next up previous contents
Next: The goodness-of-fit test and Up: Goodness-of-fit tests Previous: The equal-probabilities method of   Contents

The number of classes with equal probabilities

The power of the test, i.e. the probability that we do not accept the hypothesis when it is false [Kendall and Stuart, 1979, §22.8] is in general a function of the number of classes $k$ and should, of course, be as large as possible Furthermore, $k$ should not be chosen too large or too small, since we cannot expect the theory to hold in these cases. A rough indication is [Kendall and Stuart, 1979, exam. 30.4]:
k = 3.2{(n-1)}^{2/5}
\end{displaymath} (5.62)

where $n$ is the number of observations. Furthermore, the number of observations in each class should at least be equal to 5 [Hogg and Craig, 1978, p.271].