نشریه اندیشه آماری
سال بیست و چهارم شماره 2 (پیاپی 48، پاییز و زمستان 1398)

In the analysis of Bernoulli's variables, an investigation of the their dependence is of the prime importance. In this paper, the distribution of the Markov logarithmic series is introduced by the execution of the first-order dependence among Bernoulli variables. In order to estimate the parameters of this distribution, maximum likelihood, moment, Bayesian and also a new method which called the expected Bayesian method (E-Bayesian) are employed. In continuation, using a simulation study, it is shown that the expected Bayesian estimator out performed over the other estimators.

The analysis of discrete mixed responses is an important statistical issue in various sciences. Ordinal and overdispersed binomial variables are discrete. Overdispersed binomial data are a sum of correlated Bernoulli experiments with equal success probabilities. In this paper, a joint model with random effects is proposed for analyzing mixed overdispersed binomial and ordinal longitudinal responses. In this model, we assume an overdispersed binomial response variable follows Beta-Binomial distribution and use a latent variable approach for modeling the ordinal response variable. Also, the model parameters are estimated via the Maximum Likelihood method, and the estimates are evaluated with a simulation study via the Monte Carlo method. Finally, an application of the proposed model to real data is introduced.

In the present era, classification of data is one of the most important issues in various sciences in order to detect and predict events. In statistics, the traditional view of these classifications will be based on classic methods and statistical models such as logistic regression. In the present era, known as the era of explosion of information, in most cases, we are faced with data that cannot find the exact distribution. Therefore, the use of data mining and machine learning methods that do not require predetermined models can be useful. In many countries, the exact identification of the type of groundwater resources is one of the important issues in the field of water science. In this paper, the results of the classification of a data set for groundwater resources were compared using regression, neural network, and support vector machine. The results of these classifications showed that machine learning methods were effective in determining the exact type of springs.

One of the most common censoring methods is the progressive type-II censoring. In this method of censoring, a total of $n$ units are placed on the test, and at the time of failure of each unit, some of the remaining units are randomly removed. This will continue to record $m$ failure times, where $m$ is a pre-determined value, and then the experiment ends. The problem of determining the optimal censoring scheme in the progressive type-II censoring has been studied so far by considering different criteria. Another issue in the progressive type-II censoring is choosing the sample size at the start of the experiment, namely $n$. In this paper, assuming the Pareto distribution for the data, we will determine the optimal sample size, $n_ {opt}$, as well as the optimal censoring scheme by means of the Fisher Information. Finally, to evaluate the results, numerical calculations have been presented by using $R$ software.

The minimum density power divergence method provides a robust estimate in the face of a situation where the dataset includes a number of outlier data.In this study, we introduce and use a robust minimum density power divergence estimator to estimate the parameters of the linear regression model and then with some numerical examples of linear regression model, we show the robustness of this estimator in the face of a dataset which includes a number of outliers.

Sometimes, in practice, data are a function of another variable, which is called functional data. If the scalar response variable is categorical or discrete, and the covariates are functional, then a generalized functional linear model is used to analyze this type of data. In this paper, a truncated generalized functional linear model is studied and a maximum likelihood approach is used to estimate the model parameters. Finally, in a simulation study and two practical examples, the model and methods presented are implemented.

Life testing often is consuming a very long time for testing. Therefore, the engineers and statisticians are looking for some approaches to reduce the running time. There is a recommended method for reducing the time of failure, such that the stress level of the test units will increase, and then they will fail earlier than normal operating conditions. These approaches are called accelerated life tests. One of the most common tests is called the step stress accelerated life test. In this procedure, the stress applied to the units under the test is increased step by step at a predetermined time. The most important aspect to deal with the step stress model is the optimization of test design. In order to optimize the test plan, the best time to increase the level of stress should be chosen. In this paper, at first the step stress testing described. Then, this test is used for exponential lifetime distribution. Since life data are often not complete, this model is applied to type I censored data. By minimizing the asymptotic variance of the maximum likelihood estimator of reliability at time $xi$, the optimal test plan will be obtained. Finally, the simulation studies and one real data are discussed to illustrate the results. A sensitivity analysis shows that the proposed optimum plan is robust.

A sequence of functions (curves) collected over time is called a functional time series. Functional time series analysis is one of the popular research areas in which statistics from such data are frequently observed. The main purpose of the functional time series is to predict and describe random mechanisms that resulted in generating the data. To do so, it is needed to decompose functional time series into trend, periodic, and error components. However, we need to identify and recognize these components beforehand. Hence, in this study, a non-parametric method is presented for detecting and testing the existence of a process in a functional time series using record functions. Then, we implement and use this method for investigating the application of this method in a real functional time series. The effectiveness of this method for determining the trend in a set of real data on fertility rates in Australia has been investigated.

Unfortunately, in the recent few years, decreasing the level of educational motivation among students at various levels of education has been increasing and has plagued many of the university educational units in the country and this can have devastating and irreparable effects in the near future. In this research, using statistical analysis of the number of questionnaires, social factors are investigated on the motivation of Bushehr Persian Gulf University students and relationship between each of variables such as job status, the income of educated, gender, marital status, family income, presented content in the textbooks, university educational facilities, and weather conditions have been studied with students' educational motivation. The unequal two-stage cluster sampling method is used to collect the sample data and the required information and multivariate factor analysis used to examine the internal correlations of variables and to find the main factors by predicting the variance of variables by the factors. Also, the effect of variables on the students' educational motivation has been studied.

The Kumaraswamy distribution is a two-parameter distribution on the interval (0,1) that is very similar to beta distribution. This distribution is applicable to many natural phenomena whose outcomes have lower and upper bounds, such as the proportion of people from society who consume certain products in a given interval. In this paper, we introduce the family of Kumaraswamy-G distribution, and we detect its hazard rate function, reversible hazard rate function, mean residual life function, mean past life function, and the behavior of each of them. Also, we investigate the stochastic order concept of the family of Kumaraswamy-G distribution. Finally, in the form of a practical example, we analyze the suitability of the Kumaraswamy distribution for real data.

The Area under the ROC Curve (AUC) is a common index for evaluating the ability of the biomarkers for classification. In practice, a single biomarker has limited classification ability, so to improve the classification performance, we are interested in combining biomarkers linearly and nonlinearly. In this study, while introducing various types of loss functions, the Ramp AUC method and some of its features are introduced as a statistical model based on the AUC index. The aim of this method is to combine biomarkers in a linear or non-linear manner to improve the classification performance of the biomarkers and minimize the experimental loss function by using the Ramp AUC loss function. As an applicable example, in this study, the data of 378 diabetic patients referred to Ardabil and Tabriz Diabetes Centers in 1393-1394 have been used. RAUC method was fitted to classify diabetic patients in terms of functional limitation, based on the demographic and clinical biomarkers. Validation of the model was assessed using the training and test method. The results in the test dataset showed that the area under the RAUC curve for classification of the patients according to the functional limitation, based on the linear kernel pf biomarkers was 0.81 and with a kernel of the radial base function (RBF) was equal to 1.00. The results indicate a strong nonlinear pattern in the data so that the nonlinear combination of the biomarkers had higher classification performance than the linear combination.

In this paper, the difference between classical regression and fuzzy regression is discussed. In fuzzy regression, nonphase and fuzzy data can be used for modeling. While in classical regression only non-fuzzy data is used.The purpose of the study is to investigate the possibility of regression method, least squares regression based on regression and linear least squares linear regression method based on fuzzy weight calculation for non-fuzzy input and fuzzy output using symmetric triangular fuzzy numbers. Further reliability, confidence intervals and fitness fit criterion is presented for choosing the optimal model.Finally, by providing examples of the behavior of the proposed methods, the optimality of the regression hybrid model is shown by the least linear fuzzy squares.