Bootstrap estimation and model selection for multivariate normal mixtures using parallel computing with graphics processing units

In applications of multivariate finite mixture models, estimating the number of unknown components is often difficult. We propose a bootstrap information criterion, whereby we calculate the expected log-likelihood at maximum a posteriori estimates for model selection. Accurate estimation using the bootstrap requires a large number of bootstrap replicates. We accelerate this computation by employing parallel processing with graphics processing units (GPUs) on the Compute Unified Device Architecture (CUDA) platform. We conducted a runtime comparison of CUDA algorithms between implementation on the GPU and that on a CPU. The results showed significant performance gains in the proposed CUDA algorithms over multithread CPUs.     

Selecting significant effects in factorial designs: Lenth's method versus using negligible interactions

Among the many analytical techniques that have been published to analyze the significance of the effects in the absence of replications, two have emerged as the most widely used in text books as well as statistical software packages: The Lenth's method and the estimation of the variance of the effects from the values of those considered negligible. This article shows that neither is better than the other in all cases, and by analyzing the results obtained in a wide variety of situations it provides guidelines on when it is preferable to use one or the other technique.     

Power analysis of independence testing for three-way contingency tables of small sizes

The first aim of this paper is to introduce a modular test for the three-way contingency table (TT). The second aim is to describe the procedure of generating TT using the bar method. The third aim is on the one hand to suggest the measure of untruthfulness of H₀ and on the other hand to compare the quality of independence tests by using their power. Critical values for analyzed statistics were determined by simulating the Monte Carlo method.     

Estimation of the integrated volatility using noisy high-frequency data with jumps and endogeneity

In this paper, we investigate a new estimator of the integrated volatility of Itô semimartingales in the presence of both market microstructure noise and jumps when sampling times are endogenous. In the first step, our estimation wipes off the effects of the microstructure noise, and in the second step our estimator shrinks the effects of jumps. We provide consistency of the estimator when the jumps have finite variation and infinite variation and establish a central limit theorem for the estimator in a general endogenous time setting when the jumps only have finite variation. Simulation illustrates the performance of the proposed estimator.     

Confidence intervals for finite difference solutions

Although applications of Bayesian analysis for numerical quadrature problems have been considered before, it is only very recently that statisticians have focused on the connections between statistics and numerical analysis of differential equations. In line with this very recent trend, we show how certain commonly used finite difference schemes for numerical solutions of ordinary and partial differential equations can be considered in a regression setting. Focusing on this regression framework, we apply a simple Bayesian strategy to obtain confidence intervals for the finite difference solutions. We apply this framework on several examples to show how the confidence intervals are related to truncation error and illustrate the utility of the confidence intervals for the examples considered.     

Half-normal approximation for statistics of symmetric simple random walk

In 2013, Döbler used Stein’s method to obtain the uniform bounds in half-normal approximation for three statistics of a symmetric simple random walk; the maximum value, the number of returns to the origin and the number of sign changes up to a given time n. In this paper, we give the non-uniform bounds for these statistics by using Stein’s method and the concentration inequality approach.     

Generalized method of moments for an extended gamma process

In reliability theory, a widely used process to model the phenomena of the cumulative deterioration of a system over time is the standard gamma process (SGP). Based on several restrictions, such as a constant variance-to-mean ratio, this process is not always a suitable choice to describe the deterioration. A way to overcome these restrictions is to use an extended version of the gamma process introduced by Cinlar (1980), which is characterized by shape and scale functions. In this article, the aim is to propose statistical methods to estimate the unknown parameters of parametric forms of the shape and scale functions. We here develop two generalized methods of moments (Hansen 1982 Hansen, L. P. 1982. Large sample properties of generalized method of moments estimators. Econometrica 50 (4):1029–54.[Crossref], [Web of Science ®] , [Google Scholar]), based either on the moments or on the Laplace transform of an extended gamma process. Asymptotic properties are provided and a Wald-type test is derived, which allows to test SGPs against extended ones with a specific parametric shape function. Also, the performance of the proposed estimation methods is illustrated on simulated and real data.     

Semiparametric estimation of treatment effect with density ratio model

In this study we propose a unified semiparametric approach to estimate various indices of treatment effect under the density ratio model, which connects two density functions by an exponential tilt. For each index, we construct two estimating functions based on the model and apply the generalized method of moments to improve the estimates. The estimating functions are allowed to be non smooth with respect to parameters and hence make the proposed method more flexible. We establish the asymptotic properties of the proposed estimators and illustrate the application with several simulations and two real data sets.     

A note on multivariate linear regression

In this note, a hypothesis test based on relevant statistical differences is proposed for multivariate linear regression models whose design matrix rank does not equal the number of regression variables. A statistical example is also provided to illustrate the proposed hypothesis test.     

A general class of non parametric tests for comparing scale parameters

In this paper, a general class of non parametric tests is proposed for the two-sample scale problem. Testing of the scale parameter is very useful in real-life situations commonly faced in engineering, trade, cultivation, industries, medicine, etc. In all these fields, one will prefer the method that gives more consistent results. Thus, it is worthwhile to test the equality of scale parameters. The distribution of the proposed test is established. To assess the performance of the proposed test, the asymptotic efficacies are studied for some underlying distributions and the results are interpreted with useful information. To see the working of the proposed test, an illustrative example for the real-life data set is provided. The simulation study is also carried out to find the asymptotic power of the proposed test. An extension of the general class of tests to the multiple-sample problem is also discussed.