LogicTest
English
Log inSign up

Articles › Correlation Is Not Causation: Find the Cause Worth Acting On

Correlation Is Not Causation: Find the Cause Worth Acting On

LogicTest Editorial Team · Published 2026-09-01 · Updated 2026-09-01 · 5 min read

When two numbers move together, our minds quickly turn the pattern into a story: one thing must have caused the other.

Sales rise after an ad campaign, so the campaign must have worked. Teams with more meetings miss more deadlines, so cancelling meetings must solve the delay. You start a new drink and feel more alert, so the drink must be responsible.

Any of those conclusions could be true, but “happened together” is not enough. Correlation is a clue that deserves investigation. Causation supports an intervention: it tells us what may happen if we deliberately change one factor. Confusing the two can send money, time, and attention toward the wrong variable.

At least three stories fit the same pattern

When A and B move together, consider three possibilities:

  1. A causes B. The advertisement persuades more people to buy.
  2. B causes A. Rising sales encourage the company to spend more on advertising.
  3. C affects both. A holiday increases both advertising activity and natural customer demand.

The pattern could also be a coincidence or the result of a measurement change. The best response to a correlation is therefore not to tell the first plausible story. It is to search deliberately for competing explanations.

Sales increased after the campaign

Suppose a shop launches a campaign on Friday and weekend sales rise by 30 percent. That is encouraging, but it does not yet reveal how much credit the campaign deserves.

Ask:

  • Are weekends normally stronger than weekdays?
  • Did the shop also change its price, product range, or platform placement?
  • Did new and returning customers behave differently?
  • Did similar customers who never saw the campaign also buy more?

A comparable region, customer segment, or time period can provide a useful control. A stronger test would randomly show the campaign to some eligible customers and compare the groups. Experiments are not always practical, but asking whether a credible comparison exists is almost always useful.

Teams with more meetings miss more deadlines

The obvious intervention is to cut meetings. Yet the direction may be reversed: a project enters trouble first, and the team schedules more meetings in response. A common cause may be complexity, which increases both coordination and delay.

If unclear decision ownership is the real bottleneck, cancelling meetings can reduce information without fixing the delay. Continue the investigation:

  • Which meetings correlate with delay: decision meetings, routine updates, or emergency calls?
  • Did the meetings increase before or after the risk appeared?
  • Does the relationship remain after accounting for project size and dependencies?
  • When a team changes its meeting process, do outcomes improve?

A root cause is not merely the most satisfying story. It is an explanation that survives tests of timing, alternatives, and intervention.

Which life change made the difference?

If someone starts exercising, sleeps earlier, and changes their diet at the same time, then feels better, it is difficult to assign the entire effect to one habit. Their experience still matters. The claim simply needs to match the strength of the evidence.

“This bundle of changes seems helpful for me, so I will keep observing” is reasonable. “One element works for everyone” is a much larger claim. Correlation should never replace professional medical advice when treatment, medication, or significant symptoms are involved.

A six-step causal check

When you encounter “X causes Y,” work through these questions:

  1. Define the variables. How are X and Y measured, and did the definitions remain stable?
  2. Check the order. Did the proposed cause occur before the result?
  3. Find a comparison. What happened to similar cases without X?
  4. List common causes. What C could influence both X and Y?
  5. Test alternatives. Could reverse causation, selection bias, survivorship bias, or chance explain the pattern?
  6. Observe intervention. When X is deliberately changed, does Y respond in a repeatable way?

Not every real-world decision allows a randomized experiment. When evidence is incomplete, label its strength:

  • Clue: the variables move together and deserve more investigation.
  • Plausible explanation: timing and comparisons fit, and major confounders have been addressed.
  • Strong causal evidence: repeated interventions or high-quality studies show consistent results.

Write a prediction before acting

Causal thinking is valuable because it improves action. Before spending a budget, changing a process, or adopting a new routine, write a prediction that could turn out to be wrong. If the campaign works, which group should respond and when? If meetings cause delay, which metric should improve first after the process changes? If a habit helps, what observable change should appear and over what period?

This shifts us from explaining everything after the fact to testing something before we commit. Correlation opens the investigation; causal analysis tells us where the next step should go.

The short practice set at the end presents several pairs of events that move together. For each one, try to propose at least one common cause and one reverse-causation explanation.

Related topics

Correlation and CausationRoot Cause and the Causal ChainSix-Step Causal Analysis

Practice

Answers and explanations are on this page; nothing is recorded.

1. Single choice

Lao Zhang's noodle shop changed its menu at the beginning of March, and that month's customer flow dropped by 20%. He concluded that the new menu drove customers away and decided to switch back to the old menu immediately. A neighbor reminded him that road construction started on that street at the beginning of March, and pedestrian traffic visibly decreased. Which evaluation is the most accurate?

Show answer & explanation

Answer: D

  • AThe menu change happened before the drop in customers, and this order of events is enough to prove the new menu is the cause.✗ Wrong. The fact that the change came before the drop only establishes a temporal sequence, not a causal one; this is the 'post hoc, ergo propter hoc' fallacy.
  • BSince the road construction also affects customer flow, the menu is completely irrelevant, and the result would be the same whether the menu is changed or not.✗ Wrong. Discovering another possible cause does not mean the menu has no effect at all; jumping from 'there is another cause' to 'this cause is ruled out' is an unsupported conclusion.
  • CThe drop in customers is a coincidental combination of multiple factors and cannot be analyzed; one should just rely on years of experience and decide directly.✗ Wrong. Giving up analysis because of multiple possible causes is acting on intuition; the correct approach is to list possible explanations and seek evidence to distinguish them, not to refuse analysis.
  • DThe drop in customers only coincides with the menu change; the road construction might be the main cause, so one should first determine the real cause before deciding.✓ Correct. The drop merely coincides with the menu change, and the road construction may be a more direct cause; investigating first and then acting avoids the risk of fixing the wrong cause and making things worse.
Explanation: Lao Zhang saw a temporal sequence between the menu change and the drop in customers and immediately concluded a causal link, which is the fallacy of 'co-occurrence, therefore causation'. The neighbor's reminder points to an alternative explanation that should be investigated first: the road construction caused the decrease in pedestrian flow. The correct approach is to list multiple possible causes, gather evidence (such as comparing pedestrian flow on the construction section versus a non-construction section, or before versus after the construction), and verify on a small scale before deciding whether to switch back the menu.

2. Single choice

A hospital's statistics show that departments with more complaints also have doctors who attend more doctor-patient communication training hours. The director says: 'The more training, the more complaints; training is counterproductive and should be cancelled immediately.' Which of the following evaluations is most accurate?

Show answer & explanation

Answer: B

  • AThe hours and complaints rise together, showing training has no effect; after cancellation, complaints should fall, and waste is reduced.✗ Incorrect because it treats correlation as causation: the coexistence of more training hours and more complaints is directly read as 'training causes complaints,' then used for a decision, precisely leaping in causal direction.
  • BMore likely, departments with more complaints are prioritized for training; the high hours are a result of addressing the problem, reversing the causal direction.✓ Correct: it points out reverse causation—training does not attract complaints; instead, departments with more complaints are assigned more training, and the director's conclusion exactly reverses the causal direction.
  • CDoctors with more training hours are often more patient, so complaints should be fewer; the statistics may be erroneous, and a re-check is needed first.✗ Incorrect because it presupposes a conclusion based on intuition and doubts the data: it fails to propose testing for a possible reverse causal direction, instead using the subjective expectation of 'should be fewer' to bypass analyzing the correlation.
  • DTraining should be kept but changed to voluntary sign-up; only doctors who genuinely want to improve will attend, and complaints will naturally decrease.✗ Incorrect because it changes the remedy but continues the same misinterpretation: it assumes a direct causal link between training hours and complaints without ever asking whether more complaints lead to more training hours.
Explanation: Key reasoning: the data only show a positive correlation between training hours and complaints; one must first ask whether the causal direction might be reversed. Hospitals typically assign communication training as a corrective measure to departments with many complaints, so complaints come first and training hours follow; the director mistakes the direction. The options advocating cancellation and voluntary sign-up both rest on the untested misinterpretation that 'training causes complaints'; the option questioning statistics or relying on the 'more patient' intuition also fails to examine the causal direction.

3. Single choice

HR found that job postings published as short videos received notably more resumes, and the recruiting manager proposed converting all postings to short-video format. To test whether the short-video format itself attracts more applications, what is the next step that should be taken?

Show answer & explanation

Answer: A

  • APick two departments with similar job roles, keep the text-and-image format for one and switch to short videos for the other, hold all other conditions constant, and compare the number of resumes received.✓ Correct: This involves changing only the factor in question on a small scale with a control group and fixed other conditions. If the only difference is the posting format and resume counts differ, a causal explanation is supported.
  • BRetrieve application data for all postings over the past year and use a model to control for salary, location, and other factors to estimate the effect.✗ Wrong because statistical control cannot replace actual intervention: unobserved differences in historical data (such as job appeal or team reputation) may still affect both the choice of format and application volume, so the estimate remains correlational.
  • CFirst interview candidates who applied through short-video postings one by one to find out their true reasons for applying.✗ Wrong because collecting subjective feedback is not the same as testing causality: interviews can only provide clues, and the sample consists entirely of people who already applied, so it cannot answer whether changing the factor changes the outcome.
  • DDirectly switch all postings to short-video format for a three-month trial and observe whether the total number of resumes rises noticeably.✗ Wrong because the change lacks a control group: during the three months, conditions such as market trends and the number of positions also change, so even if resume volume rises, it cannot be attributed to the posting format itself.
Explanation: Key reasoning: to upgrade correlation to causation, you need to change the factor on a small scale while keeping other conditions as stable as possible. Setting up a control group with similar jobs and changing only the posting format meets this standard. Modeling historical data cannot eliminate unobserved common causes; interviews only provide directional clues, not a test; and switching everything at once lacks a control, so even an increase in resumes cannot be attributed to the format.