AP Psychology

1. Research method versus measurement tool

Learning goal: Identify study design from what researchers do.

An experiment manipulates an independent variable and compares outcomes across conditions. Random assignment helps make groups comparable. A questionnaire is a measurement tool, not necessarily the research method: it can be used inside an experiment, correlation study or survey design.

An observational or correlational study measures variation without assigning the exposure. It can establish an association, but unmeasured differences may account for it. Naturalistic observation and case study describe distinct ways of collecting information. Do not infer "experiment" merely because there are two groups.

The 2025 Chief Reader report flags naming a questionnaire or topic instead of the actual design. In a response, identify the method and cite the specific feature that makes it that method.

Try three invented summaries. (a) Researchers randomly assign 60 students to either a quiet room or a noisy room and then compare puzzle scores. This is an experiment: the researchers manipulate the room and assign at random. (b) Researchers ask 200 students how many hours they sleep and also record their quiz grades, with no manipulation. This is a correlational design, and the survey is only the tool used to measure sleep. (c) A researcher watches children at a playground and records what happens, changing nothing. This is naturalistic observation.

Notice that each label is justified by one feature: manipulation with random assignment, measurement of two variables without manipulation, or watching without intervening. In your answer, name the method and quote that feature from the scenario.

Practice the habit on any news headline about a study: find the method before judging the claim.

Worked example

In an original study, participants are randomly assigned to read a concise or detailed product description, then answer recall questions. Name the method.

1. The description style is deliberately manipulated.
2. Participants are randomly assigned to the conditions.
3. This is an experiment. The questions measure the dependent outcome; they do not change the method to a survey.

Mental model: Name the design, then show the action that defines it.

Common trap: Confusing a measurement instrument with a research method.

2. Operational definitions

Learning goal: Turn an abstract idea into a measurable variable.

A construct such as memory, anxiety or aggression is an abstract idea. An operational definition states exactly how it is measured or manipulated in a particular study. It must be observable or quantifiable.

"Memory is the ability to retain information" is a concept definition. "Memory was measured as the number of correctly recalled words from a 20-word list after five minutes" is an operational definition. The second specifies a method and a score.

Different operational definitions can measure different aspects of the same construct. A reaction-time measure and a self-report measure may not be interchangeable. The 2025 report flags broad topic descriptions that never specify the study’s measurable variable.

Practice turning vague terms into measurable ones. "Stress" could become the score on a ten-item questionnaire, heart rate in beats per minute during a task, or the number of times a participant reports feeling overwhelmed in a week. Each is a legitimate operational definition, yet each captures a different part of stress, and a study that uses one cannot claim the others.

Also separate the independent and dependent variables. The independent variable is what the researchers vary or compare, such as sleeping 6 hours or 9 hours. The dependent variable is the measured outcome, such as the number of words recalled. A good operational definition makes it possible for another researcher to repeat the procedure and get comparable data. If you cannot picture how to measure it, the definition is not yet operational. The examples here are invented for teaching.

Always ask whether a reader could repeat the measurement from your sentence alone. If not, add the missing detail, such as the time, the scale or the unit.

Worked example

In an original study, distracting text is defined as a narrative with 3 of 10 sentences containing irrelevant details. State the operational definition.

1. Identify the manipulated variable: distracting content.
2. Specify the procedure: 3 of 10 narrative sentences contain irrelevant details.
3. Do not replace this with a general definition of distraction.

Mental model: A construct is broad; its operational definition is study-specific.

Common trap: Giving a dictionary definition instead of the measurable procedure.

3. Sampling and generalizability

Learning goal: Match a claim to the people actually studied.

Random sampling and random assignment answer different questions. Sampling concerns who enters a study; assignment concerns which condition participants receive. Random assignment does not turn volunteers into a representative population sample.

To evaluate generalizability, compare the participant characteristics and recruitment to the target population. A sample of university students may not support direct generalization to young children or older adults. A very large sample can still be systematically unrepresentative.

The 2025 report flags generalizability statements that ignore participant data or rely only on sample size. Cite a relevant characteristic, then explain how it limits or supports extension. Avoid the blanket rule that all small samples are useless or all diverse samples represent everyone.

Consider an invented study of 80 first-year university students who are paid to test a memory app. The researchers conclude that the app helps "people" remember vocabulary. Ask three questions. Who was in the sample? Only first-year university students who volunteered for payment. Who is the claim about? "People" in general. How big is the gap? Large, because age, education and motivation could differ from other groups.

A strong answer cites the participant detail and states a limit: the results may apply to similar students but not necessarily to older adults or school children. A weak answer says only that 80 is a small sample. If the same study had 8,000 volunteers from one university, the gap would still remain. Sample size reduces random error; it does not fix who was left out.

Write one sentence that names who was studied and one that names who the claim is about, then compare them directly.

Worked example

A memory study uses 60 university volunteers aged 18–22. Can it automatically generalize to all ages?

1. The age range is narrow relative to the proposed population.
2. Volunteer recruitment may differ systematically from the broader population.
3. The evidence is most directly about this sampled group under these conditions.
4. Broader replication can test whether the pattern holds elsewhere.

Mental model: Generalize from participant evidence, not from sample-size slogans.

Common trap: Claiming diversity or random assignment guarantees universal results.

4. Ethics is applied, not named

Learning goal: Identify the guideline actually used and explain its implementation.

Informed consent involves participants receiving relevant information and agreeing voluntarily, subject to applicable research rules. Confidentiality protects identifiable data; anonymity means identity is not linked to responses in the first place. These are not synonyms.

Debriefing explains the study afterward, including relevant deception when used. A right to withdraw is different from compensation or a promise of privacy. A response should not invent an ethical procedure absent from the study summary.

The October 2025 official clarification asks students to identify a described ethical guideline and explain how researchers applied it. Name the guideline, quote or paraphrase the relevant action and connect it to participant protection.

Match actions to guidelines using invented descriptions. A study tells participants about the tasks, the time required and that they may leave at any point, and participants sign a form: that is informed consent plus the right to withdraw. A study replaces names with ID numbers and stores the key in a locked file: that is confidentiality. A study that misled participants about its true aim and then explained it at the end: that is deception followed by debriefing.

For each, a complete answer has two steps. Name the guideline, then describe what the researchers actually did to apply it and how it protects participants. Do not add protections the summary never mentions, such as saying data were destroyed when no destruction is described. Staying close to the text keeps the answer accurate and easy to credit.

If the summary does not state an ethical action, say so rather than guessing, and write only what is described.

Worked example

A summary says participants signed an information-and-consent form before joining and could withdraw. Explain one guideline.

1. Choose informed consent.
2. The form described the study and participants agreed before participation.
3. This connects the named guideline to the actual reported action.
4. Do not add anonymity unless the summary also describes it.

Mental model: Name the protection and its actual implementation.

Common trap: Inventing ethics procedures not present in the study summary.

5. Means, differences and effect direction

Learning goal: Explain what a statistic indicates in context.

A mean summarizes a measured outcome in a group. Reporting two means is not the same as interpreting their difference. State which group scored higher or lower, by how much, and what the score measures.

For percentage outcomes, subtracting percentages gives a difference in percentage points. A relative percentage change uses a baseline denominator. If accuracy is 71% versus 58%, the difference is 13 percentage points, not automatically a 13% relative increase.

A larger observed mean does not establish significance without an appropriate analysis. A statistically significant difference does not guarantee a large or important effect. The 2025 report flags reversing mean differences and defining a mean instead of explaining the observed result.

Work through a small invented case. Group A averages 71% correct on a recall task and group B averages 58%. Group A scored 13 percentage points higher. The relative change from B to A is 13/58, about 22%, which is a different quantity. A sentence that interprets the result in context reads: "Participants in group A recalled more items on average than participants in group B, by 13 percentage points."

Common slips to avoid: reversing which group is higher, repeating the definition of a mean instead of describing the result, and claiming the difference proves the treatment worked when the design was not an experiment. Say what was measured, which direction the difference points and how large it is. If a significance test is reported, note its result separately, because a difference can be statistically significant and still small in practice.

Worked example

Toy group A has mean 8 correct answers and group B has mean 6. Interpret the difference.

1. A’s mean exceeds B’s by 2 correct answers.
2. The observed measure is correct-answer count, not general intelligence.
3. Without uncertainty and a test, do not state that the difference is statistically significant.

Mental model: Describe direction and magnitude using the outcome’s units.

Common trap: Assuming a higher score is better without reading the measured variable.

6. Evidence needs a claim and a mechanism

Learning goal: Use a specific finding to support a defensible psychological argument.

A claim answers the question with a position. A topic description only names the subject. "Audiences affect performance" may be too broad; a defensible claim identifies when or how performance changes.

Evidence is a relevant finding, not merely a description of the study’s method. Cite the provided source and describe a result accurately. Then explain how an appropriate psychological concept connects that result to the claim.

The 2025 report flags absent citations, nonspecific evidence and weak application. A named concept alone is not reasoning: explain its mechanism in the scenario. Use different appropriate concepts when the task requires distinct reasoning, rather than repeating one explanation under new labels.

Build a short argument with an invented example. Claim: students tend to perform worse on a hard puzzle when observed by others. Evidence: in the provided study, participants who solved puzzles with an audience completed fewer puzzles than those who worked alone (describe the actual result from the source). Reasoning: arousal from being watched can interfere with complex tasks, so the audience raised arousal in a way that lowered performance on the difficult puzzle.

Check each part. The claim takes a position. The evidence cites a specific result and not the procedure alone. The reasoning explains how a named concept produces the result in this scenario. If the task asks for two different concepts, give two different mechanisms instead of relabeling the first one. Keep the language careful, with words like suggests or is consistent with, since one study rarely proves a general rule.

Worked example

Toy evidence says familiar-task performance improved with observers, while novel-task performance worsened. Build a bounded claim.

1. Claim: audience presence can improve well-learned task performance but impair difficult new tasks.
2. Evidence: specify the direction of the two reported outcomes and cite the toy source.
3. Reasoning: an audience can heighten arousal; a dominant response may help on familiar tasks and hinder unfamiliar ones.
4. The claim remains tied to these conditions rather than "audiences always help".

Mental model: Claim + cited finding + applied concept = reasoned argument.

Common trap: Treating concept names or recruitment details as evidence.

7. Learning: association versus consequence

Learning goal: Distinguish classical and operant conditioning in a scenario.

Classical conditioning links stimuli: a previously neutral cue becomes associated with a stimulus that already elicits a response. Identify the unconditioned stimulus and response before labelling the conditioned cue and response.

Operant conditioning changes the likelihood of a behavior through consequences. Reinforcement increases behavior; punishment decreases it. Positive means adding a stimulus and negative means removing one. It does not mean good or bad.

Negative reinforcement removes an aversive condition after a behavior, increasing that behavior. Punishment instead decreases behavior. These terms are editorial practice targets grounded in the curriculum, not measured frequency claims from the examiner report.

Try three invented scenarios. (1) A child hears a bell before each snack and soon drools at the bell alone. The snack is the unconditioned stimulus, drooling to snack is the unconditioned response, the bell is the conditioned stimulus, and drooling to the bell is the conditioned response. (2) A student gets praise after speaking in class and starts speaking more. Praise was added and behavior increased: positive reinforcement. (3) A student takes pain relief and the headache stops, so the student takes it more often. A bad condition was removed and behavior increased: negative reinforcement.

Ask two questions each time: is a cue being linked to another cue, or is a behavior followed by a consequence? Then ask whether the behavior rose or fell and whether something was added or removed.

A quick test helps: if a new cue is being paired with something that already triggers a reflex, think classical; if the person is choosing a behavior and then gets a result, think operant. Always check the direction of the behavior change before choosing between reinforcement and punishment, and say whether a stimulus was added or removed.

Worked example

A driver buckles up and the warning beep stops. Buckling becomes more frequent. Classify it.

1. The target behavior is buckling.
2. The behavior increases, so the process is reinforcement.
3. An aversive sound is removed, so it is negative reinforcement.
4. It is not punishment because behavior is not being reduced.

Mental model: Reinforce/inhibit behavior, add/remove consequence.

Common trap: Using positive/negative as good/bad labels.

8. Memory: encoding, storage and retrieval

Learning goal: Locate a memory problem in the process before proposing an explanation.

Encoding transforms experience into a usable memory representation. Storage concerns retention over time. Retrieval accesses retained information. A failed recall can reflect weak encoding or retrieval difficulty; it does not by itself prove the memory never existed.

Working memory temporarily maintains and manipulates information. Long-term memory differs in duration and organization. Rehearsal that processes meaning can support retention better than merely repeating sounds in some contexts, but learning effects depend on task and conditions.

Retrieval cues and context can affect access. Misinformation introduced after an event can influence later reports. Keep the source of information clear: original observation versus subsequent suggestion. This lesson extends the research-method report into curriculum-based editorial practice.

Use a retrieval-failure example. A student studies a vocabulary list for an exam but cannot recall a word in the exam room. Possible explanations differ by stage. Encoding: the student read the list without thinking about meaning. Storage: a long gap, with no review, weakened the trace. Retrieval: the word is stored but the exam question did not provide a cue that matches how it was learned.

Each explanation predicts something different. If a hint such as the first letter suddenly makes the word come back, retrieval is more likely than loss. If the student never attended to the word at all, encoding is likely the issue. For eyewitness examples, check whether a detail came from the original event or from a later question or conversation. This framework keeps answers precise and avoids the idea that memory is a perfect recording.

Name the stage you think failed and give the clue in the scenario that points to it.

Worked example

A participant remembers a list after a category cue but not during free recall. What does this support?

1. The cue changed the retrieval condition.
2. Some information may have been retained but less accessible in free recall.
3. The comparison supports a retrieval explanation, not certainty about every storage process.

Mental model: A failed output is not a complete diagnosis of the internal process.

Common trap: Treating memory as a single yes/no container.

9. Social explanations and alternative causes

Learning goal: Distinguish attribution from evidence of a cause.

Attribution explains behavior in terms of personal dispositions or situations. A person late to class may be careless, but traffic or caregiving duties are alternative explanations. An observed behavior alone rarely identifies one cause.

The fundamental attribution error describes overemphasizing dispositional explanations for others’ behavior while underweighting situations. It is not a statement that personality never matters. Apply the concept to an actual explanation, not merely to any criticism.

Group influence can change behavior through norms, conformity or authority. Do not label every group decision "groupthink"; the concept concerns problematic decision processes such as suppressing dissent, not simple agreement. These are editorial curriculum practice targets.

Test the ideas with invented cases. A coworker misses a meeting and a manager says, "He is lazy." That is a dispositional explanation. Possible situational explanations include a delayed train or an unclear calendar invite. Without more information, the evidence does not decide between them. When we judge our own lateness, we often point to traffic, which shows the same event can receive different explanations depending on who is judged.

For group influence, picture a team where everyone agrees quickly because the leader has spoken and nobody wants to disagree. Agreement alone is not groupthink, but the pattern of stifling doubts and not examining alternatives could be. When you write an answer, tie the term to specific details in the scenario and note where the evidence is limited. Avoid using a concept as a label without showing how the scenario fits it.

Say which explanation the evidence favors, and what extra information would help decide.

Worked example

A manager blames an employee’s character for a missed deadline while ignoring a documented system outage. Apply a concept.

1. The manager emphasizes disposition.
2. The outage is a relevant situational cause.
3. Ignoring the situation illustrates the fundamental attribution error in this scenario.
4. This does not prove all missed deadlines are caused by systems.

Mental model: Use concepts to explain a specific mechanism, not label people.

Common trap: Assuming one observed act reveals a stable trait.

10. A complete article-analysis response

Learning goal: Integrate design, measure, statistics, ethics and limits.

An article-analysis task asks several distinct questions. Read each task verb and supply the requested level: identify a method, describe an operational variable, interpret a statistic, explain applied ethics and evaluate generalizability. Reusing one vague paragraph for all parts leaves gaps.

Build a compact extraction table while reading: who participated, how they were recruited, what was assigned, what was measured, the result direction, the stated ethical action and the conclusion’s target. These are evidence anchors, not invented facts.

Use the current official rubric and source summary. October 2025 clarification changed the ethics wording, so a response must identify a described guideline and explain its implementation. The 2025 report’s observed errors guide practice, but the current rubric determines the required response.

Practice with a skeleton for a full response. For a described study, fill a short table before writing: Method (what the researchers did and why that makes it this design), Variables (independent and dependent, each operationally defined), Sample (who, how recruited), Result (direction and size, with units), Ethics (the named guideline and the action taken), Limits (what the sample or design prevents you from claiming).

Then write one short paragraph per task part, using the table row that matches the verb. A typical task might say identify, describe, explain, and justify. Each verb asks for a different depth, so a single paragraph rarely satisfies all of them. After drafting, reread the scenario once and underline every detail you cited. If a claim in your answer has no underlined support, either find the support or remove the claim. This check is quick and prevents unsupported statements.

Worked example

In an original study, volunteers aged 25–40 are randomly assigned to two practice schedules; correct-solution counts are compared and informed consent is described. Build a response plan.

1. Method: experiment, because practice schedule is manipulated and participants are randomly assigned.
2. Operational outcome: the specified correct-solution count, including timing and scoring rule.
3. Interpret means by group and unit, without inventing significance.
4. Ethics: informed consent linked to the reported consent procedure.
5. Generalization: the narrow volunteer age group limits claims about all ages.

Mental model: Each task verb demands a particular kind of evidence and reasoning.

Common trap: Answering the topic rather than the task.

11. A message crosses a gap: the synapse

Learning goal: Trace a neural signal and predict how a change at the synapse alters the receiving cell.

The National Institute of Neurological Disorders and Stroke (NINDS) describes the neuron as the primary functional unit of the nervous system. Every sensation, movement, thought, memory and feeling results from signals that pass through neurons. A neuron has three basic parts: a cell body, dendrites and an axon. Dendrites extend like tree branches and receive messages. The signal passes through the cell body and travels away down the axon, to another neuron, a muscle cell or a gland.

The synapse is the place where a signal passes from a neuron to another cell. When the signal reaches the end of the axon, it triggers the release of vesicles. The vesicles release chemicals called neurotransmitters into the synaptic cleft. The neurotransmitters cross the gap and attach to receptors on the neighboring cell. Those receptors can change the properties of the receiving cell, and if it is a neuron, the signal can continue.

Neurotransmitters are not all alike. Some are excitatory and make cells more active; others are inhibitory and dampen activity. According to NINDS, glutamate is a major excitatory neurotransmitter, GABA is inhibitory, and dopamine can be either, depending on the situation. Acetylcholine is excitatory and governs muscle contraction.

This gives you a way to reason about drug and disorder questions without memorizing every chemical. Ask three things: which neurotransmitter, is it excitatory or inhibitory, and does the change make more or less of its effect reach the receiving cell? NINDS notes, for example, that drugs that increase GABA levels are used to treat epileptic seizures, which fits a dampening effect on overactive neurons. State your prediction as a direction (more or less likely to fire) and name the step of the pathway that changed.

Worked example

A drug raises GABA levels in the brain. Predict the direction of its effect on neuron activity and name the step of the pathway involved.

1. GABA is an inhibitory neurotransmitter, according to NINDS.
2. More GABA in the cleft means more binding to receptors on the receiving cells.
3. Inhibitory binding dampens the activity of the receiving cell.
4. Prediction: neurons are less likely to fire. This fits the use of such drugs to treat seizures.

Mental model: Name the transmitter, say whether it excites or inhibits, then predict the effect on the receiving cell.

Common trap: Assuming that more of any neurotransmitter always means more activity.

12. Same people over time or different people at once?

Learning goal: Choose between cross-sectional and longitudinal designs and state what each can and cannot show.

A developmental question asks how something changes with age. There are two basic designs. A cross-sectional study compares different people of different ages at one point in time. A longitudinal study follows the same people over time and measures them repeatedly.

The National Institute on Aging describes the difference in the context of its Baltimore Longitudinal Study of Aging (BLSA). Before the BLSA, scientists generally conducted cross-sectional studies, comparing participants in one age group to a different set of people in another age group. The BLSA measures physical and cognitive changes in the same group of participants as they return over time. It is an observational study: researchers measure what happens and do not test an intervention such as a drug or an exercise program.

The designs fit different strengths. A cross-sectional study is quicker, but any difference between the age groups might come from age or from the fact that the groups are different people with different life experiences, so age is confounded with those differences. A longitudinal design lets you look at change within each person, which removes the difference between people. Its cost is time, and participants can drop out.

For a free-response answer, name the design, say what the data compare, and then state one strength and one limitation tied to the example. Do not claim that a cross-sectional study shows that people change with age when the groups differ in other ways. Also check the word "observational": without a manipulated variable and random assignment, it does not support a cause-and-effect conclusion, however long the study runs.

Worked example

A researcher tests memory in a group of 20-year-olds and a different group of 60-year-olds on one day. Name the design and one limit.

1. Different people at different ages, tested at one time: a cross-sectional design.
2. The comparison is between groups, not within each person.
3. Limit: the groups may differ in education, health or experience, not only in age.
4. So the result cannot show that memory changes within a person as they age.

Mental model: Cross-sectional compares different people at one time; longitudinal follows the same people.

Common trap: Treating a cross-sectional age difference as proof that individuals change.

13. Stress, anxiety and a disorder are not the same word

Learning goal: Separate ordinary stress and worry from an anxiety disorder using the criteria NIMH gives.

The National Institute of Mental Health (NIMH) separates three ideas that people often blur. Stress is generally a response to an external cause, such as a big test, and it goes away once the situation is resolved. It can be positive, such as inspiring you to meet a deadline, or negative. Anxiety is the body's reaction to stress and can occur even when there is no current threat. An anxiety disorder involves more than occasional worry: the anxiety does not go away, is felt in many situations and can get worse over time.

NIMH lists the types, including generalized anxiety disorder, panic disorder, social anxiety disorder and phobia-related disorders. It reports that about a third of U.S. adolescents and adults experience an anxiety disorder at some point in their lives. Warning signs include symptoms that interfere with everyday life, cause avoidance, or seem to be always present.

For treatment, NIMH names two main routes, psychotherapy and medication, and notes that many people benefit from both. Cognitive behavioral therapy (CBT) helps people notice automatic ways of thinking that are inaccurate or harmful, question those thoughts and change self-defeating behavior patterns. Acceptance and commitment therapy (ACT) instead encourages nonjudgmental acceptance of thoughts and engagement in meaningful activities. Medication options include SSRI and SNRI antidepressants, which may take several weeks to start working, benzodiazepines, which can lead to tolerance and dependence, and buspirone, which is not a sedative and takes three to four weeks to be fully effective.

In an answer, match evidence to the label. Worry before a test that ends after the test is stress. Worry that persists, spreads across situations and disrupts school or work fits the description of an anxiety disorder, but only a professional can diagnose it. Say that explicitly.

Worked example

A student feels tense for the week before each exam, but the feeling ends after the exam and she functions normally. Another student feels constant worry in many settings and has begun to avoid class. Which fits the NIMH description of stress, and which of an anxiety disorder?

1. Student 1: tension tied to an external cause that ends when it is resolved.
2. That matches the NIMH description of stress.
3. Student 2: worry that persists, appears in many situations and leads to avoidance.
4. That matches the signs NIMH gives for an anxiety disorder, although only a professional can diagnose it.

Mental model: Stress ends with its cause; an anxiety disorder persists, spreads and interferes.

Common trap: Calling normal worry a disorder, or diagnosing from a short description.

14. From a treatment claim to a fair test

Learning goal: Judge a treatment claim by asking what was compared and what the evidence supports.

Psychology questions often present a treatment and ask what a study supports. A strong answer separates the claim from the evidence and says exactly which comparison would be needed. The NIMH pages show how careful wording looks. They describe CBT as a research-supported type of psychotherapy, and they say that different therapies work for different types of people. They do not say any one treatment works for everyone.

Start by naming the claim in plain terms, for example that a therapy reduces anxiety. Then ask what the comparison is. Without a comparison group, a drop in anxiety could come from the passage of time, from other changes in a person's life or from expecting to improve. A fair test compares a treated group with a comparison group, and random assignment makes the groups similar on average, so a difference can be linked to the treatment.

Next, check the time frame. NIMH notes that SSRIs and SNRIs may take several weeks to start working, and that buspirone needs three to four weeks. A study that measures after one week could miss a real effect, and a short measurement does not show how long a benefit lasts. Also check what was measured. A questionnaire score, a clinician rating and avoiding fewer situations are different outcomes.

Finally, word your conclusion to match the design. Use "was associated with" for a comparison without random assignment and "caused" only when the design supports it. Mention one limit, such as a small sample or a short follow-up. This keeps the answer honest and gives a grader the specific reasoning they look for.

Worked example

In an invented study, 30 people complete 6 weeks of therapy and their average anxiety score falls. Can we conclude the therapy caused the drop?

1. There is no comparison group, so we cannot tell what would have happened without therapy.
2. Time, other life changes or expecting to improve could also explain the drop.
3. A fair test would randomly assign people to therapy or a comparison condition.
4. So the data are consistent with an effect but do not show that therapy caused it.

Mental model: Name the claim, require a comparison, check the time frame, and word the conclusion to match the design.

Common trap: Reading a before-after improvement as proof that a treatment worked.