Improved likelihood inference in generalized linear models (original) (raw)
Related papers
Improved Likelihood Ratio Statistics for Generalized Linear Models
Journal of the royal statistical society series b-methodological, 1983
This paper gives a general formula for the expected value of the likelihood ratio statistic for generalized linear models, corrected up to terms of order n-1, where n is the size of the sample. The application of this formula to some models is illustrated.
Specification tests for the response distribution in generalized linear models
Computational Statistics, 2012
Goodness-of-fit tests are proposed for the case of independent observations coming from the same family of distributions but with different parameters. The most popular related context is that of generalized linear models (GLMs) where the mean of the distribution varies with regressors. In the proposed procedures, and based on suitable estimators of the parameters involved, the data are transformed to normality. Then any test for normality for i.i.d. data may be applied. The method suggested is in full generality as it may be applied to arbitrary laws with continuous or discrete distribution functions, provided that an efficient method of estimation exists for the parameters. We investigate by Monte Carlo the relative performance of classical tests based on the empirical distribution function, in comparison to a corresponding test which instead of the empirical distribution function, utilizes the empirical characteristic function. Standard measures of goodness-of-fit often used in the context of GLM are also included in the comparison. The paper concludes with several real-data examples.
Small-sample testing inference in symmetric and log-symmetric linear regression models
Statistica Neerlandica
This paper deals with the issue of testing hypothesis in symmetric and log-symmetric linear regression models in small and moderate-sized samples. We focus on four tests, namely the Wald, likelihood ratio, score, and gradient tests. These tests rely on asymptotic results and are unreliable when the sample size is not large enough to guarantee a good agreement between the exact distribution of the test statistic and the corresponding chi-squared asymptotic distribution. Bartlett and Bartlett-type corrections typically attenuate the size distortion of the tests. These corrections are available in the literature for the likelihood ratio and score tests in symmetric linear regression models. Here, we derive a Bartlett-type correction for the gradient test. We show that the corrections are also valid for the log-symmetric linear regression models. We numerically compare the various tests, and bootstrapped tests, through simulations. Our results suggest that the corrected and bootstrapped tests exhibit type I probability error closer to the chosen nominal level with virtually no power loss. The analytically corrected tests, including the Bartlett-corrected gradient test derived in this paper, perform as well as the bootstrapped tests with the advantage of not requiring computationally-intensive calculations. We present two real data applications to illustrate the usefulness of the modified tests.
mdscore : An R Package to Compute Improved Score Tests in Generalized Linear Models
Journal of Statistical Software, 2014
Improved score tests are modifications of the score test such that the null distribution of the modified test statistic is better approximated by the chi-squared distribution. The literature includes theoretical and empirical evidence favoring the improved test over its unmodified version. However, the developed methodology seems to have been overlooked by data analysts in practice, possibly because of the difficulties associated with the computation of the modified test. In this article, we describe the mdscore package to compute improved score tests in generalized linear models, given a fitted model by the glm() function in R. The package is suitable for applied statistics and simulation experiments. Examples based on real and simulated data are discussed.
Three corrected score tests for generalized linear models with dispersion covariates
2003
We develop three corrected score tests for generalized linear models with dispersion covariates, thus generalizing the results of and . We present, in matrix notation, general formulae for the coefficients which define the corrected statistics. The formulae only require simple operations on matrices and can be used to obtain analytically closedform corrections for score test statistics in a variety of special generalized linear models with dispersion covariates. They also have advantages for numerical purposes since our formulae are readily computable using a language supporting numerical linear algebra. Two examples, namely, iid sampling without covariates on the mean or dispersion parameter oand one-way classification models, are given. We also present some simulations where the three corrected tests perform better than the usual score test, the likelihood ratio test and its Bartlett corrected version. Finally, we present a numerical example for a data set discussed by .
2004
Author Contact: Lauren Dong, Statistics Canada; e-mail: Lauren.Dong@statcan.can; FAX: (613) 951-3292 David Giles*, Dept. of Economics, University of Victoria, P.O. Box 1700, STN CSC, Victoria, B.C., Canada V8W 2Y2; e-mail: dgiles@uvic.ca; FAX: (250) 721-6214 * Corresponding co-author Abstract The empirical likelihood ratio (ELR) test for the problem of testing for normality in a linear regression model is derived in this paper. The sampling properties of the ELR test and four other commonly used tests are explored and analyzed using Monte Carlo simulation. The ELR test has good power properties against various alternative hypotheses.
Performance of tests of association in misspecified generalized linear models
Journal of Statistical Planning and Inference, 2006
We examine the effects of modelling errors, such as underfitting and overfitting, on the asymptotic power of tests of association between an explanatory variable x and an outcome in the setting of generalized linear models. The regression function for x is approximated by a polynomial or another simple function, and a chi-square statistic is used to test whether the coefficients of the approximation are simultaneously equal to zero. Adding terms to the approximation increases asymptotic power if and only if the fit of the model increases by a certain quantifiable amount. Although a high degree of freedom approximation offers robustness to the shape of the unknown regression function, a low degree of freedom approximation can yield much higher asymptotic power even when the approximation is very poor. In practice, it is useful to compute the power of competing test statistics across the range of alternatives that are plausible a priori. This approach is illustrated through an application in epidemiology.
Gradient test for generalised linear models with random effects
2016
This work develops the gradient test for parameter selection in generalised linear models with random effects. Asymptotically, the test statistic has a chi-squared distribution and the statistic has a compelling feature: it does not require computation of the Fisher information matrix. Performance of the test is verified through Monte Carlo simulations of size and power, and also compared to the likelihood ratio, Wald and Rao tests. The gradient test provides the best results overall when compared to the traditional tests, especially for smaller sample sizes.
Improved likelihood-based inference in Birnbaum–Saunders nonlinear regression models
Applied Mathematical Modelling, 2016
We address the issue of performing testing inference in Birnbaum-Saunders nonlinear regression models when the sample size is small. The likelihood ratio, Wald and score statistics provide the basis for testing inference on the parameters in this class of models. We focus on the small-sample case, where the reference chi-squared distribution gives a poor approximation to the true null distribution of these test statistics. We derive a general Bartlett-type correction in matrix notation for the score test, which reduces the size distortion of the test, and numerically compare the proposed test with the usual likelihood ratio, Wald and score tests, and with the Bartlett-corrected likelihood ratio test, and bootstrapcorrected tests. Our simulation results suggest that the proposed corrected test can be an interesting alternative to other tests since it leads to very accurate inference even for very small samples. We also present an empirical application for illustrative purposes.
Efficiency combined with simplicity: new testing procedures for Generalized Inverse Gaussian models
TEST, 2014
The standard efficient testing procedures in the Generalized Inverse Gaussian (GIG) family (also known as Halphen Type A family) are likelihood ratio tests, hence rely on Maximum Likelihood (ML) estimation of the three parameters of the GIG. The particular form of GIG densities, involving modified Bessel functions, prevents in general from a closed-form expression for ML estimators, which are obtained at the expense of complex numerical approximation methods. On the contrary, Method of Moments (MM) estimators allow for concise expressions, but tests based on these estimators suffer from a lack of efficiency compared to likelihood ratio tests. This is why, in recent years, trade-offs between ML and MM estimators have been proposed, resulting in simpler yet not completely efficient estimators and tests. In the present paper, we do not propose such a trade-off but rather an optimal combination of both methods, our tests inheriting efficiency from an ML-like construction and simplicity from the MM estimators of the nuisance parameters. This goal shall be reached by attacking the problem from a new angle, namely via the Le Cam methodology.