Use the Four Practices as One Statistical Routine
This final review is about connecting skills, not memorizing a list of formulas. A statistics task usually begins with a question, requires data suited to that question, uses analysis that matches the data, and ends with an interpretation that stays within the evidence. The four statistical practices—Formulate Questions, Collect Data, Analyze Data, and Interpret Results—are useful both for solving problems and for finding where your reasoning needs work.
In “Four Statistical Practices in Context,” you saw how the practices fit together. The course’s unit map in “Mapping Course Topics to Exam Questions” can help you recognize which tools may apply, but it does not predict a fixed exam layout. For a final review, use the practices as a route through each problem: What is being asked? What data and design can answer it? What analysis fits? What conclusion is justified?
The Four Practices: What to Check
Formulate Questions. Identify the population or cases of interest, the variables, and the claim or comparison being investigated. Ask whether the question is descriptive, asks about an association, or seeks evidence of a difference or effect. A vague question makes it difficult to choose a relevant data plan or analysis. In “Formulating Questions and Planning Data Collection,” you practiced letting the intended question guide those choices.
Collect Data. Read how the data were obtained, not only how many observations there are. Ask who could be included, who actually participated, whether a sample was selected randomly, whether treatments were randomly assigned, and whether the groups or observations are independent or paired. Check for plausible sources of bias and confounding. As emphasized in “Identifying Bias and Confounding in Scenarios,” these are different problems and require specific explanations. Random sampling and random assignment also support different kinds of conclusions, as you reviewed in “Making Conclusions Consistent with Study Design.”
Analyze Data. Match the method to the response variable, the data structure, and the question. Graphs and summaries help describe patterns; probability models address chance; inference evaluates evidence about population parameters; regression describes relationships between quantitative variables. As in “Choosing the Correct Inference Procedure,” identify the response and design before selecting an inference method, then check the conditions for that procedure. As in the regression tutorials, choose evidence that directly addresses the claim rather than reporting every available statistic.
Interpret Results. Explain what the analysis says in context, including the relevant population or cases, variables, and units. Distinguish evidence of an association from evidence of causation, and keep generalizations within the population represented by the data. “Interpreting Results Without Overclaiming” and “Writing a Complete Inference Response” provide useful models: answer the question, report the evidence accurately, and make the conclusion’s scope match the study design.
Worked Examples: Follow the Reasoning, Not Just the Topic
Worked Example: Diagnose a Survey Before Interpreting Its Result
Scenario. A town library wants to know whether its members support keeping a study room open later on weekdays. From a membership list of 600 people, staff randomly select 80 and email a survey. Fifty-eight people respond, and 42 of the respondents support the change. Describe what the result shows and identify a limitation.
Formulate Questions. The population of interest is the library’s 600 members. The response is whether a member supports later weekday hours, a categorical variable. The question can be stated as whether more than half of these members support the change.
Collect Data. The 80 people were randomly selected from the membership list, which is a probability-sampling feature. However, only 58 responded. Random selection does not guarantee that respondents represent nonrespondents: people who chose to answer may differ in their views. The survey result therefore has a possible nonresponse-bias limitation.
Analyze Data. Among respondents, the support proportion is \(42/58\approx0.724\), or about 72.4%. The response rate is \(58/80=0.725\), or 72.5%. These calculations use different denominators and answer different questions: one describes support among respondents; the other describes how many selected members replied.
Interpret Results. About 72.4% of the respondents supported later weekday hours. That percentage is not automatically the support percentage among all 600 members. The random selection is helpful, but the nonresponse could affect how well respondents represent the full membership. A careful report would describe the respondent result and note that limitation rather than asserting that 72.4% of all members support the change.
Worked Example: Connect a Group Comparison to Its Design
Scenario. In a hypothetical classroom activity, 80 volunteers are randomly assigned to use either a digital or printed review sheet; each group has 40 students. Twenty-eight digital-sheet students and 24 printed-sheet students complete a practice quiz. Summarize the observed difference and say what the design does—and does not—support.
Formulate Questions. The question is whether completion differed between students assigned to the two review-sheet formats. The response is binary: completed or did not complete. The groups contain different students, so the observed proportions can be compared.
Collect Data. The students volunteered, so they are not a random sample of all students. But the scenario says they were randomly assigned to the two formats. Random assignment makes the groups comparable in expectation and supports a cautious causal interpretation of a difference for these volunteers. It does not make the volunteers representative of all students.
Analyze Data. The digital group’s completion proportion is \(28/40=0.70\). The printed group’s is \(24/40=0.60\). The observed difference, digital minus printed, is \(0.70-0.60=0.10\), or 10 percentage points. This is a descriptive difference in these groups; a claim about a broader population difference would require an appropriate inference procedure and its conditions.
Interpret Results. In this activity, the proportion completing the quiz was 10 percentage points higher among volunteers assigned the digital sheet than among volunteers assigned the printed sheet. Because assignment was random, the comparison can support a causal conclusion about the effect of format for the participating volunteers, with appropriate caution about chance variation. Because participation was voluntary, the result should not be generalized automatically to all students. A strong response keeps the design’s support for causation separate from its limits on generalization.
Worked Example: Turn a Self-Check Into a Study Plan
Scenario. A student scores four practice areas from 0 to 2: Formulate Questions, 2; Collect Data, 1; Analyze Data, 1; Interpret Results, 0. The student has six hours available for final review and wants to allocate time according to missed points.
Find the needs. Each area has a maximum score of 2. The missed points are \(2-2=0\) for Formulate Questions, \(2-1=1\) for Collect Data, \(2-1=1\) for Analyze Data, and \(2-0=2\) for Interpret Results. The total number of missed points is \(0+1+1+2=4\).
Allocate the time. Give each missed point an equal share of the six available hours: \(6/4=1.5\) hours per missed point. Formulate Questions receives 0 hours in this first allocation, Collect Data receives 1.5 hours, Analyze Data receives 1.5 hours, and Interpret Results receives \(2(1.5)=3\) hours. The total is \(0+1.5+1.5+3=6\) hours.
Make the plan active. Use the three hours for Interpret Results to write conclusions from mixed problems and check context, scope, and design. Use the Collect Data block to critique study descriptions for sampling, assignment, bias, and confounding. Use the Analyze Data block to classify practice questions, choose methods, and verify conditions. Finish by redoing missed problems without looking at the solutions, then checking whether the same error returns.
This allocation is a starting point, not a guarantee that every skill requires equal time per missed point. If the student’s errors show a more specific cause—such as confusing random sampling with random assignment—use part of the relevant block to target that cause. Keep time for a final mixed set so the practices are connected rather than reviewed only in isolation.
Short Self-Check: Locate the Stage That Needs Work
Answer each question briefly without notes, then compare your answers with the scoring guide. Score each response 0, 1, or 2: 2 means the reasoning is accurate and specific; 1 means the main idea is present but incomplete or vague; 0 means it is missing or substantially incorrect. This is a diagnostic, not a prediction of an exam score.
Check Your Understanding
For each item, write one or two sentences. Give yourself credit for clear reasoning, not for using a particular phrase.
- A student asks, “Do people like the new school lunch?” Name one way to make the statistical question more specific by identifying a population and a measurable response.
- A researcher randomly selects people for a survey, but many selected people do not reply. Name the data-collection concern and explain why random selection alone does not remove it.
- A question compares completion proportions from two separate groups. What features of the question and data structure should guide the analysis choice?
- A randomized experiment finds a difference between treatment groups, but its participants volunteered. What kind of conclusion may random assignment support, and what limits generalization?
- A response gives a correct p-value but no conclusion. Name one element needed to complete the interpretation in context.
Scoring guide. For item 1, a strong answer names a target group and an observable response, such as the proportion of current students who select “satisfied” on a defined lunch survey. For item 2, identify possible nonresponse bias and explain that respondents may differ from those who did not reply. For item 3, identify a categorical response and two independent groups, then consider a two-proportion \(z\) procedure if the question asks for inference and its conditions are met. For item 4, random assignment can support a causal conclusion for the participants, while volunteering limits generalization to a wider population. For item 5, state the decision or evidence conclusion in context, identifying the population parameter and whether there is convincing evidence for the alternative claim.
Add the five scores for a total from 0 to 10. More important than the total, label each missed point by practice: question, data collection, analysis, or interpretation. If one answer seems to involve two practices, write down both. For example, a correct method paired with an unsupported population claim may show strength in Analyze Data but a gap in Interpret Results.
Build a Final Review Plan That Can Adapt
Use your self-check and recent practice work to choose what to review. A low score points to an area for attention; repeated errors tell you which skill within that area to practice. For example, if your plan identifies a suitable procedure but misses the 10% condition, review condition checks rather than starting the entire unit again. “Checking Conditions for Each Procedure Family” and “Writing a Complete Inference Response” are useful references for that specific gap.
For each missed or uncertain response, note whether the issue began with the question, the data plan, the analysis, or the interpretation. Also record the specific mistake, such as confusing paired and independent data.
Pair each error with an activity that directly addresses it: revise a question, critique a study design, select a procedure and check conditions, or rewrite a conclusion with appropriate scope.
Close your notes and solve a fresh problem. Write why the method fits and what the result means. Then check your work against the relevant tutorial and correct the reasoning, not just the final number.
Try a short mixed set after targeted practice. If the same error remains, give that skill more attention; if it improves, shift time to the next weakest area. Reserve time to review the whole path from question to conclusion.
Make the plan realistic. Short focused sessions with a specific task are more useful than rereading many pages without testing your understanding. Mix earlier topics into practice so you must recognize the response type and data structure, rather than being told which procedure to use. “Approaching a Multi-Part Free-Response Question” and “Time Management on the Digital Exam” can help you practice organizing responses and pacing without treating speed as a substitute for reasoning.
Common Review Mistakes and What Strong Work Shows
- Reviewing only formulas. A formula cannot fix a question that was misunderstood or data that do not support the intended claim. Strong work first identifies the target, the variables, and the design.
- Calling every design issue “bias.” Name the specific concern and explain its possible effect. For example, nonresponse may make respondents differ from selected people who did not reply; confounding offers an alternative explanation for an observed relationship.
- Treating a correct method as a complete response. For inference, include the parameter, procedure, relevant condition checks, calculation, and conclusion. For descriptive analysis, explain what the statistic or graph says in context.
- Confusing generalization with causation. Random sampling supports generalization to the population sampled when the process is sound; random assignment supports a causal comparison. State only the kind of conclusion the design supports.
- Making a study plan too broad. “Review inference” is difficult to act on. “Practice identifying paired data, defining the difference, and checking the distribution of differences” names a task you can complete and evaluate.
- Judging readiness from one score. A short check samples only a few skills. Use it alongside recent work, identify repeated error patterns, and adjust your plan when new practice gives you better evidence.
A clear final response makes the reasoning visible. It identifies what is being asked, uses evidence suited to the data and design, and states a conclusion that does not exceed what the study can support. When you review, look for that complete chain—not simply whether the last number matches.