Bootstrap simulation

In the previous page, we used random samples from a normal population to simulate the error distribution.

If a normal distribution does not seem a reasonable model, an alternative is to treat the actual data as the 'population' for the simulation and take random samples with replacement from this population. Such samples are called bootstrap samples.

A simulation with these bootstrap samples can again show the error distribution and provide approximate values for the bias and standard error.

October rainfall in Samaru, Northern Nigeria

If a normal distribution is a reasonable model for the data, then it is usually better to use it for simulations. However if normality is not a reasonable assumption, then a bootstrap simulation is an alternative that can be used to get a handle on the likely size of the errors.