ABSTRACT
The rational way to choose a sample size is by weighing the benefits you can expect in information against the cost of increasing the sample size. Sample sizes are too often determined on the basis of convention or of the available budget. A more rational method of choosing the size of the sample is by balancing the diminution of error expected with a larger sample, and its value, against the cost of increasing the sample size. The relationship of various sample sizes to various degrees of accuracy can be estimated with Monte Carlo methods. Step-wise sampling is not always practical, however, and the cable-television telephone-survey example is unusually favorable for its use. One major pitfall is that the early responses to a mail survey, for example, do not provide a random sample of the whole, and therefore it is a mistake simply to look at the early returns when the split is not wide enough to justify a verdict.
