Effective Degrees of Freedom and the Likelihood Ratio Test
研究了多个未知方差问题中，基于有效自由度的置信区间与经过Bartlett调整的似然比检验给出的置信区间之间的比较，并详细分析了方差分量问题和Behrens-Fisher问题推广这两个特例。
Normal-theory problems involving several unknown variances may require use of estimates not having simple chi-squared distributions. Confidence intervals are then commonly based on the use of ‘effective degrees of freedom’, i.e. the fitting of a gamma distribution via the first two moments, replacing parameters by estimates. The confidence intervals given by this method are compared with those resulting from the likelihood ratio test, inserting a Bartlett adjustment factor to achieve close approximation by a chi-squared distribution. Two special problems, one concerning components of variance and one a generalization of the Behrens-Fisher problem, are studied in detail.