Association is a starting point
Two events occurring together can support a causal hypothesis without establishing it. A third factor might produce both. The direction of influence might be reversed. People who select a treatment, program, or habit may differ beforehand from those who do not. A before-and-after improvement may reflect a broader change rather than the intervention.
Define the causal claim precisely. Does it say a factor contributes to an outcome, usually produces it, or is its only possible cause? Evidence of another cause is especially damaging to a sole-cause claim. It need not show that the original factor has no effect whatsoever.
Ask what the comparison holds constant
A useful comparison isolates the factor under investigation as much as feasible. If one workshop has both longer sessions and more experienced instructors, better results cannot straightforwardly be attributed to length alone. Random assignment can reduce systematic preexisting differences between groups; it does not automatically solve attrition, measurement error, or a small and unrepresentative study population.
In observational evidence, look for relevant differences and converging evidence rather than expecting perfect experimental control. Time order matters because a cause must occur before the effect it produces, but merely occurring earlier is insufficient. A plausible mechanism can strengthen an explanation while still leaving competing mechanisms possible.
Separate measured outcomes from desired outcomes
A program may increase reported satisfaction without improving retention, or raise test familiarity without improving transfer to new tasks. Name the outcome that was actually measured. Then identify the additional assumption needed to treat it as a measure of the outcome the conclusion discusses.
A surprising pair of observations often calls for a distinction rather than a contradiction. Total injuries could rise while injuries per participant fall if participation rises enough. A treatment group could improve less in raw totals because it began with substantially more severe cases.
Worked example
Participants in an optional rehearsal group deliver clearer speeches than nonparticipants. The organizer concludes that the group caused the difference.
- Alternative cause
- Students already interested in speaking may practice more and be more likely to join.
- Useful evidence
- Comparable students assigned to the group by a lottery improve more on the same independently rated task. This addresses selection, though details still matter.
- A measurement question
- Were both groups judged by the same standard, without the reviewer knowing who joined?
- A fair conclusion
- The original comparison is consistent with a benefit, but alone does not isolate the group’s causal effect.
Check your understanding
Take it into practice
Name one plausible alternative cause and one comparison that would help distinguish it from the proposed explanation.