On a preliminary test estimator for one of the parameters of the log-zero-poisson distribution

A discrete distribution called the log-zero-Poisson distribution has been recommended by Katti (c.f. Biometrics 1970) as an alternate to the negative binomial and other distributions usually called "contagious" distributions.A major problem in the use of this and all other contagious distributions has been the difficulty of obtaining the maximum likelihood esti-mates. A custom-made ad hoc estimator, λ, has been proposed for the parameter λ of this distribution in Katti and Khedr (1980). In this paper, its efficiency relative to Fisher information is studied, only to discover that λ can be 30 times better than the maximum likelihood estimate in some parts of the parameter space and much weaker in other parts.A preliminary test is recommended to choose between the estimates, and the efficiency of the procedure is tabulated. As it is to be expected, the resultant estimator equals the better of the two estimators with some error at the values of the parameters where the two estimators are equivalent.     

An algorithm to test divisibility of continuous distributions,applied to the uniform distribution

We present an algorithm to test if a continuous random variable is n-divisible. Through the algorithm, testing for divisibility is put in the format of numerical analysis amenable to computer processing. This is illustrated with the uniform distribution which is known to be non-divisible. New evidence and new proofs arising out of the insight developed with this style are given. This algorithm, along with the analytical methods should provide much needed versatility to tackle problems in this area. The fact that Bondesson [1987] works on the divisibility of the Half Cauchy Distribution gives evidence that there are many distributions which can use the combination of the analytical and the algorithmic tools to settle the divisibility issue.     

Efficiency of an adaptive estimator of a log-zero-poisson parameter

An estimator, λ is proposed for the parameter λ of the log-zero-Poisson distribution. While it is not a consistent estimator of λ in the usual statistical sense, it is shown to be quite close to the maximum likelihood estimates for many of the 35 sets of data on which it is tried. Since obtaining maximum likelihood estimates is extremely difficult for this and other contagious distributions, this estimate will act at least as an initial estimate in solving the likelihood equations iteratively. A lesson learned from this experience is that in the area of contagious distributions, variability is so large that attention should be focused directly on the mean squared error and not on consistency or unbiasedness, whether for small samples or for the asymptotic case. Sample sizes for some of the data considered in the paper are in hundreds. The fact that the estimator which is not a consistent estimator of λ is closer to the maximum likeli-hood estimator than the consistent moment estimator shows that the variability is large enough to not permit consistency to materialize even for such large sample sizes usually available in actual practice.     

A non-bayesian approach to estimation using tested priors

A two stage sampling scheme for estimating the mean of a distribution, proposed by Katti (1962), is investigated and some of its properties are generalized. The method utilizes subjective information in the analysis, but is still within the classical framework. The generalized mean square error of the estimator is computed as the loss function and is compared with that of the usual classical approach. Modifications are suggested to improve the efficiency of the estimator by incorporating the uncertainty of the subjective information. The procedure, which is referred to as "the Method of Tested Priors" is an alternate way of using subjective information without necessarily agreeing with the philosophical aspect of the Bayesian approach.     

Sampling distribution after splitting experimental units

It has been a common practice to recommend one distribution for a large class of problems. An example is the use of the logarithmic distribution for count data - the point of concern in this paper is its recommendation without regard to the size of the experimental unit. References for this particular distribution go back to Fisher et al. [1943]. Look up Douglas [1980], Patil et al. [1984] and Kotz and Johnson [1985] to pick up additional references in this area -especially through sorting the data base of the American Mathematical Society. We will show a fallacy in this structure; provide a computer algorithm find the actual distributions; and then to check on the divisibility. The language used is APL2. But the users can make up their own programs.