Quantifying variation and estimating the effects of sample size on the frequencies of linguistic variables

Estimating the frequency of linguistic variables is a fundamental task in the analysis of linguistic data. However, as it is often the case that the amount of material available from different people or text categories may vary, the simplest methods of calculating frequencies are not always appropriate. In this article, we discuss different approaches, including bootstrap methods and a Bayesian approach, and compare the results they yield with those given by some of the simple measures in common use, such as pooling and averaging. We also study the effect of sample size on the accuracy of the estimates.