Explain resampling techniques, the bootstrap among them
The idea
Resampling estimates the variability of a statistic by drawing repeatedly from the data itself rather than relying on a formula. The bootstrap draws samples of the same size with replacement, computes the statistic each time and uses the spread of those values as the standard error. Jackknife resampling instead leaves out one observation at a time. The appeal is that no distributional assumption is needed, which matters for statistics with no clean analytical standard error. The limit is that resampling can only reflect what the original sample contains: a biased sample resampled a thousand times is still biased.
A practice question
What does the bootstrap use to estimate a standard error?
- A formula derived from the normal distributionWhy not: That is the analytical route the bootstrap exists to avoid. Its appeal is needing no distributional assumption.
- Repeated samples drawn with replacement from the observed data — correct
- A second, independent datasetWhy not: It needs only the one sample. Resampling from that sample is what substitutes for data nobody has.
The bootstrap treats the sample as a stand-in for the population and resamples from it, using the spread of the resulting statistics as the standard error. No distributional assumption is required.
A harder one
Why is resampling particularly useful for unusual statistics such as a median or a Sharpe ratio?
- Those statistics have no simple analytical standard error — correct
- Those statistics cannot be computed from a single sampleWhy not: Both compute perfectly well from one sample. What is missing is a formula for the *standard error* of each.
- Resampling makes the statistics unbiasedWhy not: It estimates variability and does nothing to bias. A biased statistic stays biased in every resample.
Closed-form standard errors exist for a mean but not for many statistics of practical interest. Resampling supplies one empirically without needing the algebra.
The hardest tier
What limitation does the bootstrap share with the sample it is drawn from?
- It requires the data to be normally distributedWhy not: It requires no distributional assumption at all — that freedom is the main reason to use it.
- It can only be applied to meansWhy not: Means are where analytical formulas already work. The bootstrap earns its keep on medians, ratios and other statistics that have none.
- It cannot correct a sample that is unrepresentative of the population — correct
Resampling explores the variability within the data available. If the original sample was biased, every bootstrap replicate inherits that bias, and the confidence it produces is false confidence.
In the app: 9 practice questions on this outcome, a narrated walkthrough with diagrams, spaced review that brings it back before you forget it.
Take the free diagnostic →