A bootstrap procedure creates repeated samples from the available observations, typically sampling with replacement. Calculating a statistic on those samples produces a distribution that can help assess its variability under the procedure's assumptions.
The sampling unit and assumptions need to fit the data. Dependent observations may require grouping or other suitable methods, and resampling cannot fix biased collection, missing conditions or an evaluation that measures the wrong outcome.
ELI5
The method asks how much a measured result might vary if the observed data were represented differently. It repeatedly builds new samples from the observations and compares the results.
For example, a researcher studying daily performance could resample appropriate day-level units and recompute the average. A wide range of averages would suggest that the measured result is uncertain, rather than proving a dependable improvement.
Does resampling collect new real-world evidence?
No. It uses the available observations to assess variability under specified assumptions.
Can every observation be treated as independent?
No. Dependencies and the chosen sampling unit matter, so the procedure must suit the structure of the data.
