APPROXIMATE POWER AND SAMPLE SIZE DETERMINATION FOR COMMON ONE-SAMPLE AND 2-SAMPLE HYPOTHESIS TESTS
APPROXIMATE POWER AND SAMPLE SIZE DETERMINATION FOR COMMON ONE-SAMPLE AND 2-SAMPLE HYPOTHESIS TESTS
复制标题
DOI：
10.1177/001316447003000404
复制
发表时间：
1970-01-01
影响因子：
2.7
通讯作者：
COHEN, J
中科院分区：
文献类型：
--
作者：
COHEN, J
812教育和心理测量（y），a和N。幂== 1―~ 3，其中~是当H1为真时未能拒绝Ho的概率，即II类错误。对于任何给定的统计检验，功效、-y、a和N这四个参数是如此相关，以至于它们中的任何一个都可以被写成其余三个的函数，并且所有这四种形式的功效分析都可能用于数据分析（Cohen，1965）;这篇文章强调了作为y、α和N的函数的功率的确定以及作为期望功率、α和-y的函数的N的确定，并更简要地考虑其他形式。是什么使简化方案提出可能的是这样一个事实，即大多数统计量的抽样分布进行假设检验是正常的一个充分的近似，至少对于大样本，当他们不是，可用表变换他们。大样本限制（通常大于25或30）通常根本没有限制，因为在行为科学的大部分范围内，除非样本很大，否则效应量太小，无法获得可接受的大功效。&dquo;最后，确定准确的功效值或样本大小通常并不重要，即使是粗略估计也很有用，而且在任何情况下，其价值都比没有大得多。读者寻求更精确的值（和更详细的说明，需要较少的计算）是指
812 EDUCATIONAL AND PSYCHOLOGICAL MEASUREMENT (y), a, and N. Power== 1―~ 3, where~ is the probability of failing to reject Ho when H1 is true, a &dquo; Type II&dquo; error. For any given statistical test, the four parameters, power,-y, a and N are so related that any one of them can be written as a function of the remaining three, and all four of these forms of power analysis are potentially of use in data analysis (Cohen, 1965); this article emphasizes the determination of power as a function of y, a, and N and the determination of N as a function of desired power, a and-y, and considers the other forms more briefly. What makes the simplified scheme to be presented possible is the fact that the sampling distributions of most statistics subjected to hypothesis testing are normal to a sufficient approximation, at least for &dquo; large&dquo; samples, and when they are not, available tabled transformations of them are. The large-sample limitation (conventionally, greater than 25 or 30) is frequently no limitation at all, since over much of the behavioral science spectrum, effect sizes are too small for acceptably large power to be attained unless samples are &dquo; large. &dquo; Finally, the determination of &dquo; exact&dquo; power values or sample sizes is usually not critical-even &dquo; ballpark&dquo; estimates are quite useful, and in any case of infinitely greater value than none at all. The reader seeking more exact values (and more detailed exposition with less need for computation) is referred to