In the preceding chapters, we have discussed ways to estimate various statistics that summarise data and/or hypotheses, such as sample means and variances, parameters of models, their distributions, confidence intervals and p-values from goodness-of-fit tests. We can calibrate these if we know the sampling distribution of the relevant statistics. That is, we can place the observed value in the distribution expected (for a given hypothesis) and assess whether it is in the expected range or not. For example, in order to compute a p-value from a goodness-of-fit test, we need to know the distribution of the test statistics, or to find the variance (or bias) of some estimator we need to know the sampling distribution of the estimator. These follow from the distribution of the data and the mathematical relationship between the data and the statistic. Often this is difficult, sometimes even impossible, to perform analytically. But the Monte Carlo method makes many of these problems tractable, and provides a powerful tool for analysing data, and understanding the properties of analysis procedures and experiments.

The core of the Monte Carlo method is to generate random data and use this to compute estimates of derived quantities. We can use the Monte Carlo method to evaluate integrals, explore distributions of estimators and estimate any other quantities of sampling distributions.