Saddlepoint approximations for the generalized variance and Wilks' statistic

A simple iterative procedure is found for the exact null and alternative distributions of likelihood ratio, cumulative sum and related statistics for testing for a change in probability of a sequence of independent binomial random variables. It is concluded that the likelihood ratio test is slightly less powerful than the cumulative sum test near the middle of the sequence, but that the likelihood ratio test is much more powerful near the ends. An example from epidemiology is used to illustrate the results. Under certain conditions the cumulative sum test is identical to a two-sample Kolmogorov-Smirnov test and so the iterative procedure can be used to find the null distribution of Kolmogorov-Smirnov and related statistics.

BackgroundIn many medical studies the likelihood ratio test (LRT) has been widely applied to examine whether the random effects variance component is zero within the mixed effects models framework; whereas little work about likelihood-ratio based variance component test has been done in the generalized linear mixed models (GLMM), where the response is discrete and the log-likelihood cannot be computed exactly. Before applying the LRT for variance component in GLMM, several difficulties need to be overcome, including the computation of the log-likelihood, the parameter estimation and the derivation of the null distribution for the LRT statistic.MethodsTo overcome these problems, in this paper we make use of the penalized quasi-likelihood algorithm and calculate the LRT statistic based on the resulting working response and the quasi-likelihood. The permutation procedure is used to obtain the null distribution of the LRT statistic. We evaluate the permutation-based LRT via simulations and compare it with the score-based variance component test and the tests based on the mixture of chi-square distributions. Finally we apply the permutation-based LRT to multilocus association analysis in the case–control study, where the problem can be investigated under the framework of logistic mixed effects model.ResultsThe simulations show that the permutation-based LRT can effectively control the type I error rate, while the score test is sometimes slightly conservative and the tests based on mixtures cannot maintain the type I error rate. Our studies also show that the permutation-based LRT has higher power than these existing tests and still maintains a reasonably high power even when the random effects do not follow a normal distribution. The application to GAW17 data also demonstrates that the proposed LRT has a higher probability to identify the association signals than the score test and the tests based on mixtures.ConclusionsIn the present paper the permutation-based LRT was developed for variance component in GLMM. The LRT outperforms existing tests and has a reasonably higher power under various scenarios; additionally, it is conceptually simple and easy to implement.

For testing separate families of hypotheses, the likelihood ratio test does not have the usual asymptotic properties. This paper considers the asymptotic distribution of the ratio of maximized likelihoods (RML) statistic in the special case of testing separate scale or location-scale families of distributions. We derive saddlepoint approximations to the density and tail probabilities of the log of the RML statistic. These approximations are based on the expansion of the log of the RML statistic up to the second order, which is shown not to depend on the location and scale parameters. The resulting approximations are applied in several cases, including normal versus Laplace, normal versus Cauchy, and Weibull versus log-normal. Our results show that the saddlepoint approximations are satisfactory, even for fairly small sample sizes, and are more accurate than normal approximations and Edgeworth approximations, especially for tail probabilities that are the values of main interest in hypothesis testing problems.

For the special case of balanced one-way random effects ANOVA, it has been established that the generalized likelihood ratio test (LRT) and Wald’s test are largely equivalent in testing the variance component. We extend these results to explore the relationships between Wald’s F test, and the LRT for a much broader class of linear mixed models; the generalized split-plot models. In particular, we explore when the two tests are equivalent and prove that when they are not equivalent, Wald’s F test is more powerful, thus making the LRT test inadmissible. We show that inadmissibility arises in realistic situations with common number of degrees of freedom. Further, we derive the statistical distribution of the LRT under both the null and alternative hypotheses $$H_0$$ and $$H_1$$ where $$H_0$$ is the hypothesis that the between variance component is zero. Providing an exact distribution of the test statistic for the LRT in these models will help in calculating a more accurate p-value than the traditionally used p-value derived from the large sample chi-square mixture approximations.

In a recent paper in this journal, the use of variance-stabilising transformation techniques was proposed to overcome the problem of inadequacy in normality approximation when testing association for a low-frequency variant in a case-control study. It was shown that tests based on the variance-stabilising transformations are more powerful than Fisher's exact test while controlling for type I error rate. Earlier in the journal, another study had shown that the likelihood ratio test (LRT) is superior to Fisher's exact test, Wald's test, and Pearson's χ(2) test in testing association for low-frequency variants. Thus, it is of interest to make a direct comparison between the LRT and the tests based on the variance-stabilising transformations. In this commentary, we show that the LRT and the variance-stabilising transformation-based tests have comparable power greater than Fisher's exact test, Wald's test, and Pearson's χ(2) test.

Approximate normality in testing an exchangeable covariance structure under large- and high-dimensional settings

Transform inversions, in which density and survival functions are computed from their associated moment generating function $\mathcal{M}$, have largely been based on methods which use values of $\mathcal{M}$ in its convergence region. Prominent among such methods are saddlepoint approximations and Fourier-series inversion methods, including the fast Fourier transform. In this paper we propose inversion methods which make use of values for $\mathcal{M}$ which lie outside of its convergence region and in its analytic continuation. We focus on the simplest and perhaps richest setting for applications in which $\mathcal{M}$ is either a meromorphic function in its analytic continuation, so that all of its singularities are poles, or else the singularities are isolated essential. Asymptotic expansions of finite- and infinite-orders are developed for density and survival functions using the poles of $\mathcal{M}$ in its analytic continuation. For finite-order expansions, the expansion error is a contour integral in the analytic continuation, which we approximate using the saddlepoint method based on following the path of steepest descent. Such saddlepoint error approximations accurately determine expansion errors and, thus, provide the means for determining the order of the expansion needed to achieve some preset accuracy. They also provide an additive correction term which increases accuracy of the expansion. Further accuracy is achieved by computing the expansion errors numerically using a contour path which ultimately tracks the steepest descent direction. Important applications include Wilks’ likelihood ratio test in MANOVA, compound distributions, and the Sparre Andersen and Cramér–Lundberg ruin models.

The small sample powers of two statistics, the likelihood ratio test, and a test based on the asymptotic normality of maximum likelihood estimators (z-test) were compared in a simulation experiment. Two models were specified, one containing the Box-Cox transformation on the dependent variable only, and one containing the Box Cox transformation on both the dependent and independent variables. The transformation parameter,λ was estimated 200 times, for each of six different values of z in each of three sample sizes foi both models. At each replication. 17 hypotheses were tested using both a likelihood ratio test and a z-test. Results indicate that w hiic both likelihood ratio tests and z-tests are unbiased, in small samples the z-test is generally preferable to the likelihood ratio test.

Geographically Weighted Regression (GWR) study potential relationships in regression models that distinguish geographic spaces using non-stationary parameters to overcome spatial effects. The use of gamma regression, namely regression with the dependent variable with a gamma distribution, can be an alternative if the data do not follow a normal distribution. Gamma distribution is a continuous set of non-negative values, generally skewed to the right or positive skewness. Gamma regression is developed to be Bivariate Gamma Regression (BGR) when there are two dependent variables with gamma distribution. If the observation units are location points, spatial effects may occur. The Geographically Weighted Bivariate Gamma Regression (GWBGR) model can be a solution for spatial heterogeneity. However, during its development, many cases require information from panel data. Using panel data can provide complete information because it covers several periods, but it allows for temporal effects. This study developed a Geographically and Temporally Weighted Bivariate Gamma Regression (GTWBGR) model to handle spatial and temporal heterogeneity simultaneously. The estimation of the GTWBGR model parameters uses the Maximum Likelihood Estimation (MLE) method that followed by the numerical iteration of Berndt Hall Hall Hausman (BHHH). The simultaneous testing uses the Maximum Likelihood Ratio Test (MLRT) method to get a test statistic. With a large sample size, the distribution of the test statistic approaches chi-square. Meanwhile, partial testing uses the Z test statistic.

In this paper, the distribution and percentage points of the likelihood ratio statistic for testing the hypothesis H : μ = μ e; Σ = Σ c , where e(p × p) = (1, …, 1)’ and Σ c is a circular symmetry matrix, in a p-variate normal distribution are obtained. The distribution is derived in series form using the inverse Mellin transform and the residue theorem. Percentage points for p = 4, 5, 6 and 7 are computed using distributional results given in this article.

1. Probability Models. 1.1 Discrete Probability Models. 1.2 Conditional Probability and Independence. 1.3 Random Variables. 1.4 Expectation. 1.5 The Variance. 1.6 Covariance and Correlation. 2. Special Discrete Distributions. 2.1 The Binomial Distribution. 2.2 The Hypergeometric Distribution. 2.3 The Geometric and Negative Binomial Distributions. 2.4 The Poisson Distribution. 3. Continuous Random Variables. 4.1 Continuous RV's and Their Distributions. 4.2 Expected Values and Variances. 4.3 Transformations of Random Variables. 4.4Joint Densities. 4 Special Continuous Distributions. 4.1 The Normal Distribution. 4.2 The Gamma Distribution. 5. Conditional Distributions. 5.1 The Discrete Case. 5.2 Conditional Expectations for the Discrete Case. 5.3 Conditional Densities and Expectations for Continuous RV's. 6. Limit Laws. 6.1 Moment Generating Functions. 6.2 Convergence in Probability and in Distribution. 6.3 The Central Limit Theorem. 6.4 The Delta-Method. 7. Estimation. 7.1 Point Estimation. 7.2 The Method of Moments. 7.3 Maximum Likelihood. 7.4 Consistency. 7.5 The OMEGA-Method. 7.6 Confidence Intervals. 7.7 Fisher Information, The Cramer-Rao Bound, and Asymptotic Normality of MLE's. 7.8 Sufficiency. 8. Testing Hypotheses. 8.1 Introduction. 8.2 The Neyman-Pearson Lemma. 8.3 The Likelihood Ratio Test. 8.4 The p-Value and the Relationship Between Tests of Hypotheses and Confidence Intervals. 9. The Multivariate Normal, Chi-square, t, and F-Distributions. 9.1 The Multivariate Normal Distribution. 9.2 The Central and Noncentral Chi-Square Distributions. 9.3 Student's t-Distribution. 9.4 The F-Distribution. 10.3 Nonparametric Statistics. 10.1 The Wilcoxon Test and Estimator. 10.2 One Sample Methods. 10.3 The Kolmogorov-Smirnov Tests. 11. Linear Models. 11.1 The Principle of Least Squares. 11.2 Linear Models. 11.3 F-Tests for H0. 11.4 Two-Way Analysis of Variance. 12. Frequency Data. 12.1 Logistic Regression. 12.2 Two-Way Frequency Tables. 12.3 Chi-Square Goodness of Fit Tests. 13. Miscellaneous Topics. 13.1 Survival Analysis. 13.2 Bootstrapping. 13.3 Bayesian Statistics. 13.4 Sampling.

A random variable Y 1 is said to be smaller than Y 2 in the increasing concave stochastic order if for all increasing concave functions for which the expected values exist, and smaller than Y 2 in the increasing convex order if for all increasing convex ψ. This article develops nonparametric estimators for the conditional cumulative distribution functions of a response variable Y given a covariate X, solely under the assumption that the conditional distributions are increasing in x in the increasing concave or increasing convex order. Uniform consistency and rates of convergence are established both for the K-sample case and for continuously distributed X.

In the Gaussian sequence model Y=μ+ξ, we study the likelihood ratio test (LRT) for testing H0:μ=μ0 versus H1:μ∈K, where μ0∈K, and K is a closed convex set in Rn. In particular, we show that under the null hypothesis, normal approximation holds for the log-likelihood ratio statistic for a general pair (μ0,K), in the high-dimensional regime where the estimation error of the associated least squares estimator diverges in an appropriate sense. The normal approximation further leads to a precise characterization of the power behavior of the LRT in the high-dimensional regime. These characterizations show that the power behavior of the LRT is in general nonuniform with respect to the Euclidean metric, and illustrate the conservative nature of existing minimax optimality and suboptimality results for the LRT. A variety of examples, including testing in the orthant/circular cone, isotonic regression, Lasso and testing parametric assumptions versus shape-constrained alternatives, are worked out to demonstrate the versatility of the developed theory.

When using maximum likelihood methods to estimate genetic and environmental components of (co)variance, it is common to test hypotheses using likelihood ratio tests, since such tests have desirable asymptotic properties. In particular, the standard likelihood ratio test statistic is assumed asymptotically to follow a χ2distribution with degrees of freedom equal to the number of parameters tested. Using the relationship between least squares and maximum likelihood estimators for balanced designs, it is shown why the asymptotic distribution of the likelihood ratio test for variance components does not follow a χ2distribution with degrees of freedom equal to the number of parameters tested when the null hypothesis is true. Instead, the distribution of the likelihood ratio test is a mixture of χ2distributions with different degrees of freedom. Implications for testing variance components in twin designs and for quantitative trait loci mapping are discussed. The appropriate distribution of the likelihood ratio test statistic should be used in hypothesis testing and model selection.

The authors address likelihood ratio statistics used to test simultaneously conditions on mean vectors and patterns on covariance matrices. Tests for conditions on mean vectors, assuming or not a given structure for the covariance matrix, are quite common, since they may be easily implemented. But, on the other hand, the practical use of simultaneous tests for conditions on the mean vectors and a given pattern for the covariance matrix is usually hindered by the nonmanageability of the expressions for their exact distribution functions. The authors show the importance of being able to adequately factorize the c.f. of the logarithm of likelihood ratio statistics in order to obtain sharp and highly manageable near-exact distributions, or even the exact distribution in a highly manageable form. The tests considered are the simultaneous tests of equality or nullity of means and circularity, compound symmetry, or sphericity of the covariance matrix. Numerical studies show the high accuracy of the near-exact distributions and their adequacy for cases with very small samples and/or large number of variables. The exact and near-exact quantiles computed show how the common chi-square asymptotic approximation is highly inadequate for situations with small samples or large number of variables.