Executive PrepLab · V2 · Practice 3

Cause and correlation

V2.8 is not a question type. It is the reasoning flaw that underlies more items in this format than any other, and it appears inside Assumption, Strengthen, Weaken and Evaluate questions rather than as a category of its own.

That is exactly why it deserves its own block. If you can recognise the shape before you read the options — premises reporting an association, conclusion asserting a cause — you already know the four things the answer could be. It is the closest thing to a shortcut this format allows, and the shortcut only works if recognition is automatic.

The four live alternatives, every time:

| Alternative | What it says | |---|---| | Coincidence | The association arose by chance | | Reverse causation | B is producing A, not the other way round | | Common cause | Something else produces both | | Selection effect | The group measured selected itself |

Everything in this block is built on recognising which of the four the item is about.

Layer 1

Structure drills

Shared by all four question types below — the anatomy work that every one of them depends on. No options: write your answer, then open the model answers for the set.

For each argument: state the correlation, state the causal claim, and say which of the four alternatives is most live. Some have more than one; name the strongest.

D1. Firms that offer flexible working report lower staff turnover than firms

D2. Patients who complete the full course of physiotherapy recover more

D3. Towns with more independent bookshops have higher literacy rates.

D4. Children who attend preschool score higher on entry tests at age six.

D5. Regions that adopted the new seed variety saw yields rise 12 percent. The

D6. People who take a daily multivitamin visit their doctor less often.

D7. Departments that adopted the project-tracking software delivered more

D8. Ships that carry a particular class of cargo report more hull damage.

D9. Students who sit at the front of the lecture hall achieve higher marks.

D10. Countries with higher chocolate consumption have more Nobel laureates per

	The causal claim	Strongest alternative
D1	Flexibility → retention	Common cause. Firms that offer flexibility are typically well-resourced and well-managed, which independently reduces turnover
D2	Completing → recovery	Selection effect, and reversed. Patients who are recovering well are the ones able to complete; those in pain stop
D3	Bookshops → literacy	Reverse causation. Literate populations sustain bookshops. A bookshop cannot survive where nobody reads
D4	Preschool → performance	Selection effect. Families who enrol children in preschool differ systematically — income, parental education, prioritisation of schooling
D5	Variety → yield	Common cause, or coincidence. Regions adopting a new variety may also have adopted new fertiliser, irrigation, or simply had a good season
D6	Vitamins → less illness	Selection effect. People who take daily supplements also exercise, eat better and smoke less. The "healthy user" pattern
D7	Software → on-time delivery	Reverse causation. Well-run departments adopt tracking tools; chaotic ones do not have the discipline to
D8	Cargo → hull damage	Common cause. That cargo may be carried on particular routes, in particular seasons, by older vessels
D9	Front row → marks	Selection effect. Motivated students choose the front. The seat did not choose them
D10	Chocolate → Nobel prizes	Common cause, and a stark one. National wealth produces both. This is the shape at its most obvious, which is why it is worth having in the set

The pattern across the ten: selection effects account for four, common causes for four, reverse causation for two. Coincidence never appears as the strongest alternative — with real datasets it rarely is, and the exam reflects that. If you find yourself choosing coincidence, check whether you have missed a selection effect.

The tell for a selection effect: ask who chose to be in this group, and did anybody assign them? Nothing in the passage will tell you. You have to notice that nobody randomised.

Layer 2

Full items

Item 1

Employees at firms offering flexible working report greater job satisfaction than employees at firms that do not. A consultancy has concluded that introducing flexible working raises job satisfaction.

Which of the following, if true, most seriously weakens the consultancy’s conclusion?

Common cause. Firms that offer flexibility also pay better and train more, and both of those independently raise satisfaction. The correlation survives with flexibility contributing nothing.

A is a single-case objection to an aggregate claim. C is background. D questions the measure without touching the causal link — and note it would weaken any conclusion drawn from satisfaction data, which is a sign it is not the item’s target. E is an implementation cost.

Item 2

Patients who complete a full twelve-week course of physiotherapy after knee surgery recover more mobility than patients who discontinue treatment earlier. The clinic has concluded that completing the full course produces better outcomes and plans to require completion as a condition of treatment.

Which of the following, if true, casts the most doubt on the clinic’s reasoning?

Reversed, and a selection effect at once: patients discontinue because their recovery has stalled. Completion is a marker of good recovery rather than a cause of it — so requiring completion of patients whose recovery has stalled achieves nothing.

B is one-case. A, D and E concern the course length, the staff and the measurement instrument, none of which bears on the direction of causation.

Item 3

A study found that towns with a greater number of independent bookshops per capita have higher adult literacy rates. The researchers propose that bookshops contribute to literacy.

Which of the following, if true, most undermines the researchers’ proposal?

Reverse causation, stated explicitly and with a mechanism: bookshops cannot survive without readers, so literacy produces bookshops.

D is a single-case exception. A is a national trend with no bearing on the cross-sectional comparison. B and E are measurement and alternatives.

Item 4

Regions that adopted a new drought-tolerant seed variety recorded average yield increases of 12 percent in the following season. The agricultural ministry credits the variety and plans a national rollout.

Which of the following would be most useful to determine in evaluating the ministry’s plan?

Two-answer test. If rainfall was close to average, the yield rise cannot be attributed to a favourable season and the variety looks genuinely effective. If rainfall was unusually good, a drought-tolerant variety may have contributed nothing — and the plan would fail in a normal year.

D is the second-best answer and worth naming: training alongside the seed is a real confound. But rainfall is the more fundamental one for a drought-tolerant variety, since the whole point of the variety is what it does when rain is short. A, C and E leave the causal question untouched.

Item 5

Departments that adopted the new project-tracking software subsequently delivered a higher proportion of projects on schedule. The chief operating officer has concluded that the software improves delivery performance and will mandate it across the organisation.

The officer’s conclusion depends on which of the following assumptions?

Negate it: the adopting departments did already differ in ways affecting delivery — better managed, better staffed, more disciplined. Then the correlation is explained without the software, and the mandate achieves nothing.

A and C are implementation and cost. D is about willingness. E questions the metric rather than the inference.

Item 6

People who take a daily multivitamin visit their doctors less frequently than people who do not. A supplement manufacturer cites this in support of the claim that multivitamins reduce the incidence of illness.

Which of the following, if true, most weakens the manufacturer’s claim?

The healthy-user effect, which is the most common selection effect in health data. People who take daily supplements differ from those who do not in a bundle of ways that independently reduce illness.

D is a real objection to the proxy and weakens mildly — but it questions the measure while C dismantles the causal claim. A, B and E are cost, motivation and alternatives.

Item 7

An analysis of hospital records found that patients admitted at weekends had higher mortality than patients admitted on weekdays. A newspaper concluded that weekend staffing levels are causing avoidable deaths.

Which of the following, if true, most seriously undermines the newspaper’s conclusion?

Composition, which is a selection effect by another name. The two groups are not the same kind of patient: weekend admissions skew toward emergencies, weekday admissions include scheduled procedures on stable patients. Higher mortality follows from who is admitted, not from how they are staffed.

A supports the newspaper. C confirms the data are comparable in source. D and E concern responses and magnitude.

Item 8

Students who sit in the front third of a lecture hall achieve higher marks than students who sit at the back. A lecturer proposes assigning seats so that every student spends part of the term at the front.

Which of the following, if true, most seriously weakens the case for the lecturer’s proposal?

Selection effect with the mechanism spelled out: engagement produces both the front seat and the marks. Assigning seats moves the students without moving the engagement, so the intervention gets none of what produced the correlation.

A is one-case. B, D and E concern attendance, capacity and assessment weighting.

Item 9

A city introduced free public transport for residents over sixty-five. In the following two years, reported loneliness among that age group fell. The council attributes the improvement to increased mobility.

Which of the following, if true, most strengthens the council’s explanation?

The strongest kind of strengthener for a causal claim: it links the effect to the individuals the mechanism actually reached, and to the subgroup for whom mobility was most constrained.

B is a control and does strengthen — but it only excludes a general trend, where C shows the mechanism operating on the right people. A shows uptake without linking it to the outcome. D and E are cost and precedent.

Item 10

Firms that appointed a chief sustainability officer between 2015 and 2020 subsequently reduced their emissions more than firms that did not. A business school has concluded that appointing such an officer causes emissions reductions, and recommends the appointment.

Which of the following, if true, most seriously undermines the business school’s conclusion?

Reverse causation with a timeline: firms already committed created the post, and most had begun reducing before the appointment. The officer is a consequence of the commitment, not a cause of the reductions.

C is a data-quality objection that would apply equally to both groups. A and E are background. D is a single-case exception to an aggregate claim.

Answer as many as you like, then check — it scores what you have answered so far, across all 10 items.

Layer 3

Diagnosis

Check your answers above, then read this. The pattern of what you missed matters more than how many.

If you missed	It means	Repeat
1, 7	Common causes not recognised — you look for a single confounder rather than a bundle	D1, D5, D8, D10
2, 3, 5, 10	Reverse causation not automatic; you accept the stated direction	D3, D7
6, 8	Selection effects invisible to you because the passage never mentions them	D2, D4, D6, D9
4	Evaluate on a causal argument — you are not running yes and no separately	V2.4
9	You take a control as stronger than a mechanism	V2.2
Any where you chose a measurement objection (1D, 6D, 10C)	You attack the data rather than the inference	— see below

On measurement objections. Three items offer one — self-report, an imperfect proxy, unaudited figures — and none is the answer. They are worth understanding as a class: an objection that would weaken any conclusion drawn from the same data is not aimed at this argument’s reasoning. When an option questions how something was measured, ask whether it damages the causal link specifically or merely the evidence in general. If it is the second, it is almost never the answer.

The cross-cutting count. Of the ten items, four turn on selection effects, three on reverse causation and three on common causes. If your errors cluster in one column, that is the alternative you are not seeing — and the fix is the drill set, not more items.