← Evaluate statistically based reports
1 Reading the notation
This standard is about reading other people’s statistics critically. These are the terms you need to use precisely.📖 Terms used when evaluating reports
9 termsCover the right column and check you can say what each one means.
- Population vs sampling frame vs sample
- Population = everyone you want conclusions about. Sampling frame = the list you actually sampled from. Sample = who you got data from. Bias creeps in at each gap. Naming WHICH gap is where the marks are.
- Bias
- A systematic tendency for an estimate to be wrong in a particular direction. Bias shifts the whole interval; it does not widen it. So a bigger biased sample is more CONFIDENTLY wrong.
- Non-response bias
- Those who did not reply differ systematically from those who did.
- Self-selection bias
- People chose to take part, so those with strong opinions are over-represented. Online polls are the classic case.
- Confounding variable
- A third variable that influences both of the variables you are comparing, creating a correlation with no direct causal link. Ice cream sales and drownings: both driven by hot weather.
- Observational study vs experiment
- An experiment assigns treatments (ideally randomly) and CAN support causation. An observational study only records what happens and cannot.
- Margin of error
- How far the estimate might reasonably be from the true value. Roughly 1/√n for a proportion. Depends on sample SIZE, not on population size. A national poll needs no more people than a city one.
- Relative vs absolute risk
- Relative = the ratio between groups. Absolute = the actual probability. “Doubles your risk” means nothing until you know the baseline. 1 in a million to 2 in a million is a doubling.
- Statistical significance
- A result unlikely to be due to chance alone. Not the same as being large enough to matter. With a big enough sample, almost any tiny difference becomes detectable.