Thailand Statistician

Estimation Methods for the Modified Generalized Linear Exponential Distribution and Its Applications in Engineering Data

M.G.M. Ghazal — Wed, 30 Sep 2026 00:00:00 +0700

This paper investigates different methods for estimating the unknown parameters of the modified generalized linear exponential distribution (MGLED), a flexible lifetime model that can accommodate bathtub-shaped, increasing, and decreasing hazard rate functions. The MGLED is of practical importance in reliability and engineering applications because it includes several well-known lifetime distributions as special cases, such as the exponential, Rayleigh, Weibull, modified Weibull, and generalized linear exponential distributions. Five classical estimation methods are considered, namely maximum likelihood estimation, least squares estimation, weighted least squares estimation, Cramer von-Mises estimation, and Anderson-Darling estimation. The performance of these estimation methods is assessed through two real engineering data sets and an extensive simulation study. For the real data applications, the fitted models are compared using non-parametric goodness-of-fit measures, including the Kolmogorov-Smirnov statistic with its corresponding P-value, the Anderson-Darling statistic, and the Cramer von-Mises statistic. In addition, parametric model selection criteria, including the Akaike information criterion, the corrected Akaike information criterion, and the Hannan-Quinn information criterion, are used to evaluate the adequacy of the MGLED relative to several competing lifetime models. The simulation study compares the proposed estimation methods in terms of mean squared error and average absolute bias. The numerical results show that the MGLED provides a better fit than the competing models for the considered engineering data sets, while the simulation results indicate that the maximum likelihood estimation and Anderson-Darling estimation methods generally perform better than the other estimation methods as the sample size increases.

Novel Estimators for Enhanced Finite Population Variance Estimation with Auxiliary Data

Housila P. Singh, Shakti Prasad — Wed, 30 Sep 2026 00:00:00 +0700

This paper proposes two novel classes of efficient estimators for estimating the finite population variance under the Simple Random Sampling Without Replacement (SRSWOR) scheme by utilizing auxiliary information in sample surveys. The bias and mean squared error (MSE) expressions of the proposed classes of estimators are derived up to the first order of approximation. The optimum conditions under which the proposed estimators achieve minimum MSE are also obtained. Furthermore, regions of preference are established to determine the conditions under which the proposed estimators perform better than the existing estimators available in the literature. An empirical study based on three real data sets is carried out to evaluate the efficiency and practical applicability of the proposed estimators.

Dynamic Vertical Pricing and Strategic Timing: Analytical and Numerical Evidence from the Thai Palm Oil Supply Chain

Wichai Witayakiattilerd — Wed, 30 Sep 2026 00:00:00 +0700

This paper develops a dynamic game-theoretic model of a vertically decentralized agricultural supply chain with endogenous timing and three-layer pricing. Focusing on a factory–wholesaler–retailer structure, we derive closed-form subgame perfect Nash equilibria and establish existence and uniqueness under explicit economic conditions. Under linear demand and constant marginal costs, decentralized sequential pricing yields an exact efficiency ratio: total channel profit equals 7/16 of the integrated benchmark, implying a 56.25% structural profit loss. This loss admits a quadratic decomposition as a curvature-scaled squared retail-pricing wedge generated by uninternalized vertical pricing externalities. We further embed the pricing game in a continuous-time gradient adjustment system and establish global asymptotic stability using a Lyapunov argument. Spectral analysis links convergence speed to demand curvature, while computational sensitivity analysis illustrates robustness over economically admissible parameter regions. Calibration to Thai crude palm oil trade data (2015–2024) anchors demand at moderate inelasticity and indicates that the predicted distortion is quantitatively meaningful. Overall, the framework explains how vertically fragmented agricultural markets can be dynamically stable yet structurally inefficient.

Bayesian Prediction of Double Seasonal ARMA Processes

Ayman A. Amin — Wed, 30 Sep 2026 00:00:00 +0700

Recent developments in time series analysis extended autoregressive moving average (ARMA) models to accommodate datasets exhibiting two seasonal cycles, leading to the formulation of double seasonal ARMA (DSARMA) models. While Bayesian estimation methods have been explored for these models, Bayesian prediction has not yet been formally developed. In the current work, we advance the existing literature by establishing a Bayesian predictive framework for DSARMA models. Our approach assumes normally distributed error terms and employs both normal-gamma and Jeffreys priors on the model parameters. Under these assumptions, we derive closed-form predictive distributions for future, unobserved values, showing that they follow a univariate-t distribution. This result enables efficient multistep-ahead forecasting. To assess the forecasting accuracy of the current proposed Bayesian prediction methodology, we conduct a comprehensive simulation study and apply the framework to forecasting hourly electricity load across several European countries.

Logarithmic Transformed Topp-Leone Family of Distributions: Properties, and Applications

Suraj Yadav, Sanjay Kumar Singh, Abhimanyu Singh Yadav — Wed, 30 Sep 2026 00:00:00 +0700

This study introduces a novel family of continuous distributions called the logarithmic transformed Topp-Leone family. Key statistical properties of the proposed family are investigated, including the hazard rate, reverse hazard rate, odds function, moments, moment-generating function, residual and reverse residual life moments, entropy measures, order statistics, probability-weighted moments, and quantile function. Parameter estimation is performed using both maximum likelihood and maximum product spacing methods. To demonstrate the flexibility of the proposed family, the Lindley distribution is considered the baseline distribution, and the logarithmic transformed Topp-Leone Lindley distribution is derived, for which comprehensive statistical properties are also derived. The performance of the proposed estimators is evaluated through comprehensive Monte Carlo simulations.

Two-Stage Sampling Inspection Plan based on Mean for Lindley Distribution

Sudhansu S. Maiti, Shovan Biswas — Wed, 30 Sep 2026 00:00:00 +0700

This study develops Two-Stage Sampling (TSS) inspection plans for lifetime quality characteristics based on the mean, assuming non-normal distributions such as the exponential and the Lindley distribution. The lot mean (µ) is used as the quality measure. Unlike conventional approaches that focus solely on a fixed sample size, the proposed plan emphasizes efficiency by reducing inspection effort through a two-stage decision mechanism, resulting in time and cost savings. The optimal plan parameters are obtained using a non-linear optimization approach under specified acceptable and rejectable quality levels. A comparative analysis of sample size and Average Sample Number (ASN) between the TSS and Single Sampling (SS) plans is presented. An illustrative example is included to demonstrate the applicability of the proposed method. The results indicate that when the underlying data follow the Lindley distribution, the proposed TSS plan provides a more efficient decision-making framework with reduced average inspection requirements. Therefore, the proposed plan is recommended for use in such situations.

A Power Generalized DUS–Kumaraswamy Distribution with Applications to Parallel System Reliability

Ann Sania, Vellamattathil Mani Chacko — Wed, 30 Sep 2026 00:00:00 +0700

In reliability engineering, modeling parallel systems requires lifetime distributions that combine hazard rate flexibility with structural tractability for reliability analysis. In this article, we introduce a new lifetime distribution obtained by applying the Power Generalized DUS transformation to the Kumaraswamy distribution, referred to as the PGDUS-Kumaraswamy (PGDUS-K) distribution. The proposed model provides enhanced flexibility for modeling parallel systems and accommodates various monotone and non-monotone hazard rate shapes. Several statistical properties of the PGDUS-K distribution are derived, including moments, entropy measures, quantile function, and stress-strength reliability expressions for single- and multi-component systems. The model is invariant and closed under exponentiation, and preserves its form under maxima, making it suitable for parallel systems. Parameter estimation is carried out using method of maximum likelihood, method of maximum product of spacings, method of moments, method of least squares, and Bayesian approaches. A simulation study is conducted to evaluate the performance of the estimators. The practical utility of the model is illustrated through real data analysis, where it demonstrates superior goodness-of-fit compared with competing models based on standard model selection criteria. In addition, the derived stress–strength expressions enable direct reliability computation, demonstrating practical applicability.

Bayesian Reliability Estimation for Coupon Distribution Using Type II Censored Constant-Stress Accelerated Life Tests: Methods and Applications

RA Rather, Afaq Ahmad — Wed, 30 Sep 2026 00:00:00 +0700

In this paper, Bayesian approach is used to analyse the parametric inference based on tampered random variable for constant-stress life tests under type-II censored data. The lifetime distribution of experimental units follows the two parameter Coupon distribution with η and ϕ as scale and shape parameters respectively. The Markov Chain Monte Carlo approach with Squared error loss, Generalized entropy loss function, and Linear exponential loss function are used to derive Bayes estimators. An important consideration in Bayesian estimating problems is the selection of the loss function. As a result, the Bayesian estimators for the parameters are derived using various loss functions and a comparison with the standard maximum likelihood approach is carried out. Finally, a real data set is analyzed to demonstrate the proposed estimation procedure.

Determinants of Domestic Violence Against Husbands in Palestine

Mohamed Hammad Morzk, Farouk Tammam Shoaib, Noura Anwar Abdel-Fatah — Wed, 30 Sep 2026 00:00:00 +0700

The study aimed to investigate the demographic and socioeconomic factors that affect the husbands’ odds of experiencing domestic violence in Palestine. Violence in the Palestinian Society Survey is the main source of data, it was conducted by the Palestinian Central Bureau of Statistics in 2019. The sample was 5114 ever-married women. The descriptive analysis was used to describe the demographic and socioeconomic variables that affect the prevalence of violence against husbands. Binary logistic regression was used to study the determinants of violence against husbands. Findings revealed that younger wives are more likely to practice physical and psychological violence and the increase in the number of children within the family leads to an increase in the chances of husbands experiencing physical and sexual violence. The low education and unemployment status of wives were protective factors for husbands against domestic violence. The wives’ exposure to any form of domestic violence accompanies an elevated odd of husbands experiencing the same type of domestic violence that their wives face. Laws and policies should consider the male partner's problems from exposure to different types of domestic violence. Context-oriented premarital counseling programs for youth.

Using D-Vine Copula for Modeling of Dependency of Categorical Data

Zara Omar, Bushra Shamshad, Junaid Siddiqi — Wed, 30 Sep 2026 00:00:00 +0700

This research introduces a new approach to D-Vine copula in the measuring of association between categorical variables. This paper used the D-Vine method of defining the tree diagram of the categorical data to find the dependence of the variable. D-Vine copula shows that the association between variables is strong under the simple and conditional probabilities and distribution. It is used to generate tree diagrams using nodes and edges of probabilities and conditional probability, which illustrate probability distributions and association between the variables. The purpose of the present study is (1) to find a linkage between morning show viewers having impact of brand endorsement (2) to empirically test the impact theory by examining the effect of the morning shows and brands image. Data is collected using an online survey, of the viewers of morning shows, the sample of 1750 observation is used to test the factors. Results of the D-Vine copula shows that the morning shows have a positive and strong impact on the brand image and purchase of the fashion products. As predicted, brand endorsement is positively associated with morning shows and celebrity images.

A New Two-Parameter Discrete Weighted Lindley Distribution: Properties, Estimation and Applications

Manal Mahmoud Salem, Moshira Ahmed Ismail — Wed, 30 Sep 2026 00:00:00 +0700

A new two-parameter discrete weighted Lindley distribution is proposed. The proposed distribution can be seen as a generalization of one parameter discrete Lindley distribution which has only increasing hazard rate whereas our proposed distribution has increasing and bathtub shaped hazard rates. In addition, this distribution can describe over, equi and under dispersed data. Some properties of the proposed distribution are obtained such as cumulative distribution function, hazard rate function, second hazard rate function, mean residual life function, reversed hazard rate function, accumulated hazard rate function as well as order statistics. In addition, an analytical study of the shapes of the hazard rate function is provided. Maximum likelihood estimation of the parameters of the proposed distribution is considered. The results obtained in this study are illustrated using both simulated and four real data sets.

Financial Crisis Detection in Thailand Based on Currency Crisis Index Approach Using Artificial Neural Network

Wed, 30 Sep 2026 00:00:00 +0700

The 1997 financial crisis in Thailand had a variety of adverse effects, ranging from the depreciated baht, negative economic growth, increased unemployment, to declining stock prices. Thailand took a long time to recover from these impacts. Therefore, financial crisis detection is needed to prevent or minimize the impact of the crisis. This study aims to build a model that can predict the financial crisis in Thailand using multilayer perceptron (MLP). Four different optimizations were applied to obtain the optimal model, namely SGD, Adam, Nadam, and AdaBound. The model employs eleven macroeconomic indicators as input variables and uses the perfect signal derived from the Currency Crisis Index (CCI) as the target variable. This study used the CCI approach to determine Thailand’s crisis conditions from January 1991 to December 2022. Based on the results, the best model for predicting Thailand’s financial crisis is the MLP model with Nadam optimization trained using 11-6-1 network architecture, learning rate of 0.2, batch size of 128, and epoch of 100. The model produced an accuracy of 99.44% on validation data and 98.61% on testing data. The model prediction results suggest a low likelihood of a financial crisis in Thailand from January 2023 to December 2024.

Three and Four Component R-, A- and D-optimal Allocations for an Additive Quadratic Mixture Model

Bushra Husain, Fariha Aslam — Wed, 30 Sep 2026 00:00:00 +0700

Optimal mixture experiments are a powerful tool for systematically exploring and optimizing the composition of mixtures. In this paper, we have presented a new additive quadratic mixture model and have obtained R-, A- and D-optimal allocations for three and four component mixtures. The obtained designs have been checked for optimality using the equivalence theorem.

Triangular Model and Two of Its Variants Under Stratified Sampling

Eman Mansour, Reda Mazloum, Sally Abdelfatah — Wed, 30 Sep 2026 00:00:00 +0700

The non-randomized response (NRR) technique is a procedure that can be used in surveys with sensitive questions, and it has many advantages over the randomized response (RR) technique such as the ease of implementation and the ability to be used in self-administrated surveys as no random device is required. Many NRR models were developed to increase the efficiency and/or increase the respondents’ cooperation such as the triangular model and the alternating triangular model. In this paper, we extended the triangular model and two of its variants; the alternating triangular model and the triangular model under non-compliance; to stratified random sampling. The maximum likelihood estimation method was employed to estimate the population proportion possessing the sensitive attribute under each of the three extended models. Unbiased estimators of the sensitive proportion were derived and their sampling variances were obtained. In addition, the optimal allocation of the sample size for the three extended models was derived.

On Identifying the Factors Responsible for Frontal Lobe Dysfunction of the Human Brain: An Application of Multi-Event Survival Analysis to Wisconsin Card Sorting Data

Anirban Dutta, Jonali Gogoi, Dibyojyoti Bhattacharjee — Wed, 30 Sep 2026 00:00:00 +0700

The frontal lobe, located in the anterior region of the brain, plays a crucial role in facilitating cognitive functions. Its vulnerability to injuries due to its location can lead to impaired decision-making, increased risk-taking behaviour, and diminished working memory performance. The extent of frontal lobe dysfunction is often difficult to detect with advanced imaging techniques; thus, researchers rely on low-tech neuropsychological evaluations to assess cognitive deficits and functional impairments. One such tool is the Wisconsin Card Sorting Test (WCST), where participants encounter many errors throughout the test’s evaluation process. An individual’s performance in WCST can be considered a case of a multi-event survival process, given that the participant may experience numerous failures during the test. The study investigated the influence of various socio-demographic and lifestyle factors on the frequency of errors and, consequently, the degree of front lobe dysfunction. The Andersen-Gill (AG) model is implemented, and it revealed that alcohol abstinence and self-confidence in one’s working memory significantly affect the frequency of errors observed in the WCST. Future research should focus on exploring additional factors to enhance the models' ability to capture significant patterns in error risks and shed light on various dimensions of assessing individuals’ cognitive flexibility.