“Correlation does not prove causation” is a useful warning, but a passage question requires a more specific explanation. What alternative could produce the association? Did the proposed cause precede the outcome? Were people assigned to an exposure or did they choose it? What population does the sample represent?
Those questions turn a slogan into a method. Correlation, experimental assignment, and confounding are introduced in OpenStax Psychology. The worked examples below are fictional and designed to make the distinctions visible.
Begin with the group-formation sentence
Look for the sentence that explains how participants entered groups. “Participants chose whether to attend” differs from “participants were randomly assigned.” “Researchers sampled students from a roster” differs from “researchers assigned students to two study conditions.”
Sampling concerns who enters the study. Assignment concerns which condition a participant receives. Random sampling can help a sample represent a defined population. Random assignment balances potential causes across groups in expectation, but a particular random allocation can still leave chance imbalances. Neither phrase alone guarantees a flawless study.
Write these processes separately when reviewing a passage. Otherwise, the word “random” can hide an important design difference.
Original study: optional workshops
A university offers an optional planning workshop. At the end of term, attendees report a mean stress score of 12, compared with 16 among non-attendees. Lower scores indicate less reported stress. The survey is completed once, after the workshop period. The study records no pre-workshop stress measurement.
Which conclusion is most defensible?
A. The workshop reduced every attendee's stress by four points.
B. Attendance is associated with lower reported stress in the surveyed sample.
C. Low stress causes students to attend workshops.
D. The workshop cannot affect stress because attendance was voluntary.
B is best. A converts a between-group difference into an individual before-and-after change. C selects a reverse causal explanation without evidence. D wrongly treats a design limitation as proof that an effect is impossible.
The workshop might help. Students with different starting stress, time, motivation, or support might also differ in attendance. The study cannot separate those possibilities from the reported comparison alone.
Add time order, then identify what remains
Suppose investigators measure stress before the workshop period and again afterward. Attendees improve more than non-attendees. This helps describe change and establishes that the measured outcome change occurred after the period began.
It still does not make voluntary attendance equivalent to random assignment. Attendees may differ in other factors that influence change. Regression toward typical values, different concurrent support, or differential loss to follow-up could matter depending on the actual design.
Do not invent these as established explanations. They are alternatives the design may need to address. A good answer distinguishes “could explain” from “did explain.”
A stronger causal comparison
Now imagine eligible volunteers are randomly assigned to an immediate workshop or a delayed-workshop comparison. Both groups complete the same stress measure before and after the same interval. Follow-up is similar across groups, and investigators analyse participants according to their original assignment.
This design provides a stronger basis for estimating the effect of being assigned immediate access under those study conditions. It does not automatically establish the effect on every student at every university, or isolate which workshop component matters.
If the workshop group receives substantially more personal attention, the treatment comparison includes that attention. A more closely matched comparison could help distinguish content-specific effects if that is the research question. “Randomized” strengthens a causal design; it is not permission to broaden the conclusion indefinitely.
Interpret the measured variable precisely
A self-report stress questionnaire measures responses to that instrument. It may be useful, but it is not automatically equivalent to every physiological or clinical aspect of stress. If the passage reports survey scores, use that wording when the distinction matters.
Similarly, a study of attendance does not necessarily measure engagement. A participant can be assigned access and never attend. The effect of assignment and the effect of attendance are different questions because actual attendance may depend on participant characteristics.
Behavioural-science passages require this kind of alignment between construct, measurement, and inference. AAMC includes scientific reasoning within the psychological and social science content framework. AAMC content outline
Watch for group-to-individual leaps
Suppose cities with more public study spaces also have higher average examination scores. That city-level association does not show that students who use those spaces score higher than students who do not. The units of analysis differ.
A third city characteristic—such as educational resources—could relate to both quantities. Individual-level data might help ask the individual question, but collecting it still would not automatically establish causation.
The same caution applies when a mean changes. A lower group average does not mean every participant improved. Some may improve, remain stable, or worsen. Answer choices containing “all,” “every,” or “must” require evidence of that breadth.
Practise with a new scenario
In a fictional survey, students who join a peer discussion group report greater confidence in statistics. The investigators conclude that the group increases statistical competence.
Name two separate gaps before reading on.
First, group membership is self-selected, so starting interest or ability may differ. Second, confidence is the measured outcome; competence is a broader claim requiring an appropriate performance measure. A pre/post confidence measure would not by itself solve the competence gap.
A useful follow-up might assign access to a clearly described programme and measure performance with an appropriate assessment, while tracking participation and follow-up. Whether that design is feasible or ethical depends on the actual setting; this is a reasoning exercise, not a complete research protocol.
For your next passage, write: “The study compares ___, measures ___, and supports ___.” Then identify the smallest change that would make an overstrong answer defensible. That exercise teaches you to judge the exact claim rather than reject every observational result as useless.
Sources and further reading
- OpenStax Psychology 2e: Analyzing Findings — Experimental variables, random assignment, and limits of correlation.
- AAMC: What is on the MCAT Exam? PDF Outline — Official distinction between content knowledge and scientific reasoning skills.