摘要
本文讨论了正态分布方差只有一个变点的检验问题，我们构造了三个检验统计量，其中L检验基于非参数U统计量，B检验基于Bayes方法，R检验由极大似然比方法导出．本文给出了L、B、R检验的渐近临界值，并用MonteCarlo模拟方法研究了这三个检验与平方的CUSUM检验以及LM检验的势，并进行了比较。当变点在序列的前一半位置时，L和R检验较好，当变点在序列的后一半位置时，平方的CUSUM和B检验较好．
This paper considers the problem of testing for a change-point in variance of the sequence of normal random variables with unknown mean. We construct three tests, namely, L-test based on Lehmann's (1951) non-parametric U-statistic, B-test based on Bayesian method and R-test derived from the maximum likelihood ratio method. The approximate critical values of L, B, R tests are calculated. Monte Carlo simulation studies were carried out to calculate power of these tests, the CUSUM of squares test (Brown et al, 1975) and the LM-test (Nyblom, 1989). An empirical power comparison of the above tests suggests that when change-point is between the beginning and the mid portion, L and R tests are better, when changes-point is between the mid.portion and the end, CUSUM of squares and B tests are better.
出处
《应用概率统计》
CSCD
北大核心
1998年第2期113-121,共9页
Chinese Journal of Applied Probability and Statistics
关键词
变点
U-统计量
最大似然比
正态分布
检验
Normal random sequence,change-point, U-statistic,Bayesian methods, maximum likelihood ratio
