Modeling engineering and medical lifetime data using a flexible extension of the XShanker distribution under censoring Authors: Amal S Hassan, Ghadah Alomani, Amer Ibrahim Al-Omari, Maisaa M Hassan Published: 2026 ABSTRACT AbstractThe complexity of today’s lifetime data necessitates more adaptable probability models. To address this need, we develop the Cosine Topp-Leone XShanker Distribution (CTL–XSD), a novel two-parameter extension of the XShanker distribution. We provide the mathematical properties of the new model, including its moments, incomplete moments, identifiability, quantiles, Lorenz curve, and Bonferroni curve. Using Type-II censoring, we employ both classical and Bayesian methods to estimate the parameters, survival function, and hazard function. Using Markov Chain Monte Carlo techniques, we derive Bayesian point and credible interval estimates under both symmetric and asymmetric loss functions. Additionally, we obtain the asymptotic confidence intervals based on the Fisher information matrix. We also provide a numerical study to evaluate the performance of the various estimators. We compare our findings with those from other models in the context of lifetime data analysis. The CTL–XSD outperformed the alternative models in terms of the Kolmogorov–Smirnov test statistic, with values of 0.98886 and 0.98438 in the context of engineering and medical datasets, respectively, with the lowest values in terms of all information criteria.Keywords: Bayesian estimation, Cosine Topp-Leone–G, Identifiability, Approximate confidence intervals, Minimum expected loss function, Metropolis–Hastings algorithmSubject terms: Engineering, Mathematics and computing Introduction Reliability engineering, medical survival analysis, risk assessment, and many other domains depend on accurate modeling of lifespan and time-to-event data for decision-making. The complicated nature of today’s systems, ranging from biological processes to mechanical components, produces failure data with a wide variety of behaviors, such as rising, falling, bathtub-shaped, and inverted bathtub-shaped hazard rates. With serious financial and safety effects, using an inadequately flexible probability model can result in skewed reliability predictions, incorrect maintenance schedules, incorrect warranty cost estimates, and inaccurate clinical prognoses. As a result, statistical distributions that are both tractable for inference and flexible to complicated real-world patterns are in consistently high demand, especially when data are incomplete due to censoring. Literature review Numerous generated families of distributions have been created as a result of notable advancements in the theory of probability distribution in recent years. Among the many well-known families are the Topp Leone-G (TL–G)1, the Marshall–Olkin–G2, the Kumaraswamy–G3, the Type II generalized TL–G4, the transmuted–G5, the power Lindley–G6, the alpha power transformed–G7, Lomax–G8, the Weibull–G9, the logistic–X10, the odd inverted TL–G11, the odd-Burr–G12. In the same vein, several probability models derived from generated families have been developed, including the Type II half–logistic Rayleigh Weibull distribution13 and a new two-parameter distribution introduced14. Other notable examples15–17. The recent advancements in the field of distribution theory have been based on trigonometric transformation, which provides flexibility in data modeling. Following the development of the sine–G18 and sec–G19 families, several extensions have been introduced, including the sine alpha power–G20, weighted sine–G21, and so on. In the same direction, cosine-based models have also been introduced, such as the cosine Kumaraswamy–G22, cosine inverse Lomax–G23, and so on. Following this, Nanga et al.24 introduced the cosine TL–G (CTL–G) family, which extends the cosine–G and TL–G structures, thus addressing both monotonic and non-monotonic failure rate characteristics while maintaining the simplicity of the model through the addition of only one parameter. Our interest in this article stems from the recently proposed CTL–G family, which has the following cumulative distribution function (CDF) and probability density function (PDF) and where, ρ is the shape parameter, g(.), G(.), and are the PDF, CDF, and survival function (ST) of the baseline distribution. Recently, Etaga et al.25 introduced the XShanker distribution (XSD), a one-parameter model for lifetime data. It is formulated as a two-component mixture distribution with a mixing proportion where the components are an exponential distribution and a Shanker distribution, both sharing the same scale parameter β. The CDF and PDF of the XSD are given by: and The framework of the Shanker distribution has been greatly expanded through recent research. For instance, Al-Omari and Dobbah26 proposed a two-parameter Shanker-gamma distribution and used maximum likelihood (ML) estimation for its analysis. John et al.27 also proposed a one-parameter inverse form of the Shanker distribution. Almetwally et al.28 recently discussed the three-parameter sine Kumaraswamy Shanker distribution under adaptive Type II hybrid progressive censoring. Research gap and motivation Several trigonometric distributions have recently emerged. The sine inverted exponentiated Weibull distribution29 was developed using the sine–G generator, with properties, Bayesian and ML estimation discussed. Similarly, cosine-Weibull30, sine power unit inverse Lindley31, log-cosine-power function32, alpha power cosine-Weibull33, sine power half-logistic34, sine Type II Topp-Leone Gompertz35, and sine inverted exponentiated Pareto36 distributions have been introduced, each featuring unique statistical properties and validated using ML or Bayesian estimation on real data. Additional contributions include a quantile regression framework based on the tangent exponentiated odd log-logistic Weibull distribution37 and a comparative analysis of estimation methods for the sine Topp-Leone Fréchet distribution38. A bivariate cosine Gaussian distribution was suggested in39. However, with the exception of a single study, these works primarily focus on complete samples, leaving a significant methodological gap for censored data. In real-world reliability and survival studies, censoring is common due to time, cost, and operational constraints. Estimation methods that perform well on complete data can become unstable or biased under censoring, undermining their practical applicability. This gap highlights the need for reliable inferential techniques designed for censored lifetime data. Novelty and contributions of the study The primary novelty is the introduction of the CTL–XSD, which overcomes the XSD’s limitation to strictly increasing hazard rates by modeling increasing, decreasing, reversed J-shaped, and upside-down shaped failure patterns with a parsimonious two-parameter structure. However, its mathematical complexity necessitates numerical methods, and its behavior under censoring schemes remains unexplored. To address the inference gap for censored data, we adapt standard ML and Bayesian, along with Markov chain Monte Carlo (MC–MC) methodologies to provide a complete inferential framework under Type–II censoring (T–IIC). We use MC–MC for the calculation of Bayesian credible intervals under symmetric and asymmetric losses, and asymptotic intervals using the Fisher information matrix (FM), offering the flexibility of using different methodologies for estimating the model. The following goals serve as the motivations behind this study: To develop a more flexible two-parameter model capable of capturing a broader range of hazard rate behaviors (increasing, decreasing, reversed J-shaped, upside-down). To develop a more flexible two-parameter model capable of capturing a broader range of hazard rate behaviors (increasing, decreasing, reversed J-shaped, upside-down). To derive its key statistical properties including moments, incomplete moments (INMs), quantile function (QF), identifiability, and moment generating function (MGF). To derive its key statistical properties including moments, incomplete moments (INMs), quantile function (QF), identifiability, and moment generating function (MGF). To obtain parameter estimates and reliability characteristics using both classical (ML) and Bayesian approaches under T–IIC. To obtain parameter estimates and reliability characteristics using both classical (ML) and Bayesian approaches under T–IIC. To construct associated asymptotic confidence intervals (Asy–CIs) and Bayesian credible intervals (BCIs) of the parameters and reliability measures. To construct associated asymptotic confidence intervals (Asy–CIs) and Bayesian credible intervals (BCIs) of the parameters and reliability measures. To evaluate the performance of all estimators via a comprehensive simulation study. To evaluate the performance of all estimators via a comprehensive simulation study. To empirically validate the CTL–XSD ‘s flexibility and fit by applying it to COVID-19 mortality and engineering datasets and comparing it with established alternatives. This empirical validation demonstrates its statistical efficiency and superior modeling capability for diverse real-world applications. To empirically validate the CTL–XSD ‘s flexibility and fit by applying it to COVID-19 mortality and engineering datasets and comparing it with established alternatives. This empirical validation demonstrates its statistical efficiency and superior modeling capability for diverse real-world applications. This work is structured as follows: Sects. “Formulation of the CTL–XSD” and “Structural properties” introduce the CTL–XSD and examine its fundamental statistical properties. Section “Maximum likelihood estimation” focuses on classical inference, deriving the ML estimates (MLEs) and the Asy–CIs under T–IIC schemes. Section “Bayesian estimation” explores a Bayesian approach, discussing the Bayesian estimates (BEs) under various loss functions, along with the highest posterior density (HPD) credible intervals. Section “Numerical study & results” provides a numerical analysis to validate the theoretical findings. Section “Real data analysis” demonstrates the model’s practical utility by analyzing two real-world datasets and comparing the CTL–XSD to other rival models. In the last section, the paper comes to an end. Formulation of the CTL–XSD The fundamental concepts of the CTL–XSD, such as the PDF, CDF, ST, hazard rate function (HZ), and cumulative HZ, are explained. To demonstrate the flexibility of the model, the graphical behaviors of the CTL–XSD’s PDF and HZ are compared with those of the baseline XSD for various parameter values in Figs. 1 and 2. The following formulation results from the substitution of Eqs. (3) and (4) into Eq. (1), which yields the CDF of the CTL–XSD: The PDF and HZ plots of the XShanker distribution. The PDF and HZ plots of the CTL–XSD. The simplified form of is written as follows: where and is the set of parameters, β is the scale parameter and ρ is the shape parameter. The PDF of the CTL–XSD, for is given by substituting Eqs. (3) and (4) in Eq. (2) The ST, HZ, and cumulative HZ of the CTL–XSD are as follows: and, The integration of the CTL-G generator with the baseline XSD is motivated by the need for a flexible modeling framework that can accommodate the rich data patterns present in the real world. While the baseline XSD model is restricted to strictly increasing hazard rates, the use of the trigonometric transformation in the CTL framework allows the CTL–XSD model to accommodate decreasing, inverted bathtub, reversed J-shaped, and increasing hazard rates. Such flexibility is particularly important for modeling the failure risks of mechanical devices, which often exhibit wear-in failures, and electronic devices, which exhibit a peak failure rate during the middle of their lifetime. Figure 1 presents the baseline XSD, which exhibits a monotonically decreasing PDF and a strictly increasing hazard rate for all values of its single parameter β. While effective for certain aging processes, this rigid structure limits its ability to model complex datasets requiring unimodal densities or non-monotonic failure rates (such as bathtub or inverted bathtub shapes). This limitation necessitates the proposed CTL–XSD model in Fig. 2, which introduces additional shape flexibility to accommodate an increasing, decreasing, reversed J-shaped, and upside-down shapes. These plots confirm the model’s flexibility in capturing a broad range of empirical failure patterns in engineering and medical research. Structural properties In this section, we reformulate the PDF (6) of the CTL–XSD using an appropriate series expansion, which makes it easier to analyze and derive statistical properties. Also, the QF, identifiability, moments, MGF, and sth INM are all derived. Quantile function In statistical analysis and modeling, the QF is useful because it provides a method for estimating percentiles and other summary statistics of a probability distribution. In reliability engineering, this is used to determine the warranty period by identifying the time by which a certain percentage of items are expected to fail. The QF, say Q(q), of the CTL–XSD is essentially the inverse of the CDF of the CTL–XSD as given below: Equation (9) can be reformulated as follows: where We can multiply Eq. (10) by on both sides to obtain: Equation (11) shows that is the Lambert ω function of the real . The QF of the CTL–XSD is therefore given by: In reliability engineering, the QF given in Eq. (12), is used to determine time-to-failure percentiles, such as the median lifetime Q(0.5) for q = 0.5, the lower quartile Q(0.25) for q = 0.25, and the upper quartile Q(0.75) for q = 0.75. In addition to, Bowley’s skewness, and Moors’ kurtosis are represented as Identifiability of model A key characteristic of statistics and a requirement for accurate inference is identifiability. If it is theoretically feasible to ascertain the distinct true values of a model’s underlying parameters given an endless number of observations, the model is said to be identifiable. A model is said to be non-identifiable if several sets of parameter values yield the exact same probability distribution for the observed data. No amount of data can differentiate between those parameter groups in such case. Parameter identifiability in the CTL–XSD ensures that its parameters can be uniquely and accurately estimated from observed data. This feature is crucial for accurate model fitting and inference because it keeps the model’s interpretability intact by preventing various parameter configurations from yielding identical distributions. To prove the parameter identifiability for the CTL–XSD, we show that if two CDFs are equal for all values of w, then their corresponding parameters must also be equal. That is If then where We begin by setting the CDFs equal to each other, then where and Since the cosine function is monotonic on the interval (which corresponds to the range of the inner terms as w goes from 0 to ∞ we can apply the inverse functions to both sides of Eq. (13) Consider the asymptotic behavior as and taking the natural logarithm of both sides of previous equation: Using the Taylor expansion for which for small values of x can be reduced to Then Eq. (14) can be written as: Taking the limit as , the exponential terms and must decay at the same rate for the equality to hold across the entire domain. If one side would vanish infinitely faster than the other. Thus, we must have To prove that we show that which implies that Then Eq. (13) can be written as follows: For this equality to hold for all w > 0 (where 0 < < 1), the exponents must be identical, i.e. which completes the proof. Density expansion To aid in additional theoretical research, this subsection derives a power series representation of the CTL–XSD. We use the sine series expansion Then, by employing expansion (15) in PDF (6), we obtain the following formula Using the binomial expansion in the last term of Eq. (16) gives: Again, using binomial expansion in Eq. (17), then PDF expansion has the following form where Moments The most crucial aspects and characteristics of a probability distribution are described by moments, which are crucial statistical measurements. They shed light on the shape, dispersion, and central tendency of the distribution. The sth moment of CTL–XSD, denoted by can be determined by using Eq. (18) as the following: where Γ (.) is the gamma function. The earlier mentioned Eq. (19) is a crucial tool for determining several moments measures. Specifically, the mean obtained when s = 1, represents the average expected lifetime of a system or component and offers a basic measure of central tendency. In reliability engineering and survival analysis, this statistic is a crucial point of reference that provides a useful foundation for assessing long-term performance, scheduling maintenance, and calculating warranty durations. The variance quantifies the variability or dispersion of failure times around the mean, indicating the consistency of product performance. The skewness measure measures the asymmetry of the distribution; positive skewness is typical in lifetime data, reflecting that most failures occur early while a few units survive much longer. Kurtosis measures a distribution’s peakiness and tail heaviness. Failure times are more equally distributed and there are fewer outliers when the kurtosis value is low, indicating a flatter, more uniform distribution with lighter tails. A high kurtosis value, on the other hand, denotes a sharper center peak and heavier tails, which in terms of reliability may imply a concentration of failures around a certain time combined with a larger probability of extreme early or late failures. Knowing kurtosis is useful in survival and reliability analysis because it makes it easier to spot odd failure patterns, evaluate the risk of outlier events, and choose the right models for data with noticeable peak or tail behavior. As shown above, it is possible to obtain many moments. Additionally, some numerical values of some moments are given in Table 1. It is evident that for this table that as the value of ρ and keeping β fixed, the first four moments and variance of CTL–XSD for different combinations of the shape parameter ρ and the scale parameter β . Higher-order moments and the variance exhibit consistent stability across parameter settings, suggesting that ρ primarily regulates tail behavior while β governs scale. The joint effects of ρ and β thus provide flexibility in modeling distributions with variable mean and tail properties, enhancing the applicability of CTL–XSD in settings sensitive to both skewness and scale. Table 1Numerical outcomes for the first four moments, skewness, and kurtosis.βρMeanSkewnessKurtosis1.50.50.39918040.29279760.32706510.48945400.13345262.1259249.6007000.60.44591170.34106410.38731280.58372110.14222692.0207158.9567070.70.48687820.38666150.44600150.67681920.14961111.9364388.4703730.80.52320790.42986180.50321060.76877520.15611531.8634188.0776250.90.55574610.47090240.55901700.85961770.16204871.7966167.7456711.70.50.33620310.21156180.20357300.26320600.098529282.1402689.7434520.60.37629430.24671580.24122370.31400550.10511842.0326329.0840450.70.41158740.27999740.27794450.3642090.11059321.9485248.5945100.80.44300270.3115920.31378090.41382930.11534061.8776888.2057290.90.47123310.34166210.34877690.46288020.11960141.8145547.881758 Numerical outcomes for the first four moments, skewness, and kurtosis. Table 1 shows that changes in the shape parameter ρ have a significant impact on how the distribution behaves. In particular, skewness and kurtosis steadily decrease as ρ rises from 0.5 to 0.9, suggesting less asymmetry and lighter tails, although central moment measurements steadily rise. On the other hand, variations in the scale parameter β mostly impact the core measures’ magnitude; for the same ρ levels, β = 1.7 consistently yields lower values than β = 1.5. Nevertheless, β has a negligible effect on skewness and kurtosis, as both metrics show comparable downward tendencies in many contexts. Overall, the results show that whereas β primarily controls the distribution’s central tendency without significantly changing its shape, ρ controls the distribution’s structural features. Figure 3 shows the 3D representation of the of CTL–XSD ‘s mean, variance, coefficient of skewness and kurtosis. The 3D plots of the mean, variance, skewness, and kurtosis. Together, the two parameters a shape parameter (ρ) and a scale parameter (β) define the properties of the distribution. The distribution changes significantly when both parameters rise: it becomes more concentrated with less variation, gets a stronger positive skew, and shows a higher kurtosis, which means the peak is sharper and the tails are heavier. This thorough control over variance, asymmetry, and tail behavior emphasizes how important these factors are in determining the general shape of the distribution. The MGF derived for the CTL–XSD serves as a powerful analytical tool that facilitates the computation of higher-order moments and enables the study of linear combinations of independent lifetime variables. In practical terms, the MGF offers a strong mathematical framework for the analysis and interpretation of various datasets, making it a vital tool in the domains of wireless communications, environmental sciences, and quality and reliability engineering. The MGF of the CTL–XSD is given by Incomplete moments The INM enhances traditional moments by providing deeper understanding of the distribution of a random variable, especially in specific regions. INM is used in survival analysis to predict the time until an event, such as the recurrence of a disease, within a specific timeframe. This is vital for medical prognosis and insurance risk assessment. The following formula can be used to find the sth INM of the CTL–XSD: where γ (.) is the lower incomplete gamma function. The Lorenz and Bonferroni curves, which are defined by and respectively, are relevant applications of the first INM, where wq can be calculated numerically from Eq. (12) for a given probability. Economic, reliability, demographic, insurance, and medical fields all benefit from these curves. Maximum likelihood estimation For a variety of reasons, including time constraints, financial constraints, or other resource constraints, the majority of data encountered in reliability or lifetime testing experiments are censored. Here, we go over how to estimate ST, and HZ of the CTL–XSD based on the T–IIC scheme. Under T–IIC scheme, a random sample of size n is placed on test, and the experiment is terminated upon observing the first r failures (r < n). The observed data consist of the ordered failure times while the remaining n - r units are right-censored at time, where denotes the i-th order statistic. The likelihood function based on the observed failure items of size (r) obtained from n items, is given by: where is written for simplified form instead of Substituting the CDF given in Eq. (5) and PDF given in Eq. (6), in the likelihood function of Eq. (20), we get where The log-likelihood function of Eq. (21) is as follows: The first partial derivative of Eq. (22) with respect to the parameters β and ρ are given by: where As shown in Eqs. (23) and (24), closed-form solutions for the MLEs of parameters β and ρ are not available in closed form. Consequently, we employ the well-established Nelder–Mead iterative algorithm within the R 4.5.1 software environment to numerically solve these likelihood equations. This computational approach is essential for deriving the precise parametric estimates, and , for the unknown parameters β and ρ . Using the invariance property of ML estimators, the ST and HZ can be found by substituting and in Eqs. (7) and (8) The most popular method for creating confidence intervals for parameters is the asymptotic normal distribution of MLEs40. The asymptotic variances and covariances of the MLEs and are provided by the entries of the inverse of the FM, represented by l, k = 1, 2 and It is difficult to obtain accurate closed-form expressions for these expectations. Consequently, the expectation operator is dropped and the observed FM is used in its place41. The observed FM’s entries are the log-likelihood function’s second partial derivatives. The asymptotic variance–covariance matrix for the MLEs is then obtained by inverting the observed FM ​ According to Lawless42, in specific regularity scenarios, the MLEs are basically multivariate normally distributed, with a matrix of For the parameters β and ρ, this property enables the construction of 100(1 − \varepsilon)% Asy–CIs and where is the upper th percentile of the standard normal distribution. Furthermore, the Asy–CIs of the ST and HZ are produced after obtaining their variances. The variances of the ST and HZ can be roughly estimated using the delta method43. The following methodology enables us to approximate the variances of these functions: and where and is the approximation of covariance matrix. The two-sided 100(1−\varepsilon)% Asy–CIs of ST and HZ are provided as follows: and Bayesian estimation The effectiveness of the Bayesian approach in the estimation of the parameters and the reliability characteristics under censoring is greatly dependent on the proper selection of the loss function and the informative prior distributions. Although the selection of the prior is one of the primary aspects of the Bayesian approach, there is no clear consensus in the literature with regard to the selection of the optimal prior for the unknown parameters. Instead, the selection of the prior is typically based on mathematical properties of the distribution. Gamma priors are considered a convenient choice because the gamma distribution has been shown to be flexible, meaning it takes on various shapes depending on its parameters, thus making it more effective for estimation compared to other complex distributions. Therefore, gamma priors are used in this study because of their flexibility compared to other distributions. In this section, the Bayesian approach is considered to obtain BEs and BCIs of the parameters ST, and HZ of the CTL–XSD. We assume independent gamma informative priors for the parameters ρ and β to incorporate prior knowledge while ensuring positivity. Under the same T–IIC scheme defined in Sect. “Maximum likelihood estimation”, the suggested estimates are produced by using gamma (informative prior, INP), and uniform (non-INP, N_INP) distributions. Suppose that the joint prior distribution of parameters β and ρ are independent gamma distributions with the following PDF where are the hyper-parameters. The hyperparameters are carefully chosen to reflect the degree of uncertainty; specifically, they are determined based on the empirical Bayes method to align the prior means with the maximum likelihood estimates44. This strategy is referred to as the elicitation of hyper-parameters. Keep in mind that in the case of N_INP, the BEs are obtained when the selected hyperparameters get close to zero. The following is the posterior distribution of the parameters β and ρ obtained by combining L(ψ ) of Eq. (21) with of Eq. (25) using Bayes’ theorem: The following is an expression for the conditional posterior densities of β and ρ The loss function in Bayesian inference captures the cost of errors in estimation. It is also a critical factor that determines the estimates. Depending on the situation at hand, a symmetric or asymmetric loss function may be preferred to handle overestimation and underestimation of the parameter of interest. In the case of a symmetric loss function, the squared error loss function (SELF) is considered. It is a popular loss function in Bayesian inference because of its analytical convenience. It treats overestimation and underestimation equally. It is a quadratic loss function that charges the square of the difference between the true parameter and the estimates. The SELF is defined as The BEs of the parameters, ST, and HZ, represented by , will be obtained under the SELF as follows: In practice, for many estimation problems, the costs of overestimation and underestimation are not equal. To deal with this situation, we use the asymmetric linear exponential loss function (LLF) proposed by Varian45. The BEs of , will be obtained under the LLF as follows: For instance, in engineering, underestimation of time to failure might be more costly than overestimation of time to failure. To provide additional flexibility in controlling asymmetry, we also employ the minimum expected loss function (MLF) presented by Zellner46 that is defined by: The BEs of , is obtained under the MLF as follows: Markov Chain Monte Carlo Since the conditional distributions and given in Eqs. (26) and (27) do not correspond to well-known forms, the Metropolis–Hastings (M–H) algorithm is employed to generate samples. A normal distribution is used as the proposal distribution to reduce the rejection rate and increase efficiency. To calculate the BEs and create credible intervals for β and ρ, Gibbs sampling and the more general M–H within-Gibbs sampler, which frequently makes use of a normal proposal distribution, are two important subclasses of MC–MC algorithms (refer to47,48 for more information on this strategy). A summary of the hybrid M–H and Gibbs sampler: Hybrid M–H and Gibbs sampler. Computational complexity and comparative analysis This sub-section addresses the computational aspects of the M–H algorithm and its comparison with the ML method. Although the Bayesian approach using the M–H algorithm is computationally more expensive than the deterministic Nelder-Mead optimization utilized for MLE, it offers significant practical benefits for survival analysis and reliability engineering, particularly with small sample sizes and under censoring schemes. The increased runtime of the Bayesian approach is due to the necessity of constructing a large Markov chain (typically 10,000 iterations) to accurately approximate the entire posterior distribution. This stochastic sampling process is inherently more labor-intensive than resolving a single optimization problem to find point estimates. However, this computational expense results in a richer inferential output, including full posterior densities and credible intervals that do not rely on large-sample asymptotic approximations. These advantages are especially critical in practical situations where data may be substantially restricted or sample sizes are small, such as in clinical trials with limited patient data or reliability testing of high-value components. In these contexts, the extra computational effort is frequently outweighed by the increased stability and accuracy of Bayesian results, particularly when using informative or weakly informative priors. Numerical study & results This section presents a simulation study to assess the performance of MLEs and BEs for the parameters (β ,ρ), ST, and HZ of the CTL–XSD. All simulation studies and numerical computations presented in this section were conducted using the R programming language (version 4.5.1). Simulation design and procedure A simulation study employing R software (version 4.5.1) was used to derive the various estimates, including MLEs, BEs under SELF, LLF with υ = (2, - 2), and MLF for the parameters, ST, and HZ of the CTL–XSD. The simulation process is outlined as follows: Generate 1000 random samples from the CTL–XSD with sample sizes n = 50,100, and 150 Generate 1000 random samples from the CTL–XSD with sample sizes n = 50,100, and 150 The chosen values for the number of failures r (70%, 90%, and 100%) follow the practical guidelines in reliability theory. The determination of the optimal r is a trade-off between statistical efficiency and experimental cost. As noted by Lawless49, higher values of r yield more Fisher information, leading to more precise estimates. However, r is also constrained by the ‘total time on test’; in practice, r is typically chosen such that r/n ≥ 0.5 to ensure a stable maximum for the likelihood function (Meeker and Escobar50). The chosen values for the number of failures r (70%, 90%, and 100%) follow the practical guidelines in reliability theory. The determination of the optimal r is a trade-off between statistical efficiency and experimental cost. As noted by Lawless49, higher values of r yield more Fisher information, leading to more precise estimates. However, r is also constrained by the ‘total time on test’; in practice, r is typically chosen such that r/n ≥ 0.5 to ensure a stable maximum for the likelihood function (Meeker and Escobar50). The chosen values of υ are υ = (2, - 2) and t = 0.8 for ST and HZ evaluation. The chosen values of υ are υ = (2, - 2) and t = 0.8 for ST and HZ evaluation. Parameter values for the simulations are: Parameter values for the simulations are: The MLEs were computed using the stats4, MASS, and bbmle packages. Numerical optimization was performed via the Nelder–Mead method, utilizing the optim function within the bbmle package51. The MLEs were computed using the stats4, MASS, and bbmle packages. Numerical optimization was performed via the Nelder–Mead method, utilizing the optim function within the bbmle package51. Bayesian inference under both INP and N_INP priors was performed using MC–MC techniques based on the M–H algorithm. Bayesian inference under both INP and N_INP priors was performed using MC–MC techniques based on the M–H algorithm. The hyper-parameters c1, c2, k1, and k2 for the gamma priors hyperparameters were determined by matching the mean and variance of to the mean and variance of the gamma distributions. The hyper-parameters c1, c2, k1, and k2 for the gamma priors hyperparameters were determined by matching the mean and variance of to the mean and variance of the gamma distributions. After resolving the two equations above, the estimated hyper-parameters can now be expressed as 8)The performance for β,ρ, ST, and HZ estimates was assessed through standard measures: bias, mean squared error (MSE), coverage probability (CP), and average interval length (AIL). The performance for β,ρ, ST, and HZ estimates was assessed through standard measures: bias, mean squared error (MSE), coverage probability (CP), and average interval length (AIL). 9)The MC–MC process used 10,000 iterations with the first 20% discarded as burn-in. The HD Interval package was used for HPD credible intervals, and the num Deriv package for numerical differentiation required in interval estimation. The MC–MC process used 10,000 iterations with the first 20% discarded as burn-in. The HD Interval package was used for HPD credible intervals, and the num Deriv package for numerical differentiation required in interval estimation. Comparative analysis of results The simulation results are presented in Tables 2, 3, 4, 5, 6, 7, 8, 9 and illustrated in Figs. 4, 5, 6, 7, 8, 9, 10, 11. In these tables and figures, the BE under LLF for υ = 2 denoted by BE_LLFP, while the BE under LLF for υ = - 2 denoted by BE_LLF. Also, BE under SELF, represented as BE_SELF, and BE under MLF, represented as BE_MLF. The following observations based on these tables and figures show the analysis of the estimated parameters’ performance: Table 2Simulation numerical values for set 1 in case of MLE and Bayesian estimates under INP.MLEINPnrParametersMSEAILCPMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.00560.285395.60.0013040.0013210.0012890.0012810.13615096.8ρ0.33682.130295.60.0452610.0485620.0427910.0437850.80727296.65ST0.00050.087797.50.0003740.0003660.0003820.0003920.07004298.65HZ0.00270.203497.50.0019930.0020730.0019210.0019210.16843496.3545β0.00390.239996.40.0009640.0009760.0009550.0009490.11927297ρ0.21701.741095.30.0389970.0417280.0369670.0377230.75343096.3ST0.00050.088497.50.0003580.0003510.0003650.0003740.06919698.2HZ0.00270.197997.50.0018750.0019410.0018170.0018530.16255297.350β0.00340.224495.90.0008520.0008630.0008420.0008330.11019596.65ρ0.18021.590996.10.0350200.0375410.0332000.0338180.69699096.05ST0.00050.081497.50.0003440.0003380.0003510.0003600.06801098.95HZ0.00260.192297.50.0018460.0019090.0017900.0018230.16087797.410070β0.00260.195496.20.0006500.0006560.0006440.0006360.09802796.9ρ0.11281.273795.70.0250240.0270480.0235300.0237220.58460197.2ST0.00030.063897.50.0001760.0001750.0001780.0001810.05086298.45HZ0.00130.137997.50.0009600.0009770.0009440.0009880.11969396.1590β0.00180.163996.40.0004580.0004630.0004540.0004490.08076296.75ρ0.08101.089596.10.0214680.0229310.0204260.0206350.53733896.7ST0.00030.063697.50.0001810.0001800.0001830.0001860.05049698.85HZ0.00130.137197.50.0009680.0009850.0009530.0009890.11791596.7100β0.00160.155396.30.0004220.0004270.0004170.0004090.07728397.1ρ0.07751.055496.20.0203520.0218540.0192380.0193470.53101097.4ST0.00030.064797.50.0001780.0001760.0001800.0001820.05146498.15HZ0.00130.146497.50.0009740.0009900.0009600.0010040.11978896.9150105β0.00160.156996.80.0004140.0004180.0004100.0004040.07789996.65ρ0.06530.976695.60.0169080.0181380.0160030.0160710.47843596.75ST0.00020.052097.50.0001220.0001210.0001230.0001240.04268698.05HZ0.00080.112797.50.0006790.0006870.0006720.0006990.10099496.95135β0.00120.134796.60.0003030.0003060.0003010.0002970.06668897.2ρ0.05070.868196.10.0138730.0148160.0131970.0132600.44270696.1ST0.00020.052397.50.0001150.0001140.0001150.0001160.04139898.65HZ0.00080.111197.50.0006280.0006350.0006220.0006480.09844797.95150β0.00100.124097.40.0002570.0002600.0002550.0002510.06123797.8ρ0.04450.809096.10.0127920.0137080.0121160.0121150.40978696.85ST0.00020.054097.50.0001100.0001090.0001110.0001110.04064898.6HZ0.00090.117297.50.0006250.0006300.0006190.0006560.09917797.55 Simulation numerical values for set 1 in case of MLE and Bayesian estimates under INP. Table 3Simulation numerical values for set 1 in case of MLE and Bayesian estimates under N_INP.nrParametersMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.0026340.0026100.0026630.0028630.204497.1ρ0.0635200.0646000.0634070.0678800.979097.7ST0.0004920.0004800.0005050.0005220.078898.9HZ0.0021850.0022860.0020950.0020450.174996.9545β0.0022220.0022060.0022420.0023810.185097.35ρ0.0610690.0624640.0605590.0644830.948096.3ST0.0004930.0004810.0005060.0005230.077199.45HZ0.0021860.0022810.0021010.0020370.174096.950β0.0020800.0020670.0020960.0022170.177997.45ρ0.0576130.0588850.0572280.0609420.940796.15ST0.0004970.0004840.0005100.0005260.076399.25HZ0.0022020.0022980.0021160.0020290.173696.4510070β0.0016980.0016870.0017110.0017940.163097.8ρ0.0491010.0507540.0482410.0506060.878497.2ST0.0002590.0002550.0002630.0002670.059398.6HZ0.0011120.0011380.0010880.0010860.127496.7590β0.0013870.0013810.0013950.0014460.145497.45ρ0.0441310.0455830.0433440.0452500.811696.15ST0.0002620.0002580.0002660.0002710.058598.9HZ0.0011380.0011650.0011140.0010980.128496.55100β0.0013180.0013130.0013240.0013670.141697.15ρ0.0439440.0454020.0431500.0449490.823997.75ST0.0002710.0002670.0002750.0002800.060998.3HZ0.0011920.0012190.0011680.0011540.130597.15150105β0.0013180.0013120.0013250.0013690.140497.75ρ0.0408810.0422730.0400970.0416130.785797.1ST0.0001900.0001880.0001920.0001950.051798.1HZ0.0008120.0008240.0008000.0008010.109197.6135β0.0010830.0010790.0010880.0011190.126998.15ρ0.0389330.0401640.0382140.0395910.764397.65ST0.0002020.0002000.0002040.0002070.053097.95HZ0.0008320.0008450.0008210.0008180.109497.4150β0.0009540.0009520.0009570.0009790.118096.85ρ0.0351570.0365320.0343010.0353900.716397.45ST0.0001890.0001870.0001910.0001930.051698.65HZ0.0008350.0008480.0008240.0008280.112297.75 Simulation numerical values for set 1 in case of MLE and Bayesian estimates under N_INP. Table 4Simulation numerical values for set 2 in case of MLE and Bayesian estimates under INP.MLEINPnrParametersMSEAILCPMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.00370.235695.60.0008800.0008870.0008730.0008730.112596.85ρ0.32082.114995.80.0467440.0495310.0447830.0459180.834296.7ST0.00030.065397.50.0001840.0001810.0001870.0001900.048498.95HZ0.00190.167697.50.0012520.0012930.0012130.0011540.132296.1545β0.00330.220696.30.0007890.0007970.0007830.0007790.106697.2ρ0.26561.929395.10.0426570.0452330.0408870.0418530.776596.2ST0.00030.061397.50.0001780.0001750.0001810.0001840.047998.8HZ0.00180.159897.50.0012350.0012750.0011970.0011390.131596.550β0.00250.1915960.0006060.0006100.0006020.0005980.095297.25ρ0.16491.538096.10.0322730.0345060.0306930.0311780.676696ST0.00010.048197.50.0000920.0000910.0000930.0000940.036199.5HZ0.00090.118897.50.0006490.0006600.0006380.0006420.098496.5510070β0.00170.160896.10.0004370.0004400.0004340.0004310.079097.55ρ0.11701.308595.50.0284350.0300710.0273140.0277640.631397.20ST0.00020.047697.50.0000960.0000950.0000970.0000980.036598.80HZ0.00100.116797.50.0006690.0006810.0006580.0006540.098496.5090β0.00160.152796.50.0004070.0004110.0004040.0003980.077796.55ρ0.11211.2683960.0267400.0284820.0254970.0258450.618096.35ST0.00010.046797.50.0000920.0000920.0000930.0000940.036198.80HZ0.00100.120797.50.0006610.0006720.0006510.0006540.099297.15100β0.00160.153596.40.0003950.0003980.0003930.0003890.076796.80ρ0.09441.173595.90.0229170.0243870.0218680.0221390.565297.20ST0.00010.038997.50.0000640.0000640.0000640.0000650.030698.70HZ0.00060.096097.50.0004640.0004700.0004590.0004660.083098.15150105β0.00120.132396.80.0002970.0002990.0002950.0002930.067197.10ρ0.07321.042495.60.0195650.0207850.0187140.0189160.531796.45ST0.00010.038897.50.0000610.0000610.0000610.0000620.030198.15HZ0.00060.096597.50.0004390.0004440.0004340.0004420.082497.6135β0.00100.121996.60.0002510.0002530.0002500.0002470.059997.10ρ0.06400.969295.90.0175560.0187080.0167480.0168870.499096.40ST0.00010.039797.50.0000570.0000570.0000580.0000580.029099.15HZ0.00060.099297.50.0004280.0004320.0004240.0004390.081297.70150β0.00370.235697.30.0008800.0008870.0008730.0008730.112596.95ρ0.32082.114995.90.0467440.0495310.0447830.0459180.834296.55ST0.00030.065397.50.0001840.0001810.0001870.0001900.048498.35HZ0.00190.167697.50.0012520.0012930.0012130.0011540.132296.10 Simulation numerical values for set 2 in case of MLE and Bayesian estimates under INP. Table 5Simulation values for set 2 in case of Bayesian estimates under N_INP.nrParametersMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.0020830.0020670.0021020.0022340.181297.70ρ0.0687530.0693140.0692590.0734591.025296.65ST0.0002480.0002430.0002530.0002590.053799.40HZ0.0014570.0015180.0014010.0012710.140895.8045β0.0018020.0017910.0018160.0019130.166197.05ρ0.0667080.0676060.0668160.0705760.997697.25ST0.0002480.0002430.0002540.0002600.052799.20HZ0.0014630.0015220.0014090.0012670.138596.7550β0.0017190.0017090.0017310.0018170.162796.95ρ0.0638430.0645800.0641170.0677700.990396.5ST0.0002490.0002440.0002540.0002600.053098.85HZ0.0014720.0015310.0014180.0012620.137695.8010070β0.0013840.0013750.0013950.0014560.148196.50ρ0.0554680.0566730.0551430.0576730.936196.45ST0.0001350.0001330.0001360.0001380.041998.65HZ0.0007660.0007830.0007500.0007100.104496.7590β0.0011420.0011370.0011490.0011890.133696.65ρ0.0501010.0512980.0497040.0517950.859697.40ST0.0001350.0001330.0001370.0001390.041498.95HZ0.0007830.0008000.0007660.0007150.104796.90100β0.0011080.0011040.0011140.0011470.129597.20ρ0.0504510.0517380.0500080.0520270.885896.15ST0.0001390.0001370.0001410.0001430.042998.3HZ0.0008150.0008330.0007990.0007480.107797.25150105β0.0010930.0010880.0010990.0011340.127598.10ρ0.0478520.0490880.0473430.0491060.851096.35ST0.0000990.0000980.0001000.0001010.036498.75HZ0.0005650.0005740.0005570.0005350.090997.25135β0.0009070.0009030.0009110.0009360.116897.35ρ0.0455940.0466230.0451980.0468100.828497.40ST0.0001050.0001040.0001060.0001070.037498.25HZ0.0005900.0005990.0005810.0005530.092197.35150β0.0008170.0008140.0008200.0008390.108397.80ρ0.0419920.0432210.0414080.0427550.776796.55ST0.0000980.0000980.0000990.0001000.036997.65HZ0.0005840.0005930.0005760.0005540.093297.10 Simulation values for set 2 in case of Bayesian estimates under N_INP. Table 6Simulation numerical values of set 3 in case of MLE and Bayesian estimates under INP.MLEINPnrParametersMSEAILCPMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.01200.414995.50.0022650.0023030.0022340.0022260.181396.65ρ0.84493.374795.70.0656770.0688890.0636680.0653601.010997.85ST0.00150.154397.50.0010420.0010150.0010710.0011200.122898.10HZ0.01250.448897.50.0091540.0099820.0085340.0087900.366297.2545β0.00830.350796.30.0018030.0018280.0017840.0017830.160597.60ρ0.55032.773095.20.0603280.0630400.0587050.0601420.965196.20ST0.00150.154097.50.0009640.0009420.0009880.0010280.11969800HZ0.01150.423097.50.0079670.0085650.0075260.0079060.350397.4050β0.00730.329195.90.0016730.0016980.0016530.0016470.152797.35ρ0.46052.5431960.0573120.0599520.0557420.0570470.932696.85ST0.00150.145097.50.0009550.0009330.0009780.0010170.116597.60HZ0.01140.429997.50.0078360.0084170.0074020.0077060.342296.9510070β0.00530.282296.10.0012310.0012430.0012210.0012180.135897.35ρ0.28142.010995.70.0456430.0481780.0440420.0449440.821696.85ST0.00070.104597.50.0005170.0005110.0005250.0005360.088399.05HZ0.00520.295797.50.0042160.0044020.0040750.0041820.255496.7590β0.00380.237996.40.0009270.0009360.0009200.0009170.116696.90ρ0.20351.7267960.0421650.0441700.0409290.0417310.794098.10ST0.00080.104297.50.0005250.0005190.0005320.0005430.088397.90HZ0.00500.277397.50.0039130.0040600.0038020.0039070.247397.25100β0.00350.226996.30.0008690.0008790.0008600.0008550.112897.20ρ0.19631.679095.90.0392920.0414640.0378950.0386300.77669700ST0.00080.111397.50.0005230.0005170.0005300.0005410.088598.30HZ0.00540.297597.50.0039580.0040990.0038520.0039560.244597.20150105β0.00340.225796.80.0008390.0008460.0008330.0008300.111196.35ρ0.16221.539495.60.0344810.0362780.0333550.0339390.696796.55ST0.00050.085597.50.0003690.0003660.0003720.0003780.073997.75HZ0.00350.234597.50.0029100.0029900.0028490.0029050.212197.70135β0.00250.195896.40.0006500.0006550.0006460.0006440.099196.40ρ0.12771.377195.90.0309540.0325450.0299730.0304750.659697.45ST0.00050.085697.50.0003530.0003500.0003560.0003610.073296.75HZ0.00320.224397.50.0025920.0026550.0025450.0026050.196397.90150β0.00220.180897.30.0005660.0005710.0005620.0005580.093097.55ρ0.11261.286196.10.0290070.0306950.0278900.0283150.645897.20ST0.00050.089597.50.0003520.0003500.0003550.0003600.073297.15HZ0.00350.226497.50.0026040.0026590.0025640.0026370.198498.25 Simulation numerical values of set 3 in case of MLE and Bayesian estimates under INP. Table 7Simulation numerical values of set3 in case of Bayesian estimates under N_INP.nrParametersMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.0035340.0035360.0035420.0036300.237297.30ρ0.0780000.0782560.0790850.0818611.084697.25ST0.0011710.0011390.0012060.0012650.128199.00HZ0.0110430.0120170.0103390.0110760.401596.3545β0.0031540.0031550.0031620.0032350.220697.25ρ0.0763960.0767080.0773440.0800261.056197.95ST0.0011410.0011110.0011730.0012280.124698.75HZ0.0103020.0111280.0097060.0104120.387996.8550β0.0030770.0030800.0030810.0031460.217797.30ρ0.0751260.0754700.0760320.0786471.064597.75ST0.0011430.0011130.0011750.0012290.124698.15HZ0.0102420.0110700.0096350.0102240.390597.4510070β0.0023180.0023100.0023300.0023790.189097.20ρ0.0673780.0683390.0675180.0696241.031497.25ST0.0005960.0005870.0006060.0006220.093299.00HZ0.0052170.0054520.0050420.0052430.278696.5590β0.0020180.0020130.0020260.0020610.174597.65ρ0.0634910.0644070.0636190.0655170.984597.70ST0.0005990.0005890.0006090.0006250.091598.45HZ0.0049520.0051720.0047840.0049240.270796.00100β0.0019840.0019810.0019890.0020190.175896.60ρ0.0636620.0646700.0637300.0655210.997296.40ST0.0006220.0006130.0006320.0006480.095298.20HZ0.0051030.0053190.0049390.0050760.276796.60150105β0.0018830.0018780.0018900.0019190.167997.15ρ0.0601050.0613160.0599160.0614890.959297.25ST0.0004290.0004240.0004340.0004420.079398.00HZ0.0036220.0037330.0035360.0036130.234797.15135150β0.0016000.0015950.0016070.0016310.156697.30ρ0.0593520.0603410.0593270.0608620.936497.15ST0.0004330.0004280.0004380.0004460.079897.60HZ0.0032080.0032980.0031390.0032180.220097.75β0.0543170.0554250.0541490.0554330.883197.30ρ0.0004270.0004230.0004320.0004390.079497.80ST0.0032980.0033860.0032320.0033120.223097.35HZ0.0022430.0023020.0021940.0022190.180997.10 Simulation numerical values of set3 in case of Bayesian estimates under N_INP. 135 150 Table 8Simulation numerical values for set 4 in case of MLE and BEs under INP.MLEINPnrParametersMSE_AILCPMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.01130.403795.60.0017940.0018180.0017740.0017690.16222596.80ρ1.59084.624895.60.0739880.0766440.0726970.0742721.07269996.70ST0.00080.114497.50.0005000.0004890.0005120.0005290.08333897.75HZ0.00850.372197.50.0061570.0066730.0057410.0057220.295703970045β0.00790.342996.30.0014890.0015050.0014760.0014760.14560297.70ρ1.02453.778495.30.0694760.0718540.0683890.0697531.03652196.35ST0.00090.113697.50.0004600.0004510.0004700.0004840.08121698.65HZ0.00840.358097.50.0054620.0058590.0051420.0052320.28361996.8050β0.00700.3223960.0013990.0014160.0013860.0013830.14172097.20ρ0.85213.453595.90.0668970.0693060.0657570.0670321.01280896.70ST0.00080.109697.50.0004580.0004490.0004680.0004820.08017699.05HZ0.00810.353997.50.0054140.0058030.0050990.0051370.28122696.8510070β0.00510.274796.10.0010090.0010160.0010040.0010030.12348797.20ρ0.51252.709395.50.0568420.0591560.0556610.0566890.92415497.15ST0.00040.080997.50.0002490.0002460.0002520.0002570.06027098.40HZ0.00380.246197.50.0028680.0029860.0027710.0028000.20938497.1090β0.00360.232596.40.0007890.0007950.0007850.0007840.10695796.75ρ0.36492.308095.70.0541370.0561390.0531270.0540490.90052797.45ST0.00050.080697.50.0002570.0002540.0002600.0002640.06109998.50HZ0.00390.232497.50.0027780.0028800.0026930.0027160.20772297.60100β0.00330.222396.30.0007610.0007680.0007560.0007530.1057469700ρ0.35262.246095.90.0515520.0538490.0502070.0511560.87852197.75ST0.00050.082497.50.0002570.0002540.0002600.0002640.06104897.8HZ0.00410.261897.50.0028070.0029060.0027250.0027480.20217596.65150105β0.00320.219896.90.0007180.0007220.0007150.0007130.10461696.40ρ0.29122.059795.60.0471980.0491840.0460700.0468530.84180497.50ST0.00030.065397.50.0001830.0001820.0001850.0001870.05135797.80HZ0.00250.192497.50.0020280.0020810.0019840.0020020.17468198.05135β0.00240.191596.30.0005840.0005870.0005820.0005810.09567497.45ρ0.22861.840195.80.0435670.0452970.0426470.0433170.82159396.55ST0.00030.065997.50.0001750.0001740.0001770.0001780.05092098.25HZ0.00240.194597.50.0018470.0018920.0018100.0018310.16671898.50150β0.00210.177297.30.0005150.0005190.0005130.0005110.08984496.15ρ0.20011.711996.10.0409120.0428430.0397520.0404060.77473397.50ST0.00030.067597.50.0001760.0001750.0001770.0001790.05135997.30HZ0.00260.202197.50.0018750.0019150.0018410.0018750.16715097.50 Simulation numerical values for set 4 in case of MLE and BEs under INP. Table 9Simulation result values for set 4 in case of Bayesian estimates under N_INP.nrParametersMSE_SELFMSE_LLFMSE_LLFPMSE_MLFAILCP5035β0.0025180.0025260.0025150.0025530.19679697.05ρ0.0813310.0814010.0827380.0843161.09567997.50ST0.0005590.0005450.0005730.0005920.08795799.25HZ0.0071490.0077450.0066790.0069470.31903096.645β0.0022950.0023000.0022950.0023300.18971596.65ρ0.0809890.0810270.0823310.0839231.08543196.70ST0.0005470.0005350.0005610.0005790.08514398.65HZ0.0068240.0073520.0064070.0066750.31113796.7550β0.0022450.0022520.0022440.0022750.18686397.30ρ0.0788410.0788610.0802340.0817511.07657297.05ST0.0005440.0005310.0005570.0005750.08489899.05HZ0.0067680.0072940.0063480.0065300.31507397.3510070β0.0015910.0015890.0015960.0016180.15821896.85ρ0.0729570.0735950.0736040.0749561.06815196.80ST0.0002860.0002820.0002900.0002950.06352798.70HZ0.0034000.0035460.0032810.0033520.22561696.6090β0.0014340.0014330.0014370.0014530.14788496.95ρ0.0691380.0697160.0697950.0710091.01391797.25ST0.0002840.0002800.0002880.0002930.06274998.45HZ0.0033040.0034440.0031880.0032180.22089996.95100β0.0014350.0014350.0014370.0014520.14867997.45ρ0.0703910.0710200.0710270.0722081.02266697.20ST0.0002950.0002910.0002990.0003040.06511798.25HZ0.0034090.0035470.0032930.0033220.22765697.75150105β0.0013180.0013160.0013220.0013360.14055397.25ρ0.0674670.0681940.0679310.0690301.01785496.85ST0.0002060.0002040.0002080.0002110.05420698.25HZ0.0023600.0024290.0023010.0023200.18715996.90135β0.0011300.0011280.0011340.0011460.13251797.65ρ0.0674790.0681490.0679690.0690511.01443697.05ST0.0002090.0002070.0002120.0002140.05539697.65HZ0.0021990.0022590.0021490.0021720.18076097.75150β0.0010840.0010830.0010860.0010960.12883597.70ρ0.0627870.0636470.0630400.0639960.95260597.30ST0.0002050.0002030.0002070.0002090.05415098.35HZ0.0022430.0023020.0021940.0022190.18093197.10 Simulation result values for set 4 in case of Bayesian estimates under N_INP. The MSE behavior of ρ estimates for set 1 in case INP. The MSE behavior of estimates for set 1 in case N_INP. The MSE of behavior β estimates for set 1 in the case INP. The MSE behavior of estimates for set 1 in case N_INP. The MSE of behavior ρ estimates for set 1 in case of INP and N_INP. The MSE behavior of ρ estimates for set 1 and set 2 in the case of INP and N_INP. The MC–MC plots of β and ρ with iteration 10,0000 for set 1 in case INP. The plots of β and ρ with iteration 10,0000 for set 1 in case N_ INP. Figure 4 shows the MSE values for several estimate techniques under Set 1 (β = 0.5, ρ = 1.3). All sample sizes (105, 135, and 150) show that the MLE consistently yields a significantly higher MSE than the Bayesian-based approaches. The BEs (SELF_INP, LLF_INP, LLFP_INP, and MLF_INP), on the other hand, show similar and significantly lower MSE values, with only minor differences between them. This demonstrates that the Bayesian techniques are more resilient and efficient than the standard MLE, especially as the sample size increases. Figure 5 shows the MSE performance of several estimators for Set 1 (β = 0.5, ρ = 1.3) under the N_INP scheme. All sample sizes (105, 135, and 150) show consistently higher MSE values when using the MLE approach, but the Bayesian-based estimates (SELF_N_INP, LLF_N_INP, LLFP_N_INP, and MLF_N_INP) yield significantly lower and more stable MSE results. As the sample size grows, the MSE decreases, highlighting the Bayesian techniques’ superior efficiency and dependability over MLE. Figure 6 shows, with regard to β, the MSE of many different estimates under Set 1 (β = 0.5, ρ = 1.3). It is evident that for all sample sizes (35, 45, and 50), the MLE consistently produces larger MSE values than the Bayesian-based methods. On the other hand, even in small-sample situations, the BEs (SELF_INP, LLF_INP, LLFP_INP, and MLF_INP) produce significantly lower and more stable error levels, demonstrating their superior efficiency and robustness. In Fig. 7, the MSE of the MLE consistently has the highest values, showing it is the least effective estimate for β among the tested methods. Its MSE decreases as the sample size increases from 35 to 50. The other four estimates (MSE_SELF_N_INP, MSE_LLF_N_INP, MSE_LLFP_N_INP, and MSE_MLF_N_INP) show significantly lower MSE values. The performance of these four superior methods is very similar to each other, with their MSE values clustered tightly together. Their MSE decreases as sample size increase from 35 to 50. For all five estimates, the MSE consistently decreases as the sample size increases, which is expected for consistent estimates. Based on Fig. 8, the MSE of the INP case consistently provides a more accurate and efficient estimation of the parameter ρ compared to the N_INP case. The N_INP approach leads to a considerable increase in estimation error across all methods. This suggests that for a sample size of 135, the INP method is the preferred approach for estimating ρ with higher precision. Further investigation would be needed to understand the underlying reasons for the significant performance difference between the two cases. Figure 9 illustrates the performance of the parameter estimates for Set 1 and Set 2 under both INP and N_INP frameworks. As the parameter ρ increases from 1.3 to 1.5 (with β held constant), the MSE increases across all estimation methods, suggesting that higher values of ρ are more challenging to estimate accurately. Notably, the LLFP_INP method consistently yields the lowest MSE in both datasets. Overall, the simulation results indicate that BE (both INP and N–INP) outperforms the MLE. Based on Fig. 10, both parameters in the INP case appear to have converged to their stationary distributions, according to the diagnostic plots. With approximate central values of, the posterior estimates are clearly characterized. Strong mixing and effective sampling are demonstrated by the parameter β and ρ. The convergence, accurate posterior estimation, and high efficiency of the sampler all suggest that the results are robust and suitable for drawing conclusions about the model’s parameters. Figure 11, the MC–MC simulation for the ‘N_INP’ model for β and ρ was successful. All diagnostic plots show excellent performance. The chains have converged, the posterior distributions are well-defined, and the samples are efficient and have low autocorrelation. Regarding the interval estimation performance, the results in Tables 2, 3, 4, 5, 6, 7, 8, 9 demonstrate that both Asy–CIs and BCIs perform efficiently. Specifically, as the sample size n increases, the AIL of the intervals consistently decreases, reflecting enhanced precision in parameter estimation. Moreover, the CP for all parameters remains remarkably close to the nominal level of 95%. It is also worth noting that BCIs generally yield shorter AIL compared to Asy–CIs, especially in small sample sizes, which highlights the superiority of the Bayesian approach under the considered censoring schemes results presented in Tables 2, 3, 4, 5, 6, 7, 8, 9. Real data analysis This section illustrates the effectiveness of CTL–XSD in a data-fitting scenario using two actual datasets. The CTL–XSD is constructed with some other models, including, power inverse Topp–Leone distribution (PITLD)52, sine inverse Weibull distribution (SIWD)53, the inverse Lindley distribution) ILD)54, sine Fréchet distribution (SFD)55, inverse Weibull distribution (IWD), inverse power Lindley distribution (IPLD)56. This claim is supported using the Akaike information criteria (AIC), Hannan-Quinn IC (HQIC), corrected AIC (CAIC), Bayesian IC (BIC), and, with the best-fitting model selected accordingly. Additionally, we consider goodness-of-fit metrics such as Kolmogorov–Smirnov (KS), Cramer-von Mises (W*), Anderson–Darling (A*), and the p-value of the KS (P_KS) test. The model with the lowest values across different statistics and measures is regarded as the best fit, but the largest p-value is also considered to determine the most suitable distribution. Modeling medicine and engineering data The dataset originates from the medical field and is related to the mortality rate due to COVID-19 for 30 days (31st March to April 30, 2020) and was recorded for the Netherlands. This data was discussed by57. The observations are; 14.918, 10.656, 12.274, 10.289, 10.832, 7.099, 5.928, 13.211, 7.968, 7.584, 5.555, 6.027, 4.097, 3.611, 4.960, 7.498, 6.940, 5.307, 5.048, 2.857, 2.254, 5.431, 4.462, 3.883, 3.461, 3.647, 1.974, 1.273, 1.416, and 4.235. The second dataset originates from the reliability engineering field and is taken from tests representing the waiting times (in minutes) before service of 100 bank customers58. Its variability and skewed distribution make it ideal for engineering statistical applications, such as reliability and queueing analysis. Applying the model to this data demonstrates its ability to capture realistic system behavior in practical engineering scenarios. The second data set is given by: 0.8, 0.8, 1.3, 1.5, 1.8, 1.9, 1.9, 2.1, 2.6, 2.7, 2.9, 3.1, 3.2, 3.3, 3.5, 3.6, 4, 4.1, 4.2, 4.2, 4.3, 4.3, 4.4, 4.4, 4.6, 4.7, 4.7, 4.8, 4.9, 4.9, 5, 5.3, 5.5, 5.7, 5.7, 6.1, 6.2, 6.2, 6.2, 6.3, 6.7, 6.9, 7.1, 7.1, 7.1, 7.1, 7.4, 7.6, 7.7, 8, 8.2, 8.6, 8.6, 8.6, 8.8, 8.8, 8.9, 8.9, 9.5, 9.6, 9.7, 9.8, 10.7, 10.9, 11, 11, 11.1, 11.2, 11.2, 11.5, 11.9, 12.4, 12.5, 12.9, 13, 13.1, 13.3, 13.6, 13.7, 13.9, 14.1, 15.4, 15.4, 17.3, 17.3, 18.1, 18.2, 18.4, 18.9, 19, 19.9, 20.6, 21.3, 21.4, 21.9, 23, 27, 31.6, 33.1, 38.5. Some graphical representations of these datasets are shown in Figs. 12 and 13 for both real data sets. They include quantile–quantile (QQ) plots, violin plots, box plots, total time on test plots (TTT), histograms, and kernel density estimates. A few fundamental non-paramec charts for the medical dataset 1. A few fundamental non-parametric charts for the reliability engineering dataset 2. The MLEs of the competing models, together with their standard errors (SEs) for the recommended data sets, are presented in Table 10. Table 10Estimated parameters for both datasets using the SEs for all models.DataModelρSEβSEICTL–XSD0.782030.221890.221610.03616SPITLD0.209760.092070.819580.27795PITLD0.362720.161670.819630.27739SIWD1.381530.17993319.27499224.28033ILD46.4067995.687111.196652.50242SFD1.447290.2046464.416747.57462IWD153.1681374.746571.301030.13318IPLD1.868390.266531413.655121388.53152IICTL–XSD0.491530.068220.113840.01144SPITLD0.186510.047491.809530.37394PITLD0.323490.082451.823460.37305SIWD0.915500.062166.603510.80157ILD17.9830523.146260.314770.42641SFD0.915490.062167.860390.71310IWD6.532310.891341.162910.07995IPLD1.151710.080657.232170.90256 Estimated parameters for both datasets using the SEs for all models. Figures 14 and 15 illustrate the CTL–XSD distribution log-likelihood curve parameters for both real data sets, demonstrating the estimators’ maximizing. Profile-log-likelihood for parameters and of the medical dataset1. Profile-log-likelihood for parameters and of the reliability engineering dataset. A comprehensive list of model selection criteria is shown in Table 11. It contains values for the HQIC, BIC, CAIC, and AIC. The table also offers the values of the test statistics W*, A*, KS, and KS p. Table 11Numerical values for analyzing the first dataset and the second dataset.DataModelAICBICCAICHQICW*A*KSKS pICTL–XSD157.40890160.21130157.85340158.305500.023720.176130.076600.98886PITLD241.34110243.69720241.91260241.966200.048230.353020.118130.89115SIWD257.77970260.13580258.35110258.404800.046060.333870.308930.02049ILD241.23220243.58830241.80360241.857300.049790.365600.110490.93131SFD249.81790252.17400250.38930250.443000.055520.407630.200740.28820IWD244.34930246.70540244.92070244.974300.081420.588620.137340.75587IICTL–XSD637.44550642.65590637.56920639.554200.029450.193210.045890.98438PITLD656.96360662.17390657.08730659.072300.205301.414540.083110.49444SIWD676.71360681.92390676.83730678.822300.264071.777300.176860.00384ILD656.96360662.17390657.08730659.072300.205301.414530.083110.09445SFD672.76200677.97230672.88570674.870700.383222.504610.116680.13133IWD673.55790678.76830673.68160675.666700.394032.569290.117980.12356 Numerical values for analyzing the first dataset and the second dataset. For first data the CTL–XSD achieved the lowest information criteria values (AIC, BIC, CAIC, HQIC). It also had a very high KS p-value = 0.98886, indicating excellent fit. Other models had extremely low p-values < 0.05, so they do not adequately fit the data. For second data the CTL–XSD is the best model according to all information criteria. Its KS p-value = 0.98438, showing very strong agreement with the data. Some models like PITLD and SFD had moderate p-values (0.494440., 0.13133), but they were still inferior to CTL–XSD. Models such as SPITLD, SIWD, p < 0.05, meaning they are unsuitable. Finally, across both datasets, the CTL–XSD consistently provided the best fit, outperforming all competing models in terms of both information criteria and goodness-of-fit tests. Figures 16 and 17 show plots of the estimated CDF, histograms with the estimated PDF and probability–probability (PP) plots of the suggested model, respectively. The computed curves fit the empirical objects well. Estimated PDF, CDF, and PP plots for the COVID-19 mortality data. Estimated PDF, CDF, and PP plots for the reliability engineering data. Practical implications The comparative advantage of the CTL–XSD is demonstrated through its performance evaluation against related models, including the XSD and other competing distributions. A key strength of the CTL–XSD lies in its enhanced flexibility in capturing a wide range of hazard-rate shapes, particularly inverted bathtub and reversed J-shaped patterns, which frequently arise in complex engineering and medical datasets. Unlike simpler models, the incorporation of the CTL–G component provides greater adaptability, especially in the tail regions of the distribution. This added flexibility enables a more accurate representation of the underlying failure mechanism, as reflected in the substantial improvements observed in goodness-of-fit statistics. For the two datasets analyzed, the superior performance of the CTL–XSD further highlights its robustness in the presence of extreme and high-outlier observations. The flexible tail behavior, controlled by the model’s shape parameter, allows it to effectively accommodate extreme values without distorting parameter estimation. As a result, the presence of high outliers does not disproportionately influence the model fit, making the CTL–XSD particularly suitable for real-world applications characterized by heavy-tailed or extreme data. This robustness is visually supported by the PDF, CDF, and PP plots (Figs. 16 and 17), where the CTL–XSD closely follows the empirical distribution, including in the tail regions, while maintaining overall fit accuracy. Although the datasets considered represent real-world extreme scenarios, the resilience of the CTL–XSD against high outliers is further confirmed by its superior goodness-of-fit measures, including AIC, BIC, and K–S statistics. We have added a brief clarification emphasizing that the model’s flexible tail structure enables it to accommodate extreme observations more effectively than traditional distributions. The strong empirical performance of the CTL–XSD in both engineering and medical applications underscores its practical relevance. In engineering reliability analysis, the model facilitates accurate failure-time modeling, supporting improved maintenance planning and optimized system lifecycle management. In medical survival analysis, it assists in identifying high-risk periods, thereby enhancing clinical decision-making and resource allocation. Overall, the flexibility and robustness of the CTL–XSD establish it as a reliable and powerful statistical tool for data-driven decision-making in critical real-world contexts. Conclusion Lifetime distributions are fundamental tools for modeling reliability and survival data. This study introduces a new hybrid model, the CTL–XSD distribution, formed by combining the XSD with the CTL generator. Some key statistical properties are derived to demonstrate the model’s flexibility. Parameter estimation and reliability measures (survival and hazard functions) are developed under T–IIC using both ML and Bayesian approaches. Asy–CIs and BCIs under SELF, LLF, and MLF are constructed via MC–MC simulation. A comprehensive simulation study compares the estimators using MSE, AIL, and CP. Results show that Bayesian estimators based on N_INP priors generally outperform MLEs, producing smaller MSEs and shorter interval lengths while maintaining CPs near 95%. The model’s applicability is illustrated using engineering and COVID-19 survival datasets, where it successfully captures wear-out and inverted bathtub-shaped hazard behaviors and outperforms competing models. A limitation of this study is the focus on T–IIC only. Future work may consider progressive, hybrid, or adaptive censoring schemes, as well as E-Bayesian estimation under alternative loss functions. Acknowledgements The authors express their gratitude to Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R226), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. Abbreviations Akaike information criterion Average interval length Asymptotic confidence intervals Bayesian credible intervals Bayesian estimates Bayesian information criteria Corrected AIC Cumulative distribution function Cosine Topp–Leone XShanker Distribution Coverage probability Fisher information matrix Hannan–Quinn information criteria Hazard rate function Highest posterior density The inverse Lindley distribution Informative prior Kolmogorov–Smirnov Linear exponential loss function Incomplete moments Inverse Weibull distribution Inverse power Lindley distribution Maximum likelihood ML estimates Metropolis–Hastings Markov chain Monte Carlo Moment generating function Mean squared error Minimum expected loss function Non-INP Probability density function Power inverse Topp–Leone distribution Probability–probability Quantile function Quantile–quantile Standard error Sine Fréchet distribution Survival function Squared error loss function Sine inverse Weibull distribution Type–II censoring Topp Leone–G Total time on test XShanker distribution Author contributions Amal S. Hassan: Methodology; Writing—original draft; Writing—review; Formal analysis, Software & editing; Ghadah Alomani, Methodology; Writing—original draft; Formal analysis; Software & editing; Amer Ibrahim Al-Omari, Methodology; Writing—original draft; Writing—review; Formal analysis, Software & editing; Maisaa M. Hassan, Methodology; Writing—original draft; Writing—review; Formal analysis, Software & editing; All authors have read and agreed to the published version of the manuscript. Funding This research was funded by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R226), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. Data availability All data generated or analysed during this study are included in these published articles [[10.3390/e22030346](https:/doi.org/10.3390/e22030346) and [10.1016/j.rinp.2021.104012](https:/doi.org/10.1016/j.rinp.2021.104012)]. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated or analysed during this study are included in these published articles [[10.3390/e22030346](https:/doi.org/10.3390/e22030346) and [10.1016/j.rinp.2021.104012](https:/doi.org/10.1016/j.rinp.2021.104012)].