\frac{3^2}{10}+\frac{3.5^2}{10}} \approx(-0.070,\ 6.070)

Interpret: We are 95% confident that the true difference in mean weekly water use, new schedule minus usual schedule, is between about \(-0.07\) and \(6.07\) liters. The interval estimates the difference, in the order defined. The flowchart selects a two-sample t interval because the groups are independent and the question asks for estimation.

Common Mistakes and AP Exam Tips

Starting with the word “mean” and stopping there: One-sample, paired, and two-sample problems all involve means. Identify the design and parameter before choosing a procedure.
Calling any two columns paired: Pairing comes from the study design, such as repeated measurements on the same people or deliberate matching. A shared table or equal sample sizes do not create pairs.
Using a two-sample procedure for repeated measurements: For paired data, define the difference in a clear order and analyze one sample of differences. Check the distribution of those differences.
Using a paired procedure for unlinked groups: When there is no genuine link between observations in the groups, use an unpooled two-sample t procedure and check each group’s data.
Choosing interval or test from the design: Design selects the parameter and procedure family; the question’s goal selects interval or test. A paired study could ask for either one.
Skipping the conditions after naming a procedure: A procedure name is not a conditions check. Address randomness, independence and the 10% condition when relevant, and the appropriate distribution condition.
Leaving the parameter or units vague: Name the population mean or mean difference in context. For paired or two-sample inference, state the order of subtraction so the sign has a clear meaning.

A concise AP response can make the decision path visible: “The response is quantitative. Because the same units are measured twice, I will analyze the after-minus-before differences; the question asks for an estimate, so I will use a paired t interval.” This identifies the data type, design, parameter, and goal before calculations begin.

Key takeaway: Follow the flowchart in sequence: quantitative response, design-based parameter, matching t procedure, then interval or test according to the question’s goal. Check the selected procedure’s conditions explicitly.

Check Your Understanding

For each situation, identify the design, target parameter, and appropriate form of mean inference. Note which conditions would need to be checked.

A random sample of 24 rechargeable lamps is tested, and the question asks whether their population mean operating time exceeds 30 hours. Which procedure fits, and what parameter does it address?
A researcher measures the same 18 stream sites before and after a cleanup. The question asks for a plausible range for the true mean change in nitrate concentration, defined as after minus before. What procedure fits, and which data should be checked for compatibility with t inference?
Two independent random samples of 12 households compare weekly electricity use under two billing plans. The question asks whether the population means differ. What procedure and parameter fit?
A survey records whether each of 200 students owns a bicycle and asks for the population proportion who do. Should the mean-inference flowchart continue past its first decision?
A prompt asks both for a confidence interval for the mean difference between two independent groups and whether zero is a plausible value. What should the response include?