Range tests for the dispersion of several location parameters (Q2368336)

Different variation measures \(d_ i(\theta)\) are introduced to measure the variability among \(k\) populations which belong to location models \(f(x-\theta_ i)\), \(\theta=(\theta_ 1,\dots,\theta_ k)\). Instead of \(H_ 0: \theta_ 1=\cdots =\theta_ k\) the hypotheses \(H_{0,i}: d_ i(\theta)\leq \delta\) vs. \(H_{1,i}: d_ i(\theta)>\delta\) are studied. Under the assumption that \(f(x-\theta)\) has monotone likelihood ratio for the statistic \(X_{(k)}-X_{(1)}\) the least favorable configuration on \(H_{0,i}\) is established. For the normal distribution and for different significance levels the percentage points of the Studentized range test for testing dispersion of normal means are given in the case that \(d_ 3(\theta)=\max_{i\neq j}| \theta_ i-\theta_ j|\) is used as dispersion measure. Furthermore, confidence intervals for \(d_ i(\theta)\) are constructed with the help of the range statistic \(X_{(k)}-X_{(1)}\).