520 |
|
|
|b The main purpose of this paper is to provide a confidence interval based sample size determination method when a confidence interval is constructed for the purpose of both interval estimation and hypothesis testing. Therefore, both the interval's excluding ability and accuracy of the interval need to be targeted when calculating a sample size. In this paper W and R represent each event of achieving a length less than 5 and having an excluding ability for a true difference of d by a confidence interval, respectively. The paper is based on the method of directly controlling P(W R), which can be found in several literatures including Jiroutek et. al. (2003). An alternative method is presented for controlling the lower bound of P(W R) using the Bonferroni's inequality and concluded that the method that directly controls P(W R) is more efficient, yielding a smaller sample size, than the method controlling the lower bound of P(W R) in the situation where P(W R)<1 In this paper, three discrete and two continuous cases were discussed. The closed form formula of event probabilities, P(W), P(R) and P(W R), in each distributions are included. Three discrete cases are Bernoulli (p), negative- binomial (r(known), p), and Poisson. Two continuous cases are normal (known), o2) and exponential
|