Computing the exact distribution of the Bartlett's test statistic by numerical inversion of its characteristic...

Area of Science:

• Statistics
• Statistical Inference
• Hypothesis Testing

Background:

• Exact statistical inference often results in complex probability distributions for estimators and test statistics.
• Characteristic functions can define these exact distributions, but analytical inversion is often computationally challenging.
• The null distribution of Bartlett's test statistic is one such case, crucial for variance homogeneity testing.

Purpose of the Study:

• To present the explicit characteristic function of the corrected Bartlett's test statistic.
• To provide a computationally fast and efficient implementation for evaluating its exact null distribution.
• To facilitate accurate testing of homogeneity of variances in normal populations, even with unequal sample sizes.

Main Methods:

• Derivation of the explicit characteristic function for the corrected Bartlett's test statistic.
• Implementation of numerical inversion techniques for the characteristic function.
• Application of the method for evaluating exact null distributions in statistical inference.

Main Results:

• The explicit characteristic function of the corrected Bartlett's test statistic is presented.
• A computationally efficient algorithm for numerical inversion is developed and implemented.
• The method successfully evaluates the exact null distribution for variance homogeneity testing.

Conclusions:

• Numerical inversion of characteristic functions offers an efficient alternative to analytical methods for complex statistical distributions.
• The presented implementation provides a valuable tool for accurate statistical inference, particularly in testing variance homogeneity.
• This work addresses a gap in current statistical software by enabling the use of exact distributions for the corrected Bartlett's test.