Bootstrap resampling is a practical way to estimate uncertainty when you only have one sample from a population. Instead of assuming a perfect formula for the sampling distribution, the bootstrap uses the data you already collected as a stand-in for the population. It is especially useful for estimating standard errors and confidence intervals for statistics such as means, medians, proportions, and differences between groups.
The main idea is to resample from one sample many times to see how much the statistic changes.
Understanding Statistics: Bootstrap Resampling
A bootstrap works because the observed sample contains clues about the larger population. Think of the sample as a small bag of possible outcomes. Each resample builds a plausible version of what another study might have found.
Some original observations are missing from a given resample, while others occur several times. In a large resample, only about sixty three percent of the original observations appear at least once. This changing mix makes the calculated result move around.
That movement is evidence about random sampling error. A tight cluster of results suggests a precise estimate. A wide cluster suggests that the available data leave more uncertainty.
The shape of the bootstrap results matters, not just their spread. A roughly balanced, bell shaped set of results often gives an interval that is easy to interpret. A skewed set of results can produce an uneven interval, with more distance on one side of the estimate.
This is common for waiting times, incomes, house prices, or measurements that cannot go below zero. The median can behave differently from the average because it depends on the middle positions of ordered data. Correlation can behave strangely when a few unusual points control the relationship.
Plotting a histogram of the resampled results is therefore important. Students should check for skew, gaps, separate clusters, and extreme values before trusting a single uncertainty number.
In real investigations, bootstrap methods are useful when a formula is difficult or unknown. A biology class might estimate the typical leaf length from a field sample. A sports project might compare the median running times of two training groups.
A school survey might estimate the difference in proportions of students who use public transport. The procedure can be done with a spreadsheet, calculator software, or a short program. Computers make thousands of repetitions practical, but the computer is not supplying new information.
It is only showing how sensitive the chosen statistic is to the data already collected. Careful sampling at the start remains essential.
Bootstrap conclusions are strongest when the original observations are reasonably independent and represent the population of interest. They can fail with tiny samples, especially if the sample misses an important type of person or event. A resampling method cannot repair bias from a convenience sample.
It can repeat that bias many times very efficiently. Dependence needs special care too. Daily temperatures, repeated measurements from one patient, and responses from students in the same class are connected.
Treating connected observations as separate can make uncertainty look too small. When learning this topic, first state the population, the sampling unit, and the statistic.
Then distinguish natural variation in the population from uncertainty caused by taking one sample. That distinction is the main reason bootstrap results are useful.
Key Facts
- A bootstrap sample is made by sampling n observations with replacement from the original sample of size n.
- Resampling with replacement means the same data point can appear more than once in one bootstrap sample.
- For each bootstrap sample, compute the statistic of interest, such as x̄, median, p̂, or r.
- The bootstrap sampling distribution is the distribution of the statistic values from many bootstrap samples.
- Bootstrap standard error can be estimated by SE_boot = standard deviation of bootstrap statistics.
- A 95% percentile bootstrap confidence interval is approximately the 2.5th percentile to the 97.5th percentile of the bootstrap statistics.
Vocabulary
- Bootstrap resampling
- A method that repeatedly samples from the original data with replacement to estimate the variability of a statistic.
- Resampling with replacement
- A sampling process where each selected observation is put back before the next draw, so it can be chosen again.
- Bootstrap sample
- One new sample of the same size as the original sample, created by drawing observations with replacement from the original data.
- Sampling distribution
- The distribution of a statistic over many repeated samples or simulated resamples.
- Confidence interval
- A range of plausible values for a population parameter based on the observed data and its estimated uncertainty.
Common Mistakes to Avoid
- Sampling without replacement, because this only reshuffles the original data and does not create realistic variation in the statistic.
- Changing the bootstrap sample size, because each bootstrap sample should usually have the same size n as the original sample.
- Bootstrapping raw data that are not representative, because the method cannot fix bias from a poor sampling design.
- Treating a 95% confidence interval as a 95% probability that the true parameter is inside this one interval, because the parameter is fixed and the interval is random.
Practice Questions
- 1 An original sample has 8 values. You create 1000 bootstrap samples. How many observations should be drawn in each bootstrap sample, and are repeated values allowed?
- 2 The 2.5th percentile of 2000 bootstrap means is 12.4 and the 97.5th percentile is 18.9. What is the 95% percentile bootstrap confidence interval?
- 3 Explain why resampling with replacement from the original sample can help estimate uncertainty even when you do not know the population distribution.