Asymptotical problems of sequential interval and point estimation

The accuracy of interval estimation systems is usually measured using interval lengths for given covering probabilities. The confidence intervals are the intervals of a fixed width if the length of the interval is determined, i.e., not random, and tends to zero for a given covering probability. We consider two important directions of statistical analysis -sequential interval estimation with confidence intervals of fixed width and sequential point estimation with asymptotically minimum risk. Two statistical models are used to describe the basis problems of sequential interval estimation by confidence intervals of a fixed width and point estimation. A review of data on nonparametric sequential estimation is carried out and new original results obtained by the authors are presented. Sequential analysis is characterized by the fact that the moment of termination of observations (stopping time) is random and is determined depending on the values of the observed data and on the adopted measure of optimality of the constructed statistical estimate. Therefore, to solve the asymptotic problems of sequential estimation, the methods of summation of random variables are used. To prove the asymptotic consistency of the confidence intervals of a fixed width, we used a method based on application of limit theorems for randomly stopped random processes. General conditions of the consistency and efficiency of sequential interval estimation of a wide class of functionals of an unknown distribution function are obtained and verified by sequential interval estimation of an unknown probability density of asymptotically uncorrelated and linear processes. Conditions of the regularity are specified that provide the property of being an estimate with an asymptotically minimum risk for a wide class of estimates and loss functions. Those conditions are verified by sequential point estimation of an unknown distribution function.

Approximation of multivariate distribution functions

In the paper the unknown distribution function is approximated with a known distribution function by means of Taylor expansion. For this approximation a new matrix operation — matrix integral — is introduced and studied in [PIHLAK, M.: Matrix integral, Linear Algebra Appl. 388 (2004), 315–325]. The approximation is applied in the bivariate case when the unknown distribution function is approximated with normal distribution function. An example on simulated data is also given.

Predicting the Distribution of Aflatoxin Test Results from Farmers’ Stock Peanuts

Suitability of the negative binomial function for use in estimating the distribution of sample aflatoxin test results associated with testing farmers1 stock peanuts for aflatoxin was studied. A 900 kg portion of peanut pods was removed from each of 40 contaminated farmers1 stock lots. The lots averaged about 4100 kg. Each 900 kg portion was divided into fifty 2.26 kg samples, fifty 4.21 kg samples, and fifty 6.91 kg samples. The aflatoxin in each sample was quantified by liquid chromatography. An observed distribution of sample aflatoxin test results consisted of 50 aflatoxin test results for each lot and each sample size. The mean aflatoxin concentration, m; the variance, s2xamong the 50 sample aflatoxin test results; and the shape parameter, k, for the negative binomial function were determined for each of the 120 observed distributions (40 lots times 3 sample sizes). Regression analysis indicated the functional relationship between k and m to be k = 0.000006425m0.8047. The 120 observed distributions of sample aflatoxin test results were compared to the negative binomial function by using the Kolmogorov–Smirnov (KS) test. The null hypothesis that the true unknown distribution function was negative binomial was not rejected at the 5% significance level for 114 of the 120 distributions. The negative binomial function failed the KS test at a sample concentration of 0 ng/g in all 6 of the distributions where the negative binomial function was rejected. The negative binomial function always predicted a smaller percentage of samples testing 0 ng/g than was actually observed. However, the negative binomial function did fit the observed distribution for sample test results at a concentration greater than 0 in 4 of the 6 cases. As a result, the negative binomial function provides an accurate estimate of the acceptance probabilities associated with accepting contaminated lots of farmers' stock peanuts for various sample sizes and various sample acceptance levels greater than 0 ng/g.

Non-linear electrostatic plasma waves

A solution of the one-dimensional time-independent Vlasov–Poisson system, including the condition of no net current, is constructed for a large class of electric potential functions φ(x), including periodic and solitary waves, as well as monotonic transitions. Displaced Maxwell distributions for the free particles, and an unshifted Maxwell distribution for the trapped ions, are used. The condition of positiveness of the unknown distribution function for the trapped electrons (which can be split into two parts, one depending solely on the chosen wave, the other on the given distributions) is examined. It is shown, for a non-linear wavethat this condition plays no rôle when the thickness l of the wave considerably exceeds the Debye length λD(l ≫ λD). If, however, l lies in the neighbourhood of λD(l ≳ λD), there exist limits for the free parameters. The smallest thickness lmin results if the Mach number M lies in the range 1 > M > 2 and the electronion—temperature ratio θ = Te/Ti exceeds 10. This confirms the view that the wave is a steepened ion-acoustic wave. lmin decreases with decreasing amplitude ϕmax of the wave and decreasing number of trapped ions, but does not lie below the Debye length as long as the wave is non-linear. In the linear case, the condition of positiveness imposes no restrictions.

Robust Sequential Confidence Intervals for the Behrens–Fisher Problem*

Summary The problem of providing a bounded length (sequential) confidence interval for the median of a symmetric (but otherwise unknown) distribution based on a general class of one-sample rank-order statistics was investigated in (Sen & Ghosh, 1971). The purpose of the present note is to indicate how the techniques developed there can be extended to the two-sample problem. It has been shown that in particular for the Behrens- Fisher situation (see e.g., Høyland, 1965 or Ramchandramurty, 1966), when the proposed procedure is based on the “normal-scores” statistic, under very general conditions on the unknown distribution function (d.f.), it is asymptotically at least as efficient as an analogous procedure suggested in (Robbins, Simons & Starr, 1967) and (Srivastava, 1970).