Bootstrapping

Bootstrapping is a resampling method that uses samples randomly drawn from the already collected sample with replacement.

Imagine a paleontologist trying to determine the mean wing length of a prehistoric insect species with only five fossil specimens.

A higher sample size is desirable to make better inferences, but there is no way to obtain more fossils. In such cases, the bootstrap resampling method is beneficial.

These data from five specimens give a mean length of 10.7 cm.

To begin bootstrapping, randomly draw samples from the original sample set.

Notice that this sample has an identical sample size to the original one, but some values are repeated. This occurs because the bootstrap resampling is entirely random.

Several such bootstrap samples are drawn to estimate the mean wing length distribution. This way, confidence intervals can also be obtained to estimate the population mean more accurately.

Bootstrapping is easy and cost-effective, but it relies on a limited sample. If such a sample is biased or collected erroneously, the bootstrap resampling will remain as biased or erroneous as the original sample.