The document describes a three-parameter generalized inverse Weibull (GIW) distribution that can model failure rates. Key properties of the GIW distribution include:
- It reduces to the inverse Weibull distribution when the shape parameter γ equals 1.
- Its probability density function, survival function, and hazard function are defined.
- Formulas are provided for the moments, moment generating function, and Shannon entropy of the GIW distribution.
- Methods are described for maximum likelihood estimation of the GIW distribution parameters from censored lifetime data.
This document summarizes a talk on supersymmetric Q-balls and boson stars in (d+1) dimensions. It introduces Q-balls and boson stars as non-topological solitons stabilized by a conserved Noether charge. It discusses properties like existence conditions and the thin wall approximation for Q-balls. For boson stars, it covers different models and properties like rotating and charged boson stars. The document also discusses applications like the AdS/CFT correspondence and holographic superconductors using boson stars in anti-de Sitter spacetime.
This document discusses prior selection for mixture estimation. It begins by introducing mixture models and their common parameterization. It then discusses several types of weakly informative priors that can be used for mixture models, including empirical Bayes priors, hierarchical priors, and reparameterizations. It notes challenges with using improper priors for mixture models. The document also discusses saturated priors when the number of components is not known beforehand. It covers Jeffreys priors for mixtures and issues around propriety. It proposes some reparameterizations of mixtures, like using moments or a spherical reparameterization, that allow proper Jeffreys-like priors to be defined.
This document discusses using the Wasserstein distance for inference in generative models. It begins by introducing ABC methods that use a distance between samples to compare observed and simulated data. It then discusses using the Wasserstein distance as an alternative distance metric that has lower variance than the Euclidean distance. The document covers computational aspects of calculating the Wasserstein distance, asymptotic properties of minimum Wasserstein estimators, and applications to time series data.
Heatwaves are affecting the human mortality and the global climate adversely. There are many existing ad hoc definitions of heatwaves in literature. Here, we are identifying a notion of
intensity and duration of the possible heatwaves and finding a proper definition using their distributions. We are modelling these two factors and attempting to find the predictive ability of our model. We have also obtained some preliminary results on obtaining the probability distribution of duration and intensity for a sequence of stationary random variables. We are exploring these concepts on a heat data of Atlanta from the year 1991-2012. The next goals are to relax the stationary assumption and obtaining more general results.
We consider a general approach to describing the interaction in multigravity models in a D-dimensional space–time. We present various possibilities for generalizing the invariant volume. We derive the most general form of the interaction potential, which becomes a Pauli–Fierz-type model in the bigravity case. Analyzing this model in detail in the (3+1)-expansion formalism and also requiring the absence of ghosts leads to this bigravity model being completely equivalent to the Pauli–Fierz model. We thus in a concrete example show that introducing an interaction between metrics is equivalent to introducing the graviton mass.
This document discusses priors for mixture models. It introduces weakly informative priors like symmetric empirical Bayes priors and dependent priors. Improper independent priors are problematic for mixtures. Reparameterization techniques are discussed to define proper Jeffreys priors, including expressing components as local perturbations, using moments, and spherical reparameterization. Specific examples for Gaussian and Poisson mixtures show valid reparameterizations that lead to proper posteriors.
This document discusses Bayesian model comparison in cosmology using population Monte Carlo methods. It provides background on key questions in cosmology that can be addressed using cosmic microwave background data from experiments like WMAP and Planck. Population Monte Carlo and adaptive importance sampling methods are introduced to help approximate Bayesian evidence for different cosmological models given the immense computational challenges of working with this cosmological data.
This document provides solutions to logistics management assignment problems. It addresses 6 problems related to network flows, facility location, and vehicle routing. For problem 1, it shows that the difference between the sum of outdegrees and indegrees in a network is always zero. For problem 2, it proves Goldman's majority theorem and provides an algorithm to solve the 1-median problem. The solutions utilize concepts like isthmus edges, node weights, and network separation.
This document summarizes a talk on supersymmetric Q-balls and boson stars in (d+1) dimensions. It introduces Q-balls and boson stars as non-topological solitons stabilized by a conserved Noether charge. It discusses properties like existence conditions and the thin wall approximation for Q-balls. For boson stars, it covers different models and properties like rotating and charged boson stars. The document also discusses applications like the AdS/CFT correspondence and holographic superconductors using boson stars in anti-de Sitter spacetime.
This document discusses prior selection for mixture estimation. It begins by introducing mixture models and their common parameterization. It then discusses several types of weakly informative priors that can be used for mixture models, including empirical Bayes priors, hierarchical priors, and reparameterizations. It notes challenges with using improper priors for mixture models. The document also discusses saturated priors when the number of components is not known beforehand. It covers Jeffreys priors for mixtures and issues around propriety. It proposes some reparameterizations of mixtures, like using moments or a spherical reparameterization, that allow proper Jeffreys-like priors to be defined.
This document discusses using the Wasserstein distance for inference in generative models. It begins by introducing ABC methods that use a distance between samples to compare observed and simulated data. It then discusses using the Wasserstein distance as an alternative distance metric that has lower variance than the Euclidean distance. The document covers computational aspects of calculating the Wasserstein distance, asymptotic properties of minimum Wasserstein estimators, and applications to time series data.
Heatwaves are affecting the human mortality and the global climate adversely. There are many existing ad hoc definitions of heatwaves in literature. Here, we are identifying a notion of
intensity and duration of the possible heatwaves and finding a proper definition using their distributions. We are modelling these two factors and attempting to find the predictive ability of our model. We have also obtained some preliminary results on obtaining the probability distribution of duration and intensity for a sequence of stationary random variables. We are exploring these concepts on a heat data of Atlanta from the year 1991-2012. The next goals are to relax the stationary assumption and obtaining more general results.
We consider a general approach to describing the interaction in multigravity models in a D-dimensional space–time. We present various possibilities for generalizing the invariant volume. We derive the most general form of the interaction potential, which becomes a Pauli–Fierz-type model in the bigravity case. Analyzing this model in detail in the (3+1)-expansion formalism and also requiring the absence of ghosts leads to this bigravity model being completely equivalent to the Pauli–Fierz model. We thus in a concrete example show that introducing an interaction between metrics is equivalent to introducing the graviton mass.
This document discusses priors for mixture models. It introduces weakly informative priors like symmetric empirical Bayes priors and dependent priors. Improper independent priors are problematic for mixtures. Reparameterization techniques are discussed to define proper Jeffreys priors, including expressing components as local perturbations, using moments, and spherical reparameterization. Specific examples for Gaussian and Poisson mixtures show valid reparameterizations that lead to proper posteriors.
This document discusses Bayesian model comparison in cosmology using population Monte Carlo methods. It provides background on key questions in cosmology that can be addressed using cosmic microwave background data from experiments like WMAP and Planck. Population Monte Carlo and adaptive importance sampling methods are introduced to help approximate Bayesian evidence for different cosmological models given the immense computational challenges of working with this cosmological data.
This document provides solutions to logistics management assignment problems. It addresses 6 problems related to network flows, facility location, and vehicle routing. For problem 1, it shows that the difference between the sum of outdegrees and indegrees in a network is always zero. For problem 2, it proves Goldman's majority theorem and provides an algorithm to solve the 1-median problem. The solutions utilize concepts like isthmus edges, node weights, and network separation.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
On the stability and accuracy of finite difference method for options pricingAlexander Decker
1) The document discusses finite difference methods for pricing options, specifically the implicit and Crank-Nicolson methods.
2) It analyzes the stability and accuracy of each method. The Crank-Nicolson method is found to be more accurate and converges faster than the implicit method.
3) Numerical examples are provided to demonstrate the convergence properties of each method compared to the true Black-Scholes value. The results show that the Crank-Nicolson method converges to the solution faster than the implicit method as the time and price steps approach zero.
A NEW APPROACH TO M(G)-GROUP SOFT UNION ACTION AND ITS APPLICATIONS TO M(G)-G...ijscmcj
In this paper, we define a new type of M(G)-group action , called M(G)-group soft union(SU) action and M(G)-ideal soft union(SU) action on a soft set. This new concept illustrates how a soft set effects on an M(G)-group in the mean of union and inclusion of sets and its function as bridge among soft set theory, set theory and M(G)-group theory. We also obtain some analog of classical M(G)- group theoretic concepts for M(G)-group SU-action. Finally, we give the application of SU-actions on M(G)-group to M(G)-group theory.
Polyadic systems and their representations are reviewed and a classification of general polyadic systems is presented. A new multiplace generalization of associativity preserving homomorphisms, a ’heteromorphism’ which connects polyadic systems having unequal arities, is introduced via an explicit formula, together with related definitions for multiplace representations and multiactions. Concrete examples of matrix representations for some ternary groups are then reviewed. Ternary algebras and Hopf algebras are defined, and their properties are studied. At the end some ternary generalizations of quantum groups and the Yang-Baxter equation are presented.
Time series analysis, modeling and applicationsSpringer
- The document proposes a novel composition forecasting model based on the Choquet integral with respect to a completed extensional L-measure and M-density fuzzy measure.
- It compares the performance of this model to other composition forecasting models, including ones based on extensional L-measure, L-measure, lambda-measure, and P-measure, as well as ridge regression and multiple linear regression models.
- Experimental results on grain production time series data show that the proposed Choquet integral composition forecasting model with completed extensional L-measure and M-density outperforms the other models.
Selection of bin width and bin number in histogramsKushal Kumar Dey
This document discusses various statistical methods for determining the optimal number of bins and bin width in histograms. It reviews methods such as Sturges' rule, Scott's rule, Bayesian optimal binning, and others. It presents the concepts behind each method and compares their performance in simulations. The simulations involved generating data from distributions like the chi-square, normal, and uniform, and comparing the methods based on how close their histogram estimates were to the true densities. Scott's rule and Freedman-Diaconis modification generally performed best at minimizing errors. The document also discusses generalizing some methods to multivariate histograms and potential areas for further research.
An Altering Distance Function in Fuzzy Metric Fixed Point Theoremsijtsrd
The aim of this paper is to improve conditions proposed in recently published fixed point results for complete and compact fuzzy metric spaces. For this purpose, the altering distance functions are used. Moreover, in some of the results presented the class of t-norms is extended by using the theory of countable extensions of t-norms. Dr C Vijender"An Altering Distance Function in Fuzzy Metric Fixed Point Theorems" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-1 | Issue-5 , August 2017, URL: http://www.ijtsrd.com/papers/ijtsrd2293.pdf http://www.ijtsrd.com/mathemetics/other/2293/an-altering-distance-function-in-fuzzy-metric-fixed-point-theorems/dr-c-vijender
Polyadic systems and their representations are reviewed and a classification of general polyadic systems is presented. A new multiplace generalization of associativity preserving homomorphisms, a 'heteromorphism' which connects polyadic systems having unequal arities, is introduced via an explicit formula, together with related definitions for multiplace representations and multiactions. Concrete examples of matrix representations for some ternary groups are then reviewed.
MSC classes: 16T05, 16T25, 17A42, 20N15, 20F29, 20G05, 20G42, 57T05
comments on exponential ergodicity of the bouncy particle samplerChristian Robert
The document summarizes recent work on establishing theoretical convergence rates for the bouncy particle sampler (BPS), a non-reversible Markov chain Monte Carlo algorithm. The main results show that under certain conditions on the target distribution, including having exponentially decaying tails, the BPS exhibits exponential ergodicity. A central limit theorem is also established. The analysis considers different cases for thin-tailed, thick-tailed, and transformed target distributions.
This document discusses various methods for approximating marginal likelihoods and Bayes factors, including:
1. Geyer's 1994 logistic regression approach for approximating marginal likelihoods using importance sampling.
2. Bridge sampling and its connection to Geyer's approach. Optimal bridge sampling requires knowledge of unknown normalizing constants.
3. Using mixtures of importance distributions and the target distribution as proposals to estimate marginal likelihoods through Rao-Blackwellization. This connects to bridge sampling estimates.
4. The document discusses various methods for approximating marginal likelihoods and comparing hypotheses using Bayes factors. It outlines the historical development and connections between different approximation techniques.
This document discusses weak isomorphism and isomorphism between fuzzy graphs. Some key points:
1. Weak isomorphism preserves the weights of nodes but not necessarily the weights of edges, while isomorphism preserves weights of both nodes and edges.
2. If two fuzzy graphs are isomorphic, they have the same order (number of nodes) and size (sum of edge weights).
3. Isomorphism between fuzzy graphs is proved to be an equivalence relation as it is reflexive, symmetric and transitive.
4. Weak isomorphism is proved to be a partial order relation as it satisfies reflexivity, anti-symmetry and transitivity.
5. Some properties of self-complement
Many areas of machine learning and data mining focus on point estimates of key parameters. In transportation, however, the inherent variance, and, critically, the need to understand the limits of that variance and the impact it may have, have long been understood to be important. Indeed, variance and other risk measures that capture the cost of the spread around the mean, are critical factors in understanding how people act. Thus they are critical for prediction, as well as for purposes of long term planning, where controlling risk may be equally important to controlling the mean (the point estimate).
There has been tremendous progress on large scale optimization techniques to enable the solution of large scale machine learning and data analytics problems. Stochastic Gradient Descent and its variants is probably the most-used large-scale optimization technique for learning. This has not yet seen an impact on the problem of statistical inference — namely, obtaining distributional information that might allow us to control the variance and hence the risk of certain solutions.
1. The document discusses numerical calculations of Darmon points, which are conjectural algebraic points on elliptic curves defined over real quadratic fields.
2. Darmon points are constructed using p-adic integration on the upper half-plane and cohomology classes associated to the elliptic curve.
3. Examples of calculating Darmon points are presented for various elliptic curves and real quadratic fields using Sage code implementing the necessary algorithms.
SOLVING BVPs OF SINGULARLY PERTURBED DISCRETE SYSTEMSTahia ZERIZER
In this article, we study boundary value problems of a large
class of non-linear discrete systems at two-time-scales. Algorithms are given to implement asymptotic solutions for any order of approximation.
Coordinate sampler: A non-reversible Gibbs-like samplerChristian Robert
This document describes a new MCMC method called the Coordinate Sampler. It is a non-reversible Gibbs-like sampler based on a piecewise deterministic Markov process (PDMP). The Coordinate Sampler generalizes the Bouncy Particle Sampler by making the bounce direction partly random and orthogonal to the gradient. It is proven that under certain conditions, the PDMP induced by the Coordinate Sampler has a unique invariant distribution of the target distribution multiplied by a uniform auxiliary variable distribution. The Coordinate Sampler is also shown to exhibit geometric ergodicity, an important convergence property, under additional regularity conditions on the target distribution.
The document presents an algorithm to find an optimal L(2,1)-labeling for triangular windmill graphs. It begins with definitions of triangular windmill graphs, L(2,1)-labelings, and the Chang-Kuo algorithm. The Chang-Kuo algorithm is then applied to obtain an L(2,1)-labeling of a triangular windmill graph W(3,n) by iteratively finding and labeling maximal 2-stable sets. The maximum label used is the labeling number λ(G).
This document discusses nonstationary covariance modeling. It begins by introducing concepts of covariance, correlation, and how correlation affects estimation and prediction. It then discusses properties of random fields like second-order stationarity and intrinsic stationarity. The difference between these two is explained. Common parametric models for isotropic covariance and variogram functions are presented, including spherical, exponential, Gaussian, rational quadratic, and Matérn models. Parameters and properties of these models like range, smoothness, and valid dimensionality are described. Examples of each model type are shown graphically.
This document summarizes research on simplifying calculations of scattering amplitudes, especially for tree-level amplitudes. It introduces the spinor-helicity formalism for writing compact expressions for amplitudes. It then discusses color decomposition in SU(N) gauge theory and the Yang-Mills Lagrangian. Specific techniques explored include BCFW recursion relations, an inductive proof of the Parke-Taylor formula, the 4-graviton amplitude and KLT relations, multi-leg shifts, and the MHV vertex expansion. The goal is to develop recursion techniques that vastly simplify calculations compared to traditional Feynman diagrams.
In this talk, we discuss some recent advances in probabilistic schemes for high-dimensional PIDEs. It is known that traditional PDE solvers, e.g., finite element, finite difference methods, do not scale well with the increase of dimension. The idea of probabilistic schemes is to link a wide class of nonlinear parabolic PIDEs to stochastic Levy processes based on nonlinear version of the Feynman-Kac theory. As such, the solution of the PIDE can be represented by a conditional expectation (i.e., a high-dimensional integral) with respect to a stochastic dynamical system driven by Levy processes. In other words, we can solve the PIDEs by performing high-dimensional numerical integration. A variety of quadrature methods could be applied, including MC, QMC, sparse grids, etc. The probabilistic schemes have been used in many application problems, e.g., particle transport in plasmas (e.g., Vlasov-Fokker-Planck equations), nonlinear filtering (e.g., Zakai equations), and option pricing, etc.
This document presents a Bayesian joint model for longitudinal and time-to-event outcomes that allows for subpopulation heterogeneity using latent variables. It begins with motivational examples in HIV and prostate cancer research. It then reviews existing joint modeling approaches before introducing a new latent process model that models the longitudinal and survival outcomes conditional on a shared latent process. The document describes prior distributions, a Gibbs sampling algorithm, and simulation studies to evaluate the model under both Gaussian and non-Gaussian longitudinal distributions.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
On the stability and accuracy of finite difference method for options pricingAlexander Decker
1) The document discusses finite difference methods for pricing options, specifically the implicit and Crank-Nicolson methods.
2) It analyzes the stability and accuracy of each method. The Crank-Nicolson method is found to be more accurate and converges faster than the implicit method.
3) Numerical examples are provided to demonstrate the convergence properties of each method compared to the true Black-Scholes value. The results show that the Crank-Nicolson method converges to the solution faster than the implicit method as the time and price steps approach zero.
A NEW APPROACH TO M(G)-GROUP SOFT UNION ACTION AND ITS APPLICATIONS TO M(G)-G...ijscmcj
In this paper, we define a new type of M(G)-group action , called M(G)-group soft union(SU) action and M(G)-ideal soft union(SU) action on a soft set. This new concept illustrates how a soft set effects on an M(G)-group in the mean of union and inclusion of sets and its function as bridge among soft set theory, set theory and M(G)-group theory. We also obtain some analog of classical M(G)- group theoretic concepts for M(G)-group SU-action. Finally, we give the application of SU-actions on M(G)-group to M(G)-group theory.
Polyadic systems and their representations are reviewed and a classification of general polyadic systems is presented. A new multiplace generalization of associativity preserving homomorphisms, a ’heteromorphism’ which connects polyadic systems having unequal arities, is introduced via an explicit formula, together with related definitions for multiplace representations and multiactions. Concrete examples of matrix representations for some ternary groups are then reviewed. Ternary algebras and Hopf algebras are defined, and their properties are studied. At the end some ternary generalizations of quantum groups and the Yang-Baxter equation are presented.
Time series analysis, modeling and applicationsSpringer
- The document proposes a novel composition forecasting model based on the Choquet integral with respect to a completed extensional L-measure and M-density fuzzy measure.
- It compares the performance of this model to other composition forecasting models, including ones based on extensional L-measure, L-measure, lambda-measure, and P-measure, as well as ridge regression and multiple linear regression models.
- Experimental results on grain production time series data show that the proposed Choquet integral composition forecasting model with completed extensional L-measure and M-density outperforms the other models.
Selection of bin width and bin number in histogramsKushal Kumar Dey
This document discusses various statistical methods for determining the optimal number of bins and bin width in histograms. It reviews methods such as Sturges' rule, Scott's rule, Bayesian optimal binning, and others. It presents the concepts behind each method and compares their performance in simulations. The simulations involved generating data from distributions like the chi-square, normal, and uniform, and comparing the methods based on how close their histogram estimates were to the true densities. Scott's rule and Freedman-Diaconis modification generally performed best at minimizing errors. The document also discusses generalizing some methods to multivariate histograms and potential areas for further research.
An Altering Distance Function in Fuzzy Metric Fixed Point Theoremsijtsrd
The aim of this paper is to improve conditions proposed in recently published fixed point results for complete and compact fuzzy metric spaces. For this purpose, the altering distance functions are used. Moreover, in some of the results presented the class of t-norms is extended by using the theory of countable extensions of t-norms. Dr C Vijender"An Altering Distance Function in Fuzzy Metric Fixed Point Theorems" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-1 | Issue-5 , August 2017, URL: http://www.ijtsrd.com/papers/ijtsrd2293.pdf http://www.ijtsrd.com/mathemetics/other/2293/an-altering-distance-function-in-fuzzy-metric-fixed-point-theorems/dr-c-vijender
Polyadic systems and their representations are reviewed and a classification of general polyadic systems is presented. A new multiplace generalization of associativity preserving homomorphisms, a 'heteromorphism' which connects polyadic systems having unequal arities, is introduced via an explicit formula, together with related definitions for multiplace representations and multiactions. Concrete examples of matrix representations for some ternary groups are then reviewed.
MSC classes: 16T05, 16T25, 17A42, 20N15, 20F29, 20G05, 20G42, 57T05
comments on exponential ergodicity of the bouncy particle samplerChristian Robert
The document summarizes recent work on establishing theoretical convergence rates for the bouncy particle sampler (BPS), a non-reversible Markov chain Monte Carlo algorithm. The main results show that under certain conditions on the target distribution, including having exponentially decaying tails, the BPS exhibits exponential ergodicity. A central limit theorem is also established. The analysis considers different cases for thin-tailed, thick-tailed, and transformed target distributions.
This document discusses various methods for approximating marginal likelihoods and Bayes factors, including:
1. Geyer's 1994 logistic regression approach for approximating marginal likelihoods using importance sampling.
2. Bridge sampling and its connection to Geyer's approach. Optimal bridge sampling requires knowledge of unknown normalizing constants.
3. Using mixtures of importance distributions and the target distribution as proposals to estimate marginal likelihoods through Rao-Blackwellization. This connects to bridge sampling estimates.
4. The document discusses various methods for approximating marginal likelihoods and comparing hypotheses using Bayes factors. It outlines the historical development and connections between different approximation techniques.
This document discusses weak isomorphism and isomorphism between fuzzy graphs. Some key points:
1. Weak isomorphism preserves the weights of nodes but not necessarily the weights of edges, while isomorphism preserves weights of both nodes and edges.
2. If two fuzzy graphs are isomorphic, they have the same order (number of nodes) and size (sum of edge weights).
3. Isomorphism between fuzzy graphs is proved to be an equivalence relation as it is reflexive, symmetric and transitive.
4. Weak isomorphism is proved to be a partial order relation as it satisfies reflexivity, anti-symmetry and transitivity.
5. Some properties of self-complement
Many areas of machine learning and data mining focus on point estimates of key parameters. In transportation, however, the inherent variance, and, critically, the need to understand the limits of that variance and the impact it may have, have long been understood to be important. Indeed, variance and other risk measures that capture the cost of the spread around the mean, are critical factors in understanding how people act. Thus they are critical for prediction, as well as for purposes of long term planning, where controlling risk may be equally important to controlling the mean (the point estimate).
There has been tremendous progress on large scale optimization techniques to enable the solution of large scale machine learning and data analytics problems. Stochastic Gradient Descent and its variants is probably the most-used large-scale optimization technique for learning. This has not yet seen an impact on the problem of statistical inference — namely, obtaining distributional information that might allow us to control the variance and hence the risk of certain solutions.
1. The document discusses numerical calculations of Darmon points, which are conjectural algebraic points on elliptic curves defined over real quadratic fields.
2. Darmon points are constructed using p-adic integration on the upper half-plane and cohomology classes associated to the elliptic curve.
3. Examples of calculating Darmon points are presented for various elliptic curves and real quadratic fields using Sage code implementing the necessary algorithms.
SOLVING BVPs OF SINGULARLY PERTURBED DISCRETE SYSTEMSTahia ZERIZER
In this article, we study boundary value problems of a large
class of non-linear discrete systems at two-time-scales. Algorithms are given to implement asymptotic solutions for any order of approximation.
Coordinate sampler: A non-reversible Gibbs-like samplerChristian Robert
This document describes a new MCMC method called the Coordinate Sampler. It is a non-reversible Gibbs-like sampler based on a piecewise deterministic Markov process (PDMP). The Coordinate Sampler generalizes the Bouncy Particle Sampler by making the bounce direction partly random and orthogonal to the gradient. It is proven that under certain conditions, the PDMP induced by the Coordinate Sampler has a unique invariant distribution of the target distribution multiplied by a uniform auxiliary variable distribution. The Coordinate Sampler is also shown to exhibit geometric ergodicity, an important convergence property, under additional regularity conditions on the target distribution.
The document presents an algorithm to find an optimal L(2,1)-labeling for triangular windmill graphs. It begins with definitions of triangular windmill graphs, L(2,1)-labelings, and the Chang-Kuo algorithm. The Chang-Kuo algorithm is then applied to obtain an L(2,1)-labeling of a triangular windmill graph W(3,n) by iteratively finding and labeling maximal 2-stable sets. The maximum label used is the labeling number λ(G).
This document discusses nonstationary covariance modeling. It begins by introducing concepts of covariance, correlation, and how correlation affects estimation and prediction. It then discusses properties of random fields like second-order stationarity and intrinsic stationarity. The difference between these two is explained. Common parametric models for isotropic covariance and variogram functions are presented, including spherical, exponential, Gaussian, rational quadratic, and Matérn models. Parameters and properties of these models like range, smoothness, and valid dimensionality are described. Examples of each model type are shown graphically.
This document summarizes research on simplifying calculations of scattering amplitudes, especially for tree-level amplitudes. It introduces the spinor-helicity formalism for writing compact expressions for amplitudes. It then discusses color decomposition in SU(N) gauge theory and the Yang-Mills Lagrangian. Specific techniques explored include BCFW recursion relations, an inductive proof of the Parke-Taylor formula, the 4-graviton amplitude and KLT relations, multi-leg shifts, and the MHV vertex expansion. The goal is to develop recursion techniques that vastly simplify calculations compared to traditional Feynman diagrams.
In this talk, we discuss some recent advances in probabilistic schemes for high-dimensional PIDEs. It is known that traditional PDE solvers, e.g., finite element, finite difference methods, do not scale well with the increase of dimension. The idea of probabilistic schemes is to link a wide class of nonlinear parabolic PIDEs to stochastic Levy processes based on nonlinear version of the Feynman-Kac theory. As such, the solution of the PIDE can be represented by a conditional expectation (i.e., a high-dimensional integral) with respect to a stochastic dynamical system driven by Levy processes. In other words, we can solve the PIDEs by performing high-dimensional numerical integration. A variety of quadrature methods could be applied, including MC, QMC, sparse grids, etc. The probabilistic schemes have been used in many application problems, e.g., particle transport in plasmas (e.g., Vlasov-Fokker-Planck equations), nonlinear filtering (e.g., Zakai equations), and option pricing, etc.
This document presents a Bayesian joint model for longitudinal and time-to-event outcomes that allows for subpopulation heterogeneity using latent variables. It begins with motivational examples in HIV and prostate cancer research. It then reviews existing joint modeling approaches before introducing a new latent process model that models the longitudinal and survival outcomes conditional on a shared latent process. The document describes prior distributions, a Gibbs sampling algorithm, and simulation studies to evaluate the model under both Gaussian and non-Gaussian longitudinal distributions.
call for papers, research paper publishing, where to publish research paper, journal publishing, how to publish research paper, Call For research paper, international journal, publishing a paper, IJEI, call for papers 2012,journal of science and technology, how to get a research paper published, publishing a paper, publishing of journal, research and review articles, engineering journal, International Journal of Engineering Inventions, hard copy of journal, hard copy of certificates, journal of engineering, online Submission, where to publish research paper, journal publishing, international journal, publishing a paper, hard copy journal, engineering journal
Sequential experimentation in clinical trialsSpringer
This chapter discusses time-sequential clinical trial designs where the primary endpoint is survival time. It begins with an overview of survival analysis methodology, which must be extended to account for the sequential nature of interim analyses in time-sequential trials. The seminal Beta-Blocker Heart Attack Trial (BHAT) is described as an example of an early time-sequential trial. Key developments following BHAT include methods that account for two time scales: the information accumulated over time and calendar time of interim analyses. Nelson-Aalen and Kaplan-Meier estimators are also summarized as tools for survival analysis in time-sequential settings.
An investigation of inference of the generalized extreme value distribution b...Alexander Decker
This document presents an investigation of parameter estimation for the generalized extreme value distribution based on record values. Maximum likelihood estimation is used to estimate the parameters β (scale parameter) and ξ (shape parameter). Likelihood equations are derived and solved numerically. Bootstrap and Markov chain Monte Carlo methods are proposed to construct confidence intervals for the parameters since intervals based on asymptotic normality may not perform well due to small sample sizes of records. Bayesian estimation of the parameters using MCMC is also investigated. An illustrative example involving simulated records is provided.
Redundancy in robot manipulators and multi robot systemsSpringer
This document summarizes how variational methods have been used to analyze three classes of snakelike robots: (1) hyper-redundant manipulators guided by backbone curves, (2) flexible steerable needles, and (3) concentric tube continuum robots. For hyper-redundant manipulators, variational methods are used to constrain degrees of freedom and provide redundancy resolution. For steerable needles, they generate optimal open-loop plans and model needle-tissue interactions. For concentric tube robots, they determine equilibrium conformations dictated by elastic mechanics principles. The document reviews these applications and illustrates how variational methods provide a natural analysis tool for various snakelike robots.
International Refereed Journal of Engineering and Science (IRJES)irjes
The core of the vision IRJES is to disseminate new knowledge and technology for the benefit of all, ranging from academic research and professional communities to industry professionals in a range of topics in computer science and engineering. It also provides a place for high-caliber researchers, practitioners and PhD students to present ongoing research and development in these areas.
Spillover Dynamics for Systemic Risk Measurement Using Spatial Financial Time...SYRTO Project
Spillover Dynamics for Systemic Risk Measurement Using Spatial Financial Time Series Models. Andre Lucas. Amsterdam - June, 25 2015. European Financial Management Association 2015 Annual Meetings.
1) The lecture discusses the time domain analysis of continuous time linear and time-invariant systems. It covers topics such as impulse response, convolution, and how the output of an LTI system can be determined from its impulse response and the input signal.
2) An example of analyzing the voltage response of an RC circuit to an arbitrary input is presented. The output is the sum of the zero-input response, due to initial conditions, and zero-state response, which is a convolution of the impulse response and input signal.
3) Detectors of high energy photons can be modeled as having an exponential decay impulse response. Examples of characterizing real detectors through measurements of energy resolution, timing resolution, and coincidence point spread
This document provides an overview of statistical inference concepts including:
1. Best unbiased estimators, which have minimum mean squared error for a given parameter. The best unbiased estimator, if it exists, must be a function of a sufficient statistic.
2. Sufficiency and the Rao-Blackwell theorem, which states that conditioning an estimator on a sufficient statistic produces a uniformly better estimator.
3. The Cramér-Rao lower bound, which provides a lower bound on the variance of unbiased estimators. Examples are given to illustrate key concepts like when the bound may not hold.
4. Examples are worked through to find minimum variance unbiased estimators, maximum likelihood estimators, and confidence intervals for various distributions
On estimating the integrated co volatility usingkkislas
This document proposes a method to estimate the integrated co-volatility of two asset prices using high-frequency data that contains both microstructure noise and jumps.
It considers two cases - when the jump processes of the two assets are independent, and when they are dependent. For the independent case, it proposes an estimator that is robust to jumps. For the dependent case, it proposes a threshold estimator that combines pre-averaging to remove noise with a threshold method to reduce the effect of jumps. It proves the estimators are consistent and establishes their central limit theorems. Simulation results are also presented to illustrate the performance of the proposed methods.
This document summarizes techniques for approximating marginal likelihoods and Bayes factors, which are important quantities in Bayesian inference. It discusses Geyer's 1994 logistic regression approach, links to bridge sampling, and how mixtures can be used as importance sampling proposals. Specifically, it shows how optimizing the logistic pseudo-likelihood relates to the bridge sampling optimal estimator. It also discusses non-parametric maximum likelihood estimation based on simulations.
This document summarizes a presentation on testing for volatility transmission between international markets using high frequency data. It discusses using realized volatility to estimate true latent volatility processes while controlling for jumps and microstructure noise. The presentation focuses on testing for transmission of only extreme or large volatility values between markets. A quantile model is used to define extreme periods, and cross-covariances are computed to test for non-causality between markets' extreme periods using Ljung-Box statistics. Simulations are performed based on a three-regime smooth-transition model to assess the test in finite samples.
The document provides an overview of sampling and pulse modulation. It discusses key topics such as:
1) The sampling theorem which states that a signal with bandwidth B can be reconstructed from samples taken at a rate greater than 2B.
2) How sampling a signal mathematically is achieved by multiplying it by an impulse train.
3) In both the time and frequency domains, sampling a signal is equivalent to multiplying/convolving the signal by an impulse train.
Tales on two commuting transformations or flowsVjekoslavKovac1
1) The document summarizes recent work on ergodic averages and flows for commuting transformations. It discusses convergence results for single and double linear ergodic averages in L2 and almost everywhere, as well as providing norm estimates to quantify the rate of convergence.
2) It also considers double polynomial ergodic averages and provides proofs for almost everywhere convergence in the continuous-time setting. Open problems remain for the discrete-time case.
3) An ergodic-martingale paraproduct is introduced, motivated by an open question from 1950. Convergence in Lp norm is shown, while almost everywhere convergence remains open.
This document summarizes statistical tests for comparing a sample distribution to a theoretical distribution. It discusses Kolmogorov-Smirnov tests (TN,1 and TN,2) that compare the empirical distribution function to the theoretical one. It also discusses Cramér–von Mises tests (TN,3 and TN,4) that integrate the squared differences between the empirical and theoretical distributions. The document proves that TN,1, TN,2, and TN,3 converge in distribution to integrals involving a Brownian bridge process. It also discusses that TN,1 and TN,2 are consistent under certain alternatives, while TN,3 and TN,4 are consistent under other alternatives. The document introduces a self-normalized test statistic
The document discusses the gauge integral, a generalization of the Riemann integral. It defines a gauge integral as a limit of Riemann sums taken over δ-fine partitions, where δ is a gauge function. It proves some basic properties of gauge integrals, including existence of δ-fine partitions, linearity, and a Cauchy criterion for integrability. The gauge integral includes the Riemann and Lebesgue integrals as special cases and has fewer technical assumptions than the Lebesgue integral.
Numerical simulation on laminar free convection flow and heat transfer over a...eSAT Journals
Abstract
In the present numerical study, laminar free-convection flow and heat transfer over an isothermal vertical plate is presented. By
means of similarity transformation, the original nonlinear coupled partial differential equations of flow are transformed to a pair
of simultaneous nonlinear ordinary differential equations. Then, they are reduced to first order system. Finally, NewtonRaphson
method and adaptive Runge-Kutta method are used for their integration. The computer codes are developed for this numerical
analysis in Matlab environment. Velocity and temperature profiles for various Prandtl number are illustrated graphically. Flow
and heat transfer parameters are derived as functions of Prandtl number alone. The results of the present simulation are then
compared with experimental data published in literature and find a good agreement.
Keywords: Free Convection, Heat Transfer, Matlab, Numerical Simulation, Vertical Plate.
This document discusses implicit differentiation and exponential growth and decay models. It contains:
1) An example of using implicit differentiation to find the derivative of a circle equation and the equation of the tangent line.
2) An explanation of how exponential growth and decay models take the form of y' = ky, leading to solutions of y = Ce^kt where k is the constant relative growth or decay rate.
3) An example modeling world population growth from 1950-2020 using an exponential growth model that estimates the 1993 population and predicts the 2020 population.
On Some Measures of Genetic Distance Based on Rates of Nucleotide SubstitutionJustine Leon Uro
The document presents a general DNA base-nucleotide substitution model and discusses three special cases: the three-substitution-type (3ST) model, two-substitution-type (2ST) model, and the Jukes-Cantor model. The 3ST model considers transitions and two types of transversions, while the 2ST model and Jukes-Cantor model further simplify the substitution rates. Differential equations are derived to model the change in base probabilities over time under each model.
Similar to Thegeneralizedinverse weibulldistribution .... (20)
Independent Study - College of Wooster Research (2023-2024) FDI, Culture, Glo...AntoniaOwensDetwiler
"Does Foreign Direct Investment Negatively Affect Preservation of Culture in the Global South? Case Studies in Thailand and Cambodia."
Do elements of globalization, such as Foreign Direct Investment (FDI), negatively affect the ability of countries in the Global South to preserve their culture? This research aims to answer this question by employing a cross-sectional comparative case study analysis utilizing methods of difference. Thailand and Cambodia are compared as they are in the same region and have a similar culture. The metric of difference between Thailand and Cambodia is their ability to preserve their culture. This ability is operationalized by their respective attitudes towards FDI; Thailand imposes stringent regulations and limitations on FDI while Cambodia does not hesitate to accept most FDI and imposes fewer limitations. The evidence from this study suggests that FDI from globally influential countries with high gross domestic products (GDPs) (e.g. China, U.S.) challenges the ability of countries with lower GDPs (e.g. Cambodia) to protect their culture. Furthermore, the ability, or lack thereof, of the receiving countries to protect their culture is amplified by the existence and implementation of restrictive FDI policies imposed by their governments.
My study abroad in Bali, Indonesia, inspired this research topic as I noticed how globalization is changing the culture of its people. I learned their language and way of life which helped me understand the beauty and importance of cultural preservation. I believe we could all benefit from learning new perspectives as they could help us ideate solutions to contemporary issues and empathize with others.
Abhay Bhutada, the Managing Director of Poonawalla Fincorp Limited, is an accomplished leader with over 15 years of experience in commercial and retail lending. A Qualified Chartered Accountant, he has been pivotal in leveraging technology to enhance financial services. Starting his career at Bank of India, he later founded TAB Capital Limited and co-founded Poonawalla Finance Private Limited, emphasizing digital lending. Under his leadership, Poonawalla Fincorp achieved a 'AAA' credit rating, integrating acquisitions and emphasizing corporate governance. Actively involved in industry forums and CSR initiatives, Abhay has been recognized with awards like "Young Entrepreneur of India 2017" and "40 under 40 Most Influential Leader for 2020-21." Personally, he values mindfulness, enjoys gardening, yoga, and sees every day as an opportunity for growth and improvement.
STREETONOMICS: Exploring the Uncharted Territories of Informal Markets throug...sameer shah
Delve into the world of STREETONOMICS, where a team of 7 enthusiasts embarks on a journey to understand unorganized markets. By engaging with a coffee street vendor and crafting questionnaires, this project uncovers valuable insights into consumer behavior and market dynamics in informal settings."
[4:55 p.m.] Bryan Oates
OJPs are becoming a critical resource for policy-makers and researchers who study the labour market. LMIC continues to work with Vicinity Jobs’ data on OJPs, which can be explored in our Canadian Job Trends Dashboard. Valuable insights have been gained through our analysis of OJP data, including LMIC research lead
Suzanne Spiteri’s recent report on improving the quality and accessibility of job postings to reduce employment barriers for neurodivergent people.
Decoding job postings: Improving accessibility for neurodivergent job seekers
Improving the quality and accessibility of job postings is one way to reduce employment barriers for neurodivergent people.
How Does CRISIL Evaluate Lenders in India for Credit RatingsShaheen Kumar
CRISIL evaluates lenders in India by analyzing financial performance, loan portfolio quality, risk management practices, capital adequacy, market position, and adherence to regulatory requirements. This comprehensive assessment ensures a thorough evaluation of creditworthiness and financial strength. Each criterion is meticulously examined to provide credible and reliable ratings.
Lecture slide titled Fraud Risk Mitigation, Webinar Lecture Delivered at the Society for West African Internal Audit Practitioners (SWAIAP) on Wednesday, November 8, 2023.
3. F. R. S. de Gusmão et al.
1 Introduction
The inverse Weibull (IW) distribution has received some attention in the literature.
Keller and Kamath (1982) study the shapes of the density and failure rate functions
for the basic inverse model. The random variable Y has an inverse Weibull distribution
if its cumulative distribution function (cdf) takes the form
G(t) = exp −
α
t
β
, t > 0, (1)
where α > 0 and β > 0. The corresponding probability density function (pdf) is
g(t) = βαβ
t−(β+1)
exp −
α
t
β
. (2)
The IW distribution is also a limiting distribution of the largest order statistics. Drapella
(1993) and Mudholkar and Kollia (1994) suggested the names complementary Weibull
and reciprocal Weibull for the distribution (2). The corresponding survival and hazard
functions are
SG(t) = 1 − G(t) = 1 − exp −
α
t
β
and
hG(t) = βαβ
t−(β+1)
exp −
α
t
β
1 − exp −
α
t
β −1
,
respectively. The kth moment about zero of (2) is
E(T k
) = αk
(1 − kβ−1
),
where (·) is the gamma function.
The rest of the paper is organized as follows. In Sect. 2, we define the general-
ized inverse Weibull (GIW) distribution. Section 3 provides a general formula for its
moments. General expansions for the moments of order statistics of the new distri-
bution are given in Sect. 4. We discuss in Sect. 5 maximum likelihood estimation
with censored data. Section 6 provides some properties of a mixture of two GIW
distributions. Sections 7 and 8 discuss the log-generalized inverse Weibull distribu-
tion and the corresponding regression model, respectively. Some diagnostic measures
for the log-generalized inverse Weibull model are proposed in Sect. 9. Section 10
defines two type of residuals. Section 11 illustrates two applications to real data
sets including the model fitting and influence analysis. Section 12 ends with some
conclusions.
123
4. The generalized inverse Weibull distribution
2 The generalized inverse Weibull distribution
Let G(t) be the cdf of the inverse Weibull distribution discussed by Drapella (1993),
Mudholkar and Kollia (1994) and Jiang et al. (1999), among others. The cdf of the
GIW distribution can be defined by elevating G(t) to the power of γ > 0, say F(t) =
G(t)γ = exp −γ α
t
β
. Hence, the GIW density function with three parameters
α > 0, β > 0 and γ > 0 is given by
f (t) = γβαβ
t−(β+1)
exp −γ
α
t
β
, t > 0. (3)
We can easily prove that (3) is a density function by substituting u = −γ αβt−β. The
inverse Weibull distribution is a special case of (3) when γ = 1. If T is a random
variable with density (3), we write T ∼ GIW(α, β, γ ). The corresponding survival
and hazard functions are
S(t) = 1 − F(t) = 1 − exp −γ
α
t
β
and
h(t) = γβαβ
t−(β+1)
exp −γ
α
t
β
1 − exp −γ
α
t
β −1
,
respectively.
We can simulate the GIW distribution using the nonlinear equation
t = α
− log(u)
γ
−1/β
, (4)
where u has the uniform U(0, 1) distribution. Plots of the GIW density and survival
function for selected parameter values are given in Figs. 1 and 2, respectively.
2.1 Characterizing the failure rate function
The first derivative h (t) = dh(t)/dt to study the hazard function shape is given by
h (t) = h(t)t−(β+1)
γβαβ
1 − exp −γ
α
t
β −1
− (β + 1)tβ
.
We can easily show that h(t) is unimodal with a maximum value at t∗, where t∗
satisfies the nonlinear equation
γ
α
t∗
1 − exp −γ
α
t∗
β −1
= 1 + β−1
.
123
5. F. R. S. de Gusmão et al.
Fig. 1 Plots of the GIW density for some values of α with β = 5 and γ = 0.5
Fig. 2 Plots of the GIW survival functions for some values of α with β = 5 and γ = 0.5
Evidently, lim
t→0
h(t) = lim
t→∞
h(t) = 0.
3 A general formula for the moments
We hardly need to emphasize the necessity and importance of the moments in any
statistical analysis especially in applied work. Some of the most important features
and characteristics of a distribution can be studied through moments (e.g., tendency,
dispersion, skewness and kurtosis). If the random variable T has the GIW density
function (3), the kth moment about zero is
123
6. The generalized inverse Weibull distribution
E(T k
) = γ
k
β αk
(1 − kβ−1
). (5)
The moment generating function M(z) of T for |z| < 1 is
M(z) = E(ezT
) =
n
k=0
zk
k!
γ
k
β αk
1 − kβ−1
.
Hence, for |z| < 1, the cumulant generating function of T is
K(z) = log[M(z)] = log
n
k=0
zk
k!
γ
k
β αk
1 − kβ−1
. (6)
The first four cumulants obtained from (6) are
κ1 = γ
1
β α 1(β), κ2 = γ
2
β α2
{ 2(β) − 1(β)2
},
κ3 = γ
3
β α3
{ 3(β) − 3 2(β) 1(β) + 2 1(β)3
}
and
κ4 = γ
4
β α4
{ 4(β) − 4 3(β) 1(β) − 3 1(β)2
+ 12 2(β) 1(β)2
− 6 1(β)4
},
where k(β) = (1 − kβ−1) for k = 1, . . . , 4. The skewness and kurtosis are then
ρ3 = κ3/κ
3
2
2 and ρ4 = κ4/κ2
2 , respectively.
The Shannon entropy of a random variable T with density f (t) is a measure of the
uncertainty defined by E{− log[ f (T )]}. The Shannon entropy for the GIWdistribution
can be expressed as
E{− log[ f (T )]} = − log(αβγ ) + (β + 1){log(α) + β−1
[log(γ ) + 0.577216]} + 1,
where 0.577216 is the approximate value for Euler’s constant.
4 Moments of order statistics
Order statistics make their appearance in many areas of statistical theory and practice.
Let the random variable Tr:n be the rth order statistic (T1:n ≤ T2:n ≤ · · · ≤ Tn:n) in a
sample of size n from the GIW distribution, for r = 1, . . . , n. The pdf of Tr:n is given
by
fr:n(t) = Cr:n f (t)F(t)r−1
[1 − F(t)]n−r
, t > 0,
where f (t) comes from (3), F(t) = exp −γ α
t
β
and Cr:n = n!/[(r −1)!(n −r)!].
123
7. F. R. S. de Gusmão et al.
The kth moment about zero of the rth order statistic is
µ(k)
r:n = E(T k
r:n) = Cr:n
∞
0
f (t)F(t)r−1
[1 − F(t)]n−r
dt.
After some algebra, we can obtain
µ(k)
r:n = Cr:nαk
γ k/β
[1 − (k/β)]
n−r
i=0
n − r
i
(−1)i
(r + i)
k
β −1
. (7)
An alternative formula for this moment (Barakat and Abdelkader 2004) is
E(T k
r:n) = k
n
j=n−i+1
(−1)j−n+i−1 j − 1
n − i
n
j
Ij (k), (8)
where Ij (k) denotes the integral
Ij (k) =
∞
0
tk−1
[1 − F(t)]j
dt. (9)
From Eqs. 8 and 9, we have
E(T k
r:n)=k(γ αβ
)
k
β
k(β)
n
j=n−i+1
j
l=0
(−1)j−n+i−1 j − 1
n − i
n
j
(−1)l j
l
l
k
β −1
.
(10)
For γ = 1, we obtain the moments of order statistics of the IW distribution. Equations
7 and 10 are the main results of this section.
5 Maximum likelihood estimation with censored data
Let Ti be a random variable distributed as (3) with the vector of parameters θ =
(α, β, γ )T . The data encountered in survival analysis and reliability studies are often
censored. A very simple random censoring mechanism that is often realistic is one
in which each individual i is assumed to have a lifetime Ti and a censoring time Ci ,
where Ti and Ci are independent random variables. Suppose that the data consist of
n independent observations ti = min(Ti , Ci ) for i = 1, . . . , n. The distribution of Ci
does not depend on any of the unknown parameters of Ti . Parametric inference for
such model is usually based on likelihood methods and their asymptotic theory. The
censored log-likelihood l(θ) for the model parameters reduces to
123
8. The generalized inverse Weibull distribution
l(θ) = r log(γ ) + log(β) + β log(α) − (β + 1)
i∈F
log(ti ) − γ αβ
i∈F
t
−β
i
+
i∈C
log 1 − exp −γ
α
ti
β
,
where r is the number of failures and F and C denote the sets of uncensored and
censored observations, respectively.
The score functions for the parameters α, β and γ are given by
Uα(θ) =
rβ
α
− γβαβ−1
i∈F
t
−β
i + γβαβ−1
i∈C
t
−β
i
1 − ui
ui
,
Uβ(θ) =
r
β
+ r log(α) −
i∈F
log(ti ) − γ αβ
i∈F
t
−β
i log
α
ti
+γ αβ
i∈C
t
−β
i log
α
ti
1 − ui
ui
and
Uγ (θ) =
r
γ
− αβ
i∈F
t
−β
i + αβ
i∈C
t
−β
i
1 − ui
ui
,
where ui = 1 − exp −γ α
ti
β
is the ith transformed observation.
The MLE θ of θ is obtained by solving the nonlinear likelihood equations Uα(θ) =
0, Uβ(θ) = 0 and Uγ (θ) = 0 using iterative numerical techniques such as the
Newton–Raphson algorithm. We employ the programming matrix language Ox
(Doornik 2007).
For interval estimation of (α, β, γ ) and hypothesis tests on these parameters, we
derive the observed information matrix since the expected information matrix is very
complicated and requires numerical integration. The 3×3 observed information matrix
J(θ) is
J(θ) = −
⎛
⎝
Lαα Lαβ Lαγ
· Lββ Lβγ
· · Lγ γ
⎞
⎠ ,
whose elements are given in Appendix 1. Under conditions that are fulfilled for param-
eters in the interior of the parameter space but not on the boundary, the asymptotic
distribution of
√
n(θ − θ) is N3(0, I (θ)−1
),
where I (θ) is the expected information matrix. This asymptotic behavior is valid if
I (θ) is replaced by J(θ), i.e., the observed information matrix evaluated at θ. The
multivariate normal N3(0, J(θ)−1) distribution can be used to construct approximate
confidence regions for the parameters and for the hazard and survival functions. The
123
9. F. R. S. de Gusmão et al.
asymptotic normality is also useful for testing goodness of fit of the GIW distribution
and for comparing this distribution with some of its special sub-models using one of
the three well-known asymptotically equivalent test statistics—namely, the likelihood
ratio (LR) statistic, Wald and Rao score statistics.
6 Mixture of two GIW distributions
Mixture distributions have been considered extensively by many authors. For an excel-
lent survey of estimation techniques, discussion and applications, see Everitt and Hand
(1981), Maclachlan and Krishnan (1997) and Maclachlan and Peel (2000). Recently,
AL-Hussaini and Sultan (2001), Jiang et al. (2001) and Sultan et al. (2007) reviewed
some properties and estimation techniques of finite mixtures for some lifetime models.
The mixture of two generalized inverse Weibull (MGIW) distributions has density
given by
f (t; θ) =
2
i=1
pi fi (t; θi ),
where 2
i=1 pi = 1, θ = (θT
1 , θT
2 )T , θ1 = (α1, β1, γ1)T , θ2 = (α2, β2, γ2)T and
fi (t; θi ) is the density function of the ith component given by
fi (t; θi ) = γi βi α
βi
i t−(βi +1)
exp −γi
αi
t
βi
, t, αi , βi , γi > 0, i = 1, 2.
Different shapes of the MGIW density are shown in Figs. 3 and 4. The cdf of the
MGIW distribution can be expressed as
F(t; θ) =
2
i=1
pi Fi (t; θi ),
where Fi (t; θi ) = exp −γi
αi
t
βi
is the cdf of the ith GIW distribution for i = 1, 2.
6.1 Properties
Here, we derive some properties of the MGIW distribution by extending the corre-
sponding results for the GIW distribution. The kth moment of the random variable T
following the MGIW distribution (5) is given by
E(T k
) =
2
i=1
pi γ
k
βi
i αk
i (1 − kβ−1
i ).
123
10. The generalized inverse Weibull distribution
Fig. 3 Plots of the MGIW density for some values of γ2 with α1 = 1, β1 = 2, γ1 = 3, α2 = 5 and β2 = 6
Fig. 4 Plots of the MGIW density for some values of γ1 and γ2 with α1 = 1, β1 = 2, α2 = 5 and β2 = 6
The corresponding survival and the hazard functions are
S(t) =
2
i=1
pi 1 − exp −γi
αi
t
βi
123
11. F. R. S. de Gusmão et al.
and
h(t) =
2
i=1 pi γi βi α
βi
i t−(βi +1) exp −γi
αi
t
βi
2
i=1 pi 1 − exp −γi
αi
t
βi
,
respectively. If γi = 1 and αi = 1/δi , the results derived by Sultan et al. (2007) are
obtained as a special case.
6.2 Identifiability
Let φ be a transform associated with each Fi ∈ having the domain of definition
Dφi with linear map M : Fi → φi . If there exists a total ordering ( ) of such that
(i) F1 F2 (F1, F2 ∈ ) ⇒ Dφ1 ⊆ Dφ2 and
(ii) for each F1 ∈ , there is some s1 ∈ Dφ1 , φ1(s) = 0 such that lims→s1 φ2(s)/
φ1(s) = 0 for F1 < F2 (F1, F2 ∈ ),
the class of all finite mixing distributions is identifiable relative to (Chandra
1977).
Using Chandra’s approach (assumptions (i) and (ii)) and the results by Sultan et al.
(2007), we now prove the following proposition.
Proposition The class of all finite mixing distributions relative to the GIW distribution
is identifiable.
Proof If Ti is a random variable with the GI W distribution, the sth moment is
φi (s) = E(T s
i ) = γ
s/βi
i α−s
i (1 − sβ−1
i ), i = 1, 2.
From the cdf of Ti we have
F1 < F2 when β1 = β2, γ1 = γ2 and α1 < α2 (11)
and
F1 < F2 when α1 = α2 >
1
t
, γ1 = γ2 and β1 < β2. (12)
Let Dφ1 (s) = (−∞, β1), Dφ2 (s) = (−∞, β2) and s = β1. From the facts (11) and
(12), we have
lim
s→β1
φ1(s) = γ
β1/β1
1 α
−β1
1 (1 − β1β−1
1 ) = (0+) = ∞. (13)
When α1 = α2 > 1/t,γ1 = γ2 and β1 < β2, we obtain
lim
s→β1
φ2(s) = γ
β1/β2
1 α
−β1
1 (1 − β1β−1
2 ) > 0. (14)
123
12. The generalized inverse Weibull distribution
Finally, the results (13) and (14) lead to
lim
s→β1
φ2(s)
φ1(s)
= 0,
and then the identifiability is proved.
It is a much more difficult problem to study the identifiability of a mixture of n ≥ 3
distributions. Lack of identifiability gives no guarantee of convergence to the true
values of parameters and therefore usually gives rise to confusing results.
7 The log-generalized inverse Weibull distribution
Let T be a random variable having the GIW density function (3). The random variable
Y = log(T ) has a log-generalized inverse Weibull (LGIW) distribution whose density
function, parameterized in terms of σ = 1/β and µ = log(α), is given by
f (y; γ, σ, µ)=
γ
σ
exp −
y − µ
σ
− γ exp −
y − µ
σ
, −∞ < y < ∞
(15)
where γ > 0, σ > 0 and −∞ < µ < ∞. Four density curves (15) for selected
parameter values are given in Fig. 5 to show great flexibility of the new parameter γ .
The corresponding survival function to (15) is
S(y) = 1 − exp −γ exp −
y − µ
σ
.
We now define the standardized random variable Z = (Y −µ)/σ with density function
π(z; γ ) = γ exp[−z − γ exp(−z)], −∞ < z < ∞ and γ > 0. (16)
The inverse extreme value distribution follows as a special case when γ = 1.
The kth ordinary moment of the standardized distribution (16), say µk = E(Zk),
is
µk =
∞
−∞
zk
γ exp −z − γ exp(−z) dz. (17)
Substituting u = exp(−z), we can rewrite Eq. 17 as
µk = (−1)k
γ
∞
0
log(u)
k
exp(−γ u)du. (18)
123
13. F. R. S. de Gusmão et al.
Fig. 5 Plots of the LGIW density for some parameter values
The integral in (18) follows from Prudnikov et al. (1986) and Nadarajah (2006) as
∞
0
[log(u)]k
exp(−γ u)du =
∂k[γ −a (a)]
∂ak
a=1
.
Inserting the last equation in (18), yields the kth moment of Z
µk = (−1)k
γ
∂k[γ −a (a)]
∂ak
a=1
. (19)
The ordinary moments of Y can be obtained via binomial expansion and Eq. 19.
8 The log-generalized inverse Weibull regression model
In many practical applications, the lifetimes ti are affected by explanatory variables
such as the cholesterol level, blood pressure and many others. Let xi = (xi1, . . . , xip)T
be the explanatory variable vector associated with the ith response variable yi for
i = 1, . . . , n. Consider a sample (y1, x1), . . . , (yn, xn) of n independent observations,
where each random response is defined by yi = min{log(ti ), log(ci )}. We assume
non-informative censoring and that the observed lifetimes and censoring times are
independent.
For the first time, we construct a linear regression model for the response variable
yi based on the LGIW density given by
yi = xT
i β + σzi , i = 1, . . . , n, (20)
123
14. The generalized inverse Weibull distribution
where the random error zi follows the distribution (16), β = (β1, . . . , βp)T , σ > 0
and γ > 0 are unknown scalar parameters and xi is the explanatory variable vector
modeling the location parameter µi = xT
i β. Hence, the location parameter vector
µ = (µ1, . . . , µn)T of the LGIW model has a linear structure µ = Xβ, where
X = (x1, . . . , xn)T is a known model matrix. The log-inverse Weibull (LIW) (or the
inverse extreme value) regression model is defined by Eq. 20 with γ = 1.
Let F and C be the sets of individuals for which yi is the log-lifetime or log-
censoring, respectively. The total log-likelihood function for the model parameters
θ = (γ, σ, βT
)T can be written from Eqs. 16 and 20 as
l(θ)=r log(γ )−log(σ) −
i∈F
yi − xT
i β
σ
−γ
i∈F
exp −
yi − xT
i β
σ
+
i∈C
log 1 − exp −γ exp −
yi − xT
i β
σ
, (21)
where r is the observed number of failures. The MLE θ of θ can be obtained by maxi-
mizing the log-likelihood function (21). We use the matrix programming language Ox
(MAXBFGS function) (see Doornik 2007) to compute this estimate. From the fitted
model (20), the survival function for yi can be estimated by
S(yi ; ˆγ , ˆσ, β
T
) = 1 − exp − ˆγ exp −
yi − xT
i β
ˆσ
.
Under general regularity conditions, the asymptotic distribution of
√
n(θ − θ)
is multivariate normal Np+2(0, K(θ)−1), where K(θ) is the expected information
matrix. The asymptotic covariance matrix K(θ)−1 of θ can be approximated by the
inverse of the (p+2)×(p+2) observed information matrix J(θ) and then the asymp-
totic inference for the parameter vector θ can be based on the normal approximation
Np+2(0, J(θ)−1) for θ. The observed information matrix is
J(θ) = {− ¨Lr,s} =
⎛
⎝
−Lγ γ −Lγ σ −Lγβj
· −Lσσ −Lσβj
· · −Lβj βs
⎞
⎠ ,
whose elements are given in Appendix 2.
The asymptotic multivariate normal Np+2(0, J(θ)−1) distribution can be used to
construct approximate confidence regions for some parameters in θ and for the hazard
and survival functions. In fact, an 100(1 − α)% asymptotic confidence interval for
each parameter θr is given by
AC Ir = ˆθr − zα/2 − ¨L
r,r
, ˆθr + zα/2 − ¨L
r,r
,
123
15. F. R. S. de Gusmão et al.
where − ¨L
r,r
denotes the rth diagonal element of the inverse of the estimated observed
information matrix J(θ)−1 and zα/2 is the quantile 1 − α/2 of the standard nor-
mal distribution. The asymptotic normality is also useful for testing goodness of
fit of some sub-models and for comparing some special sub-models using the LR
statistic.
The interpretation of the estimated coefficients could be based on the ratio of median
times (see Hosmer and Lemeshow 1999) which holds for continuous or categorical
explanatory variables. When the explanatory variable is binary (0 or 1), and consider-
ing the ratio of median times with x = 1 in the numerator, if ˆβ is negative (positive),
it implies that the individuals with x = 1 present reduced (increased) median survival
time in [exp( ˆβ) × 100%] as compared to those individuals in the group with x = 0,
assuming the other explanatory variables fixed.
We are also interested to investigate if the LIW model is a good model to fit the
data under investigation. Clearly, the LR statistic can be used to discriminate between
the LIW and LGIW models since they are nested models. The hypotheses to be tested
in this case are H0 : γ = 1 versus H1 : γ = 1 and the LR statistic reduces to
w = 2{l(θ) − l(θ)}, where θ is the MLE of θ under H0. The null hypothesis is
rejected if w > χ2
1−α(1), where χ2
1−α(1) is the quantile of the chi-square distribution
with one degree of freedom.
9 Sensitivity analysis
9.1 Global influence
A first tool to perform sensitivity analysis is by means of the global influence starting
from case-deletion. Case-deletion is a common approach to study the effect of drop-
ping the ith observation from the data set. The case-deletion model corresponding to
model (20) is given by
yl = xT
l β + σzl, l = 1, . . . , n, and for l = i. (22)
From now on, a quantity with subscript (i) refers to the original quantity with the ith
observation deleted. Let l(i)(θ) be the log-likelihood function for θ under model (22)
and θ(i) = ( ˆγ(i), ˆσ(i), β
T
(i))T be the MLE of θ obtained by maximizingl(i)(θ). To assess
the influence of the ith observation on the unrestricted estimate θ = ( ˆγ , ˆσ, β
T
)T , the
basic idea is to compare the difference between θ(i) and θ. If deletion of an observa-
tion seriously affect the estimates, more attention should be paid to that observation.
Hence, if θ(i) is far from θ, then the ith case is regarded as an influential observation.
A first measure of the global influence can be expressed as the standardized norm of
θ(i) − θ (so-called generalized Cook distance) given by
GDi (θ) = (θ(i) − θ)T
J(θ)−1
(θ(i) − θ).
Another alternative is to assess the values GDi (β) and GDi (γ, σ) which reveal the
impact of the ith observation on the estimates of β and (γ, σ), respectively. Another
123
16. The generalized inverse Weibull distribution
well-know measure of the difference between θ(i) and θ is the likelihood distance
L Di (θ) = 2{l(θ) − l(θ(i))}.
Further,wecanalsocompute ˆβj − ˆβj(i)( j = 1, . . . , p)toobtainthedifferencebetween
β and β(i). Alternative global influence measures are possible. We can develop the
behavior of a test statistic, such as the Wald test for explanatory variable or censoring
effect, under a case deletion scheme.
9.2 Local influence
As a second tool for sensitivity analysis, we now describe the local influence method
for the LGIW regression model with censored data. Local influence calculation can be
carried out for the model (20). If likelihood displacement L D(ω) = 2{l(θ) − l(θω)}
is used, where θω is the MLE under the perturbed model, the normal curvature for
θ at the direction d, d = 1, is given by Cd(θ) = 2|dT T
J(θ)−1 d| (see Cook
1986). Here, is a (p + 2) × n matrix depending on the perturbation scheme, whose
elements are given by ji = ∂2l(θ|ω)/∂θj ∂ωi , i = 1, . . . , n and j = 1, . . . , p + 2
evaluated at θ and ω0, and ω0 is the no perturbation vector. For the LGIW model, the
elements of J(θ) are given in Appendix 2. We can calculate normal curvatures Cd(θ),
Cd(γ ), Cd(σ) and Cd(β) to perform various index plots, for instance, the index plot
of dmax, the eigenvector corresponding to Cdmax , the largest eigenvalue of the matrix
B = T
J(θ)−1 and the index plots of Cdi (θ), Cdi (γ ), Cdi (σ) and Cdi (β), so-called
the total local influence (see, for example, Lesaffre and Verbeke 1998), where di is an
n×1 vector of zeros with one at the ith position. Hence, the curvature at direction di
has the form Ci = 2| T
i J(θ)−1
i |, where T
i is the ith row of . It is usual to point
out those cases such that
Ci ≥ 2 ¯C, ¯C =
1
n
n
i=1
Ci .
9.3 Curvature calculations
Here, under three perturbation schemes, we calculate the matrix
= ( ji )(p+2)×n =
∂2l(θ|ω)
∂θi ωj (p+2)×n
, j = 1, . . . , p + 2 and i = 1, . . . , n,
from the log-likelihood function (21). Let ω = (ω1, . . . , ωn)T be the vector of weights.
123
17. F. R. S. de Gusmão et al.
9.3.1 Case-weight perturbation
In this case, the log-likelihood function has the form
l(θ|ω) = log(γ ) − log(σ)
i∈F
ωi −
i∈F
ωi zi − γ
i∈F
ωi exp(−zi )
+
i∈C
ωi log {1 − exp[−γ exp(−zi )]} ,
where zi = (yi − xT
i β)/σ, 0 ≤ ωi ≤ 1 and ω = (1, . . . , 1)T . From now on, we take
= ( 1, . . . , p+2)T , ˆzi = (yi − xT
i
ˆβ)/ˆσ and ˆhi = exp[− ˆγ exp(−ˆzi )].
The elements of the vector 1 have the form
1i =
ˆγ −1 − exp(−ˆzi ) if i ∈ F
ˆhi exp(−ˆzi )(1 − ˆhi )−1 if i ∈ C.
On the other hand, the elements of the vector 2 are
2i =
ˆσ−1{−1 + ˆzi [−1 + ˆγ exp(−ˆzi )]} if i ∈ F
ˆγ ˆσ−1 ˆzi ˆhi exp(−ˆzi )(1 − ˆhi )−1 if i ∈ C.
The elements of the vector j for j = 3, . . . , p + 2 can be expressed as
ji =
ˆσ−1xi j 1 − ˆγ exp(ˆzi ) if i ∈ F
ˆγ ˆσ−1xi j ˆhi exp(−ˆzi )(1 − ˆhi )−1 if i ∈ C.
9.3.2 Response perturbation
We assume that each yi is perturbed as yiw = yi + ωi Sy, where Sy is a scale factor
such as the standard deviation of Y and ωi ∈ . The perturbed log-likelihood function
becomes
l(θ|ω) = r log(γ ) − log(σ) −
i∈F
z∗
i − γ
i∈F
exp(−z∗
i )
+
i∈C
log 1 − exp[−γ exp(−z∗
i )] ,
where z∗
i = (yi + ωi Sy) − xT
i β /σ. In addition, the elements of the vector 1 take
the form
1i =
ˆσ−1Sy exp(ˆzi ) if i ∈ F
ˆσ−1Sy ˆhi exp(−ˆzi ){[ ˆγ exp(−ˆzi ) − 1] 1 − ˆhi + ˆhi exp(−ˆzi )}(1 − ˆhi )−2 if i ∈ C.
123
18. The generalized inverse Weibull distribution
On the other hand, the elements of the vector 2 are
2i =
⎧
⎨
⎩
ˆσ−2Sy[1 + ˆγ ˆzi exp(−ˆzi ) − exp(−ˆzi )] if i ∈ F
ˆσ−2 ˆγ Sy ˆhi exp(−ˆzi ){ˆzi ˆγ exp(−ˆzi ) − 1 + 1}(1 − ˆhi )−1
+ˆσ−2 ˆγ Sy ˆzi ˆh2
i exp(−2ˆzi )(1 − ˆhi )−2 if i ∈ C.
The elements of the vector j , for j = 3, . . . , p + 2, can be expressed as
ji =
⎧
⎨
⎩
ˆγ ˆσ−2Sy xi j exp(−ˆzi ) if i ∈ F
− ˆγ ˆhi exp(−ˆzi ){−ˆσ−2Sy xi j [ ˆγ exp(−ˆzi ) − 1]}(1 − ˆhi )−1
+ ˆγ ˆσ−2Sy xi j ˆh2
i exp(−2ˆzi )(1 − ˆhi )−2 if i ∈ C.
9.3.3 Explanatory variable perturbation
We now consider an additive perturbation on a particular continuous explanatory vari-
able, say Xt , by making xitω = xit +ωi Sx , where Sx is a scale factor and ωi ∈ . This
perturbation scheme leads to the following expressions for the log-likelihood function
and for the elements of the matrix :
l(θ|ω) = r log(γ ) − log(σ) −
i∈F
z∗∗
i − γ
i∈F
exp(−z∗∗
i )
+
i∈C
log 1 − exp[−γ exp(−z∗∗
i )] ,
where z∗∗
i = (yi −x∗T
i β)/σ andx∗T
i β = β1+β2xi2+· · ·+βt (xit +ωi Sx )+· · ·+βpxip.
In addition, the elements of the vector 1 can be expressed as
1i =
−ˆσ−1Sx ˆβt exp(−ˆzi ) if i ∈ F
−ˆσ−1Sx ˆβt ˆhi exp(−ˆzi ){[ ˆγ exp(−ˆzi ) − 1](1 − ˆhi ) + ˆhi exp(−ˆzi )}(1 − ˆhi )−2 if i ∈ C,
the elements of the vector 2 are
2i =
⎧
⎨
⎩
ˆσ−2Sx ˆβt {−1 − ˆγ ˆzi exp(−ˆzi ) + exp(−ˆzi )} if i ∈ F
− ˆγ ˆσ−2Sx ˆβt ˆhi exp(−ˆzi ){ˆzi [ ˆγ exp(−ˆzi ) − 1] + 1}(1 − ˆhi )−1
− ˆγ ˆσ−2Sx ˆβt ˆzi ˆh2
i exp(−2ˆzi )(1 − ˆhi )−2 if i ∈ C,
the elements of the vector j , for j = 3, . . . , p + 2 and j = t, have the forms
ji =
⎧
⎨
⎩
− ˆγ ˆσ−2Sx ˆβt xi j exp(−ˆzi ) if i ∈ F
− ˆγ ˆσ−2Sx ˆβt xi j ˆhi exp(ˆzi )[γ exp(−ˆzi ) + 1](1 − ˆhi )−1
− ˆγ ˆσ−2Sx ˆβt xi j ˆh2
i exp(−2ˆzi )(1 − ˆhi )−2 if i ∈ C,
and the elements of the vector t are given by
123
19. F. R. S. de Gusmão et al.
ti =
⎧
⎪⎨
⎪⎩
ˆσ−1Sx 1 − ˆγ ˆσ−1 ˆβt xit exp(−ˆzi ) − exp(−ˆzi ) if i ∈ F
− ˆγ ˆσ−1Sx ˆhi exp(−ˆzi ){ˆσ−1 ˆβt xit [ ˆγ exp(−ˆzi ) − 1] − 1}(1 − ˆhi )−1
− ˆγ ˆσ−2Sx ˆβt xi j ˆh2
i exp(−2ˆzi )(1 − ˆhi )−2 if i ∈ C.
10 Residual analysis
In order to study departures from the error assumption as well as the presence of outli-
ers, we consider the martingale residual proposed by Barlow and Prentice (1988) and
some transformations of this residual.
10.1 Martingale residual
This residual was introduced in the counting process (see Fleming and Harrington
1991). For LGIW regression models can be written as
rMi =
1 + log 1 − exp − ˆγ exp(−ˆzi ) if i ∈ F
log 1 − exp − ˆγ exp(−ˆzi ) if i ∈ C,
where ˆzi = (yi − xT
i β)/ˆσ. The distribution of the residual rMi is skewed with a max-
imum value +1 and minimum value −∞. Transformations to achieve a more normal
shaped form would be more adequate for residual analysis.
10.2 Martingale-type residual
Another possibility is to use a transformation of the martingale residual, based on
the deviance residuals for the Cox model with no time-dependent covariates as intro-
duced by Therneau et al. (1990). We use this transformation of the martingale residual
in order to have a new residual symmetrically distributed around zero (see Ortega
et al. 2008). Hence, a martingale-type residual for the LGIW regression model can be
expressed as
rDi = sgn(rMi ) −2 rMi + δi log(δi − rMi )
1/2
, (23)
where rMi is the martingale residual.
11 Applications
11.1 Vitamin A
We illustrate the proposed model using a data set from a randomized community trial
that was designed to evaluate the effect of vitamin A supplementation on diarrheal
episodes in 1,207 pre-school children, aged 6–48 months at the baseline, who were
123
20. The generalized inverse Weibull distribution
0
0,2
0,4
0,6
0,8
1
0,4 0,80 0,2 0,6 1
i / n
TTT
Fig. 6 TTT-plot on vitamin A data
assigned to receive either placebo or vitamin A in a small city of the Northeast of
Brazil from December 1990 to December 1991 (see, for example, Barreto et al. 1994).
The vitamin A dosage was 100,000 IU for children younger than 12 months and
200,000 IU for older children, which is the highest dosage guideline established by
the World Health Organization (WHO) for the prevention of vitamin A deficiency.
The total time was defined as the time from the first dose of vitamin A until the
occurrence of an episode of diarrhea. An episode of diarrhea was defined as a sequence
of days with diarrhea and a day with diarrhea was defined when three or more liquid
or semi-liquid motions were reported in a 24-h period. The information on the occur-
rence of diarrhea collected at each visit corresponds to a recall period of 48–72h. The
number of liquid and semi-liquid motions per 24h was recorded. In many applica-
tions, there is a qualitative information about the failure rate function shape which
can help in selecting a particular model. In this context, a device called the total
time on test (TTT) plot (Aarset 1987) is useful. The TTT plot is obtained by plotting
G(r/n) = [( r
i=1 Ti:n) + (n − r)Tr:n]/( n
i=1 Ti:n) for r = 1, . . . , n against r/n
(Mudholkar et al. 1996), where Ti:n, for i = 1, . . . , n, are the order statistics of the
sample.
Figure 6 shows the TTT-plot for the current data with the convex and then concave
shape, which indicates that the hazard function has a unimodal shape. Hence, the GIW
distribution could be an appropriate model for fitting such data. The MLEs (approxi-
mate standard errors in parentheses) are: ˆα = 3.3422 (52.08), ˆβ = 0.9696 (0.0245)
and ˆγ = 8.6448 (130.66). In order to assess if the model is appropriate, Fig. 7a
plots the empirical survival function and the estimated survival function for the GIW
distribution. This distribution fits well to the data under analysis. Additionally, the
estimated hazard function in Fig. 7b has an upside-down bathtub-shaped curve.
11.2 Multiple myeloma
As a second application, we consider the data set given in Krall et al. (1975) and
reported in Lawless (2003, p. 334) connected with survival analysis. This data set has
123
21. F. R. S. de Gusmão et al.
(a) (b)
0
0,25
0,5
0,75
1
100 110 120 130
Time
EstimatedSurvivorFunction
Survival Fitted GIW
Kaplan-Meier estimate
Lower and Upper 95% C.L.
0
0,005
0,01
0,015
0,02
0,025
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
TiTime
FittedHazardRateFunction
Fig. 7 a Estimated survival function and the empirical survival for vitamin A data. b Estimated hazard rate
function for vitamin A data
recently been analyzed by Jin et al. (2003) and Ghosh and Ghosal (2006) who use a
semi-parametric accelerated failure time model. The aim of the study is to relate the
logarithm of the survival time (y) for multiple myeloma with a number of prognostic
variables for censored data. The data consist of the survival times, in months, of 65
patients who were treated with alkylating agents, of which 48 died during the study
and 17 survived. They also include measurements on each patient for the following
five predictors: logarithm of a blood urea nitrogen measurement at diagnosis (urea)
(x1); hemoglobin measurement at diagnosis (hemoglobin) (x2); age at diagnosis (age)
(x3); sex (sex) (x4, 0 for male and 1 for female) and serum calcium measurement at
diagnosis (serum) (x5).
We fit the LGIW no-intercept regression model
yi = β1xi1 + β2xi2 + β3xi3 + β4xi4 + β5xi5 + σzi , (24)
where the errors z1, . . . , z65 are independent random variables with density function
(16).
11.2.1 Maximum likelihood results
Table 1 provides the MLEs of the parameters for the LGIW and LIW regression models
fitted to the current data using the subroutine MAXBFGS in Ox. The LR statistic for
testing the hypotheses H0: γ = 1 versus H1: γ = 1, i.e. to compare the LIW and LGIW
regression models, is w = 2{−76.87 − (−82.69)} = 11.64 (P-value <0.01) which
yields favorable indications toward to the LGIW model. The LGIW model involves an
extra parameter which gives it more flexibility to fit the data, although some parame-
ters may be estimated with less precision, as is the case of β3 in this example. We note
from the fitted LGIW regression model that the age at diagnostic and the sex do not
seem to be significant. We expect that the survival time increases as the blood urea
nitrogen and the serum calcium measurements at diagnostic decrease, but it increases
123
22. The generalized inverse Weibull distribution
Table 1 MLEs of the parameters for the LGIW and LIW regression model fitted to the myeloma data
LGIW LIW
Parameter Estimate SE P-value Estimate SE P-value
γ 130.25 180.25 – 1 – –
σ 0.9179 0.1025 – 1.0525 0.1162 –
β1 –1.5459 0.3605 <0.0001 –1.1777 0.3885 0.0035
β2 0.1604 0.0456 0.0008 0.2421 0.0502 <0.0001
β3 0.0006 0.0119 0.9612 0.0309 0.0099 0.0027
β4 0.3093 0.2604 0.2393 0.8197 0.2659 0.0030
β5 – 0.1618 0.0607 0.0097 – 0.0605 0.0549 0.2745
Table 2 MLEs of the parameters for the log-Weibull model fitted to the myeloma data
Parameter Estimate SE P-value
σ 0.8822 0.0943 –
β0 4.5642 1.4489 0.0016
β1 −1.6257 0.5180 0.0017
β2 0.1181 0.0538 0.0282
β3 0.0187 0.0140 0.1815
β4 0.0352 0.2741 0.8979
β5 −0.1215 0.0883 0.1691
Table 3 Statistics AIC, BIC and CAIC for comparing the LGIW and LIW models
Model AIC BIC CAIC
LGIW 167.8 183.0 169.7
LIW 177.4 190.4 178.8
LW 174.6 189.8 176.6
as the hemoglobin measurement at diagnosis increases. As an alternative analysis, we
fit the log-Weibull (LW) regression model (see, for example, Lawless 2003) to these
data. The MLEs of the parameters for this model are given in Table 2. Further, we
provide in Table 3 the Akaike Information Criterion (AIC), the Bayesian Information
Criterion (BIC) and the Consistent Akaike Information Criterion (CAIC) to compare
the LGIW, LIW and log-Weibull regression models. The LGIW regression model out-
performs the other two models irrespective of the criteria and can be used effectively
in the analysis of these data.
123
23. F. R. S. de Gusmão et al.
(a) (b)
0
10
20
30
40
50
60
10 15 20 25 30 35 40 45 50 55 60 65
Index
GeneralizedCookDistance
5
0
2
4
6
10 15 20 25 30 35 40 45 50 55 60 65
Index
LikelihoodDistance
5
40
12
0 5 0 5
Fig. 8 a Index plot of GDi (θ) (Generalized Cook’s distance) on the myeloma data. b Index plot of L Di (θ)
(Likelihood distance) on the myeloma data
(a) (b)
-0,8
-0,7
-0,6
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
5 0
0,12
0,24
0,36
0,48
0,6
0,72
0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
5
Ci(θ)
12
Fig. 9 Index plot for θ on the myeloma data (case-weight perturbation): a dmax and b total local influence
11.2.2 Sensitivity analysis
Global influence analysis. In this section, we compute using Ox the case-deletion
measures GDi (θ) and L Di (θ) introduced in Sect. 9.1. The influence measure index
plots are given in Fig. 8, where we note that the cases 5, 12 and 40 are possible
influential observations.
Local and total influence analysis. In this section, we perform an analysis of the
local influence for the Golden shiner data set by fitting LGIW regression models.
Case-weight perturbation. We apply the local influence theory developed in
Sect. 9.2, where the case-weight perturbation is used and the value Cdmax = 1.27
was obtained as a maximum curvature. Figure 9a plots the eigenvector corresponding
to dmax, whereas Fig. 9b plots the total influence Ci . The observations 5 and 12 are
very distinguished in relation to the others.
123
24. The generalized inverse Weibull distribution
(a) (b)
-0,6
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
12
5
0
1
2
3
4
5
6
7
8
0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
5
Ci(θ)
12
Fig. 10 Index plot for θ on the myeloma data (response perturbation): a dmax and b total local influence
Response variable perturbation. The influence of perturbations on the observed
survival times is now analyzed. The value of the maximum curvature was Cdmax =
14.37. Figure 10a shows the plot of dmax versus the observation index, thus indicating
that the observations 5 and 12 are far way from the others. Figure 10b plots the total
local influence (Ci ) versus the observation index, thus suggesting that the observations
5 and 12 again stand out.
Explanatory variable perturbation. The perturbation of vectors for each continuous
explanatory variable (x1, x2, x3 and x5) are now investigated. The values for the max-
imum curvature were Cdmax = 2.05, Cdmax = 1.75, Cdmax = 1.05 and Cdmax = 1.75
for x1, x2, x3 and x5, respectively. Then plots of dmax and the total local influence Ci
against the index of the observations are shown in Figs. 11a–d and 12a–d. We note
that the observations 5 and 12 are very distinguished in relation to the others.
11.2.3 Residual analysis
In order to detect possible outliers in the fitted LGIW regression model, Fig. 13 plots
the martingale-type residualsrDmodi versus fitted values. We note that almost all obser-
vations are within the interval [−2, 2], although a few observations ( 5 and 12) appear
as possible outliers, thus indicating that the model is well fitted.
11.2.4 Impact of the detected influential observations
From the previous sections we can consider the observations 5 and 12 as possi-
ble influential. The observation 5 corresponds to one of the lowest survival times
and shows the extent of hemoglobin. The observation 12 is also one of the low-
est stroke survivors and also shows the extent of the serum calcium. Table 4 gives
the relative changes (in percentages) of the parameter estimates defined by RCθ j
=
[(θ j − θ j (I))/θ j ]×100, and the corresponding p-values, where θ j (I) denotes the
123
25. F. R. S. de Gusmão et al.
(a) (b)
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
5
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
10 15 20 25 30 35 40 45 50 55 60 65
Index
5
Ci(θ)
(c) (d)
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
1212
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0 5
0 5
0 5 0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
12
Fig. 11 Index plots for θ on the myeloma data (explanatory variable perturbation): a dmax (urea), b total
local influence (urea), c dmax (hemoglobin) and d total local influence (hemoglobin)
MLE of θ j after removing the “set I” of observations in the LGIW model fitted to the
current data. We consider the following sets: I1 = { 5}, I2 = { 12} and I3 = { 5, 12}.
Table 4 shows that the estimates of the LGIW regression model are not highly sensi-
tive under deletion of the outstanding observations. In general, the significance of the
parameter estimates does not change (at the significance level of 5%) after removing
the set I. Hence, we do not have inferential changes after removing the observations
handed out in the diagnostic plots. The large variations in the parameter estimates
occur for the estimates which are not significant, and then they could be removed
from the model.
12 Conclusions
We introduce a three parameter lifetime distribution, so-called the generalized inverse
Weibull (GIW) distribution, to extend some well-known distributions in the lifetime
123
26. The generalized inverse Weibull distribution
(a) (b)
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
10 15 20 25 30 35 40 45 50 55 60 65
Index
dmaxi(θ)
12
5
0
0,05
0,1
0,15
10 15 20 25 30 35 40 45 50 55 60 65
Index
5
Ci(θ)
12
(c) (d)
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
10 15 20 25 30 35 40 45 50 55 60 65
Index
12
dmaxi(θ)
5
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0 5
0 5
0 5 0 5 10 15 20 25 30 35 40 45 50 55 60 65
Index
12
dmaxi(θ)
5
Fig. 12 Index plot for θ on the myeloma data (explanatory variable perturbation): a dmax (age), b total
local influence (age), c dmax (serum) and d Total local influence (serum)
0,250 0,5
Fitted Value
Martingale-typeResidual(rDi)
5 12
3
2
1
0
-1
-2
-3
Fig. 13 Plot of the martingale-type residuals (rDi
) versus fitted values for the LGIW model fitted to
myeloma data
123
27. F. R. S. de Gusmão et al.
Table 4 Relative changes [-RC- in %], parameter estimates and their P-values in parentheses for the
indicated set
Dropping ˆγ ˆσ ˆβ1 ˆβ2 ˆβ3 ˆβ4 ˆβ5
– – – – – – –
All observations 130.3 0.918 −1.546 0.160 0.001 0.309 −0.162
(–) (–) (0.000) (0.000) (0.961) (0.239) (0.010)
[313] [3] [−4] [30] [808] [65] [−9]
Set I1 537.7 0.892 −1.603 0.112 −0.004 0.108 −0.177
(–) (–) (0.000) (0.030) (0.720) (0.695) (0.003)
[13] [2] [3] [−7] [−714] [−48] [−13]
Set I2 112.8 0.896 −1.504 0.172 0.005 0.459 −0.182
(–) (–) (0.000) (0.000) (0.694) (0.082) (0.003)
[−219] [5] [−1] [20] [117] [14] [−21]
Set I3 414.9 0.876 −1.555 0.128 0.000 0.265 −0.195
(–) (–) (0.000) (0.009) (0.993) (0.337) (0.001)
literature. The new distribution is much more flexible than the inverse Weibull distri-
bution and could have increasing, decreasing and unimodal hazard rates. We derive
explicit algebraic formula for the rth moment which holds in generality for any param-
eter values. We discuss maximum likelihood estimation and hypothesis tests for the
parameters.
We also propose a log-generalized inverse Weibull (LGIW) regression model for
analysis of lifetime censored data. We show how to obtain the maximum likelihood
estimates (MLEs) and develop asymptotic tests for the parameters based on the asymp-
totic distribution of the estimates. We provide applications of influence diagnostics
(global, local and total influence) in LGIW regression models to censored data. Further,
we discuss the robustness aspects of the MLEs from the fitted LGIW regression model
through residual and sensitivity analysis. The practical relevance and applicability of
the proposed regression model are demonstrated in two real data analysis.
Appendix 1: The observed information matrix for the GIW model
The elements of the observed information matrix J(θ) for the model parameters
(α, β, γ ) under censored data are given by
Lαα(θ) = −
rβ
α2
− γβ(β − 1)αβ−2
i∈F
t
−β
i
+ γβαβ−2
i∈C
t
−β
i
1 − ui
ui
β − 1 −
γβ
ui
α
ti
β
;
123
28. The generalized inverse Weibull distribution
Lαβ(θ) =
r
α
− γ αβ−1
i∈F
t
−β
i 1 + β log
α
ti
+ γ αβ−1
i∈C
t
−β
i
1 − ui
ui
1 + β log
α
ti
1 −
γ
ui
α
ti
β
;
Lγβ(θ)=−αβ
i∈F
t
−β
i log
α
ti
+αβ
i∈C
t
−β
i log
α
ti
1−ui
ui
1−
γ
ui
α
ti
β
;
Lαγ (θ) = −βαβ−1
i∈F
t
−β
i + βαβ−1
i∈C
t
−β
i
1 − ui
ui
1 −
γ
ui
α
ti
β
;
Lγ γ (θ) = −
r
γ 2
− α2β
i∈C
t
−2β
i
1 − ui
u2
i
;
and
Lββ(θ) = −
r
β2
− γ αβ
i∈F
log
α
ti
log(α) − t
−β
i log(ti )
+ γ αβ
i∈C
log
α
ti
1 − ui
ui
log(α) − t
−β
i log(ti )
−
t
−β
i γ
ui
α
ti
β
log
α
ti
.
Here, ui = 1 − exp −γ α
ti
β
.
Appendix 2: The observed information matrix for the LGIW model
We derive the second-order partial derivatives of the log-likelihood function. After
some algebraic manipulations, we obtain
Lγ γ = −
r
σ2
−
i∈C
exp(−2zi )hi (1 − hi )−2
;
Lγ σ =
i∈F
exp(−zi )(˙zi )σ +
i∈C
hi exp(−zi )(˙zi )σ [γ exp(−zi )](1 − hi )−1
+
i∈C
h2
i γ exp(−2zi )(˙zi )σ (1 − h−2
i );
123
29. F. R. S. de Gusmão et al.
Lγβj =
i∈F
exp(−zi )(˙zi )βj +
i∈C
hi exp(−zi )(˙zi )βj [γ exp(−zi )](1 − hi )−1
+
i∈C
h2
i γ exp(−2zi )(˙zi )βj (1 − h−2
i );
Lσσ =
r
σ2
+
i∈F
−(¨zi )σσ + γ exp(−zi ) − [(˙zi )σ ]2
+ (¨zi )σσ
−γ
i∈C
hi exp(−zi ) [(˙zi )σ ]2
[γ exp(−zi ) − 1] + (¨zi )σσ (1 − hi )−1
−γ
i∈C
h2
i exp(−2zi ) [(˙zi )σ ]2
(1 − hi )−2
;
Lσβj = −
i∈F
(¨zi )βj σ + γ
i∈F
− exp(−zi )(˙zi )βj (˙zi )σ + exp(−zi )(¨zi )βj σ
−γ
i∈C
hi exp(−zi ) (˙zi )βj (˙zi )σ [γ exp(−zi ) − 1] + (¨zi )βj σ (1 − hi )−1
+ γ
i∈C
h2
i exp(−2zi )(˙zi )β− j (˙zi )σ (1 − hi )−2
;
and
Lβj βs =
i∈F
−γ exp(−zi )(˙zi )βj (˙zi )βs
−γ
i∈C
hi exp(−zi ) (˙zi )βj (˙zi )βs [γ exp(−zi ) − 1] (1 − hi )−1
−γ
i∈C
h2
i exp(−2zi )(˙zi )βj (˙zi )βs (1 − hi )−2
;
where hi = exp[−γ exp(−zi )], (˙zi )σ = −zi /σ, (˙zi )βj = −xi j /σ, (˙zi )βs = −xis/σ,
(¨zi )σσ = 2zi /σ2, (¨zi )βj σ = xi j /σ2 and zi = (yi − xT
i β)/σ.
References
Aarset MV (1987) How to identify bathtub hazard rate. IEEE Trans Reliab 36:106–108
AL-Hussaini EK, Sultan KS (2001) Reliability and hazard based on finite mixture models. In: Balakrishnan
N, Rao CR (eds) Handbook of statistics, vol 20. Elsevier, Amsterdam pp 139–183
Barakat HM, Abdelkader YH (2004) Computing the moments of order statistics from nonidentical random
variables. Stat Methods Appl 13:15–26
Barlow WE, Prentice RL (1988) Residuals for relative risk regression. Biometrika 75:65–74
Barreto ML, Santos LMP, Assis AMO, Araújo MPN, Farenzena GG, Santos PAB, Fiaccone
RL (1994) Effect of vitamin A supplementation on diarrhoea and acute lower-respiratory-tract infec-
tions in young children in Brazil. Lancet 344:228–231
Chandra S (1977) On the mixtures of probability distributions. Scand J Stat 4:105–112
Cook RD (1986) Assesment of local influence (with discussion). J R Stat Soc B 48:133–169
Doornik J (2007) Ox: an object-oriented matrix programming language. International Thomson Bussiness
Press, London
123
30. The generalized inverse Weibull distribution
Drapella A (1993) Complementary Weibull distribution: unknown or just forgotten. Qual Reliab Eng Int
9:383–385
Everitt BS, Hand DJ (1981) Finite mixture distributions. Chapman and Hall, London
Fleming TR, Harrington DP (1991) Counting process and survival analysis. Wiley, New York
Ghosh SK, Ghosal S (2006) Semiparametric accelerated failure time models for censored data. In: Upad-
hyay SK, Singh U, Dey DK (eds) Bayesian statistics and its applications. Anamaya Publishers,
New Delhi pp 213–229
Hosmer DW, Lemeshow S (1999) Applied survival analysis. Wiley, New York
Jiang R, Zuo MJ, Li HX (1999) Weibull and Weibull inverse mixture models allowing negative weights.
Reliab Eng Syst Saf 66:227–234
Jiang R, Murthy DNP, Ji P (2001) Models involving two inverse Weibull distributions. Reliab Eng Syst Saf
73:73–81
Jin Z, Lin DY, Wei LJ, Ying Z (2003) Rank-based inference for the accelerated failure time model. Bio-
metrika 90:341–353
Keller AZ, Kamath AR (1982) Reliability analysis of CNC machine tools. Reliab Eng 3:449–473
Krall J, Uthoff V, Harley J (1975) A step-up procedure for selecting variables associated with survival.
Reliab Eng Syst Saf 73:73–81
Lawless JF (2003) Statistical models and methods for lifetime data. Wiley, New York
Lesaffre E, Verbeke G (1998) Local influence in linear mixed models. Biometrics 54:570–582
Maclachlan GJ, Krishnan T (1997) The EM algorithm and extensions. Wiley, New York
Maclachlan G, Peel D (2000) Finite mixture models. Wiley, New York
Mudholkar GS, Kollia GD (1994) Generalized Weibull family: a structural analysis. Commun Stat Ser A
23:1149–1171
Mudholkar GS, Srivastava DK, Kollia GD (1996) A generalization of the Weibull distribution with appli-
cation to the analysis of survival data. J Am Stat Assoc 91:1575–1583
Nadarajah S (2006) The exponentiated Gumbel distribution with climate application. Environmetrics
17:13–23
Ortega EMM, Paula GA, Bolfarine H (2008) Deviance residuals in generalized log-gamma regression mod-
els with censored observations. J Stat Comput Simul 78:747–764
Prudnikov AP, Brychkov YA, Marichev OI (1986) Integrals and series. Gordon and Breach Science Pub-
lishers, New York
Sultan KS, Ismail MA, Al-Moisheer AS (2007) Mixture of two inverse Weibull distributions: properties
and estimation. Comput Stat Data Anal 51:5377–5387
Therneau TM, Grambsch PM, Fleming TR (1990) Martingale-based residuals for survival models. Biomet-
rika 77:147–160
123
View publication statsView publication stats