Match the conclusion to the evidence collected
A random sample supports inference to its target population. Random assignment supports comparing treatments causally. Neither phrase is interchangeable with simply having many participants or seeing a large difference.
- Mean is sensitive to extreme values.
- Median is the middle value after ordering.
- Probability is favorable outcomes divided by total possible outcomes.
- A sample must be representative before generalizing.
- Correlation does not prove causation.
A large survey with a narrow audience
An online fitness forum surveys 5,000 members and finds that 72% exercise daily. Can researchers conclude that 72% of all adults exercise daily?
No. The participants came from a fitness forum rather than a representative sample of all adults. A large response count reduces some random fluctuation within that audience but does not make the audience representative of the wider population.
Keep the setup accurate
- Using the wrong denominator in probability.
- Confusing median with mean.
- Assuming a trend line passes through every point.
Use the method while it is fresh.
What conclusion is safer?
Describe the surveyed respondents or the population justified by the actual sampling process. Avoid widening a conclusion from an interested subgroup to all adults without evidence that the selection supports that step.