Causal reasoning · 1/6

Association is a starting point, not a cause

Turn an observed difference into a precise, testable causal question.

About 15–20 minutes. The following cases are fictional.

Record your first judgment

Employees using AI finish 12 tasks a day; nonusers finish eight. A manager says AI raises productivity by 50%. How much does the evidence establish?

My initial judgment, confidence and reasons

Understand the judgment process

Twelve is 50% above eight: an observed group difference. It does not tell us how the same employees would perform without AI under comparable conditions. Task difficulty and quality may differ too.

AI could help. Skilled employees could be more likely to adopt it. Roles with easily counted tasks could also adopt it more. These explanations can coexist; alternatives identify missing evidence rather than establish that AI is ineffective.

Ask instead: among eligible employees, how does access to a specified AI tool for two weeks change average quality-approved output compared with no access? Specify population, intervention, comparison, outcome and follow-up before interpreting the numbers.

Checks you can perform

Separate the observed difference from the proposed cause.
Specify population, intervention, comparison, outcome and period.
List two explanations compatible with the observation.
Name the comparison needed next.

Guided practice

Rewrite the claim in one sentence and propose a comparison that reduces differences in prior skill.

My analysis and checking plan

Explore the explanation

In this sample, AI users completed 50% more tasks on average; role, skill and quality differences remain unresolved. Randomize access within similar roles and baseline skill groups, apply common quality criteria, and check use and sharing between groups.

Apply it in a new context

A platform reports that weekly note-takers score ten points higher and promises everyone a ten-point gain. What can you retain?

My independent response

Compare after attempting

Retain the sample association. Prior knowledge, study time and course choice may explain it; a mean difference is not an individual effect. Compare baseline characteristics and improve assignment to the methods.

Preserve a revision record

What changed your judgment? What still supports it? When will you review?

Failure to establish causation does not establish no effect. An average difference is not an individual promise.

Check whether you supplied inspectable evidence, a feasible next step and revision conditions. Revealing an answer is not mastery.

Source and scope

Hernán & Robins — Causal Inference: What If (chapters 1–3, 7–8)

The source provides conceptual or methodological background. Cases, steps and exercises are authored here; they are not the original experiment or evidence of this course’s effectiveness.

I have completed this section; uncheck to undo

Completion is a personal record, not proof of mastery or certification.