مجله علوم آماری
سال دوازدهم شماره 1 (پیاپی 23، بهار و تابستان 1397)

Two stage linear models are applicable when the data of some dependent and independent variables was obtained at to time stage, and we want to use from the data of two stage for linear model fitting. In this article we introduce multistage and, as a special case, two-stage linear models. Then we obtain the parameter estimation by two methods and show that the estimation are the same for methods. Since the expression of estimations are very complicated we give some R program for computing the parameter estimation of two-stage linear models, then show its application in an illustrative example. Also we propose a very simple computational methods for parameter estimation which did not need to complicated expression and give and R program for it.

Introducing some efficient model selection criteria for mixed models is a substantial challenge; Its source is indeed fitting the model and computing the maximum likelihood estimates of the parameters. Data cloning is a new method to fit mixed models efficiently in a likelihood-based approach. This method has been popular recently and avoids the main problems of other likelihood-based methods in mixed models. A disadvantage of data cloning is its inability of computing the maximum of likelihood function of the model. This value is a key quantity in proposing and calculating information criteria. Therefore, it seems that we can not, directly, define an appropriate information criterion by data cloning approach. In this paper, this believe is broken and a criterion based on data cloning is introduced. The performance of the proposed model selection criterion is also evaluated by a simulation study.

In this paper, we further investigate stochastic comparisons of the lifetime of parallel systems with heterogeneous independent Pareto components in term of the star order and convex order. It will be proved that the lifetime of a parallel system with heterogeneous independent components from Pareto model is always smaller than from the lifetime of another parallel system with homogeneous independent components from Pareto model in the sense of convex order. Also, under a general condition on the scale parameters, it is proved a result involving with star order.

In this paper, we study the maintenance method in a system. We also consider a system that begin at time zero with most efficienty. After the first failure it is repaired, but we assume that the lifetime of the system is stochastically less than its lifetime at time zero. It is repaired after the second failure and after the third failur it is checked whether the system should be dismanteld or completly repaired. During the performance of the system preventive maintenance could be used to increase the lifetime of the system. Because these actions are costly, we discuss a method for optimizing the cost of preventive maintenance. Finally, we provide some illustrative examples.

The nonparametric and semiparametric regression models have been improved extensively in the field of cross-sectional study and independent data, but their improvement in the field of longitudinal data is restricted to the recent years or decade. Since the common methods for correlated data have a much lower ability rather than for the independent data, we should use the models which consider the correlation among the data. The mixed and marginal models consider the correlation factor among the data, and so obtain a better fit for that. Furthermore, the semiparametric regression has more flexibility compared to the parametric and nonparametric regression. Consequently, based on the properties of the longitudinal data, the marginal longitudinal semiparametric regression with the penalized spline estimations, is a suitable choice for the analysis of the longitudinal data. In this article, the semiparametric regression with different coefficients which specifies the relationship between a response variable and an explanatory variable based on another explanatory variable is assessed. In addition, Bayesian inference on the nonparametric model for a simulated data and the marginal longitudinal semiparametric model for a real data have been done by standard software; and the results have good performance.

Nowadays, the use of various censorship methods has become widespread in industrial and clinical tests. Type I and Type II progressive censoring are two types of these censors. The use of these censors also has some disadvantages. This article tries to reduce the defects of the type I progressive censoring by making some change to progressive censorship. Considering the number and the time of the withdrawals as a random variable, this is done. First, Type I, Type II progressive censoring and two of their generalizations are introduced. Then, we introduce the new censoring based on the Type I progressive censoring and its probability density function. Also, some of its special cases will be explained and a few related theorems are brought. Finally, the simulation algorithm is brought and for comparison of introduced censorship against the traditional censorships a simulation study was done.

The presence of outliers in data set may affect structure of analysis of variance test so that test results led to wrong acceptance or rejection of null hypothesis. In this paper the method of robust permutation distribution of F statistic based on trimmed mean is proposed. This method by permutation distribution of a function of trimmed mean, reduces the sensitivity to classical assumptions such as normality and presence of outlier and it guarantees the reliability of result. The proposed method is compared with robust analysis of variance based of forward search approach. The proposed method, unlike the forward search-based approach is free of restricted parametric assumptions and computationally spend less time. Numerically assessment results on type I error and power of test, demonstrate good performance of this robust method in comparison with competitor method.

In this paper a method has been given to detect the shocks in structural time series using Kalman filter algorithm. As the Kalman filter algorithm is used for state space forms which include ARMA models as an especial case, the suggested method can be used for more general time series than linear models. Five shocks; additive outlier, level change, seasonal change, periodic change and slope change have been reviewed with this method. The performance of suggested method has been shown via a simulation study. The marriage data set from England has been considered as a real data set to study.

In this article, with the help of exponentiated-G distribution, we obtain extensions for the Probability density function and Cumulative distribution function, moments and moment generating functions, mean deviation, R$acute{e}$nyi and Shannon entropies and order Statistics of this family of distributions. We use maximum liklihood method of estimate the parameters and with the help of a real data set, we show the Risti$acute{c}$-Balakrishnan-$G$ family of distributions is a proper model for lifetime distribution.

Prediction of spatial variability is one of the most important issues in the analysis of spatial data. So predictions are usually made by assuming that the data follow a spatial model. In General, the spatial models are the spatial autoregressive (SAR), the conditional autoregressive and the moving average models.
In this paper, we estimated parameter of SAR(2,1) model by using maximum likelihood and obtained formulas for predicting in SAR models, including the prediction within the data (interpolation) and outside the data (extrapolation). Finally, we evaluate the prediction methods by using image processing data.

This paper introduces a new distribution based on extreme value distribution. Some properties and characteristics of the new distribution such as distribution function, moment generating function and skewness and kurtosis are studied. Finally, by computing the maximum likelihood estimators of the new distribution's parameters, the performance of the model is illustrated via two real examples.

Suppose that we have a random sample from one-parameter Rayleigh distributioný. ýIn classical methodsý, ýwe estimate the interesting parameter based on the sample information andý ýwith usual estimatorsý. ýSometimes in practiceý, ýthe researcher has some information about the unknowný ýparameter in the form of a guess valueý. ýThis guess is known as nonsample informationý. ýIn this caseý, ýlinear shrinkage estimators are introducedý ýby combining nonsample and sample information which have smaller risk than usual estimators in the vicinity ofý ýguess and true valueý. ýIn this paperý, ýsome shrinkage testimators are introduced using different methods based oný ývicinity of guess value and true parameter and their risks are computed under the entropy loss functioný. ýThený, ýthe performance ofý ýshrinkage testimators and the best linear estimator is calculated via the relative efficiency of themý. ýThereforeý, ýthe results are applied for the type-II censored data.

According to multiple sources of errors, shape data are often prone to measurement error. Ignoring such error, if does exists, causes many problems including the biasedness of the estimators. The estimators coming from variables without including the measurement errors are called naive estimators. These for rotation and scale parameters are biased, while using the Procrustes matching for two dimensional shape data. To correct this and to improve the naive estimators, regression calibration methods that can be obtained through the complex regression models and invoking the complex normal distribution, as well as the conditional score are proposed in this paper. Moreover, their performance are studied in simulation studies. Also, the statistical shape analysis of the sand hills in Ardestan in Iran is undertaken in presence of measurement errors.

In this paper, based on generalized order statistics the Bayesian and maximum liklihood estimations of the parameters, the reliability and the hazard functions of Gompertz distribution are investigated. Specializations to Bayesian and maximum liklihood estimators, some lifetime parameters of progressive II censoring and record values are obtained. Also by using two real data sets and simulated data accurations of different estimates of the parameters are compared. Next the Bayesian and maximum liklihood estimates of the Gompertz distribution are compared with Weibull and Lomax distrtibutions.