Biometrika - Advance Access

Bias reduction in exponential family nonlinear models

Kosmidis, I., Firth, D. — Mon, 09 Nov 2009 19:52:52 PST

In Firth (1993, Biometrika) it was shown how the leading term in the asymptotic bias of the maximum likelihood estimator is removed by adjusting the score vector, and that in canonical-link generalized linear models the method is equivalent to maximizing a penalized likelihood that is easily implemented via iterative adjustment of the data. Here a more general family of bias-reducing adjustments is developed for a broad class of univariate and multivariate generalized nonlinear models. The resulting formulae for the adjusted score vector are computationally convenient, and in univariate models they directly suggest implementation through an iterative scheme of data adjustment. For generalized linear models a necessary and sufficient condition is given for the existence of a penalized likelihood interpretation of the method. An illustrative application to the Goodman row-column association model shows how the computational simplicity and statistical benefits of bias reduction extend beyond generalized linear models.

Maximum likelihood estimation using composite likelihoods for closed exponential families

Mardia, K. V., Kent, J. T., Hughes, G., Taylor, C. C. — Thu, 29 Oct 2009 22:34:34 PDT

In certain multivariate problems the full probability density has an awkward normalizing constant, but the conditional and/or marginal distributions may be much more tractable. In this paper we investigate the use of composite likelihoods instead of the full likelihood. For closed exponential families, both are shown to be maximized by the same parameter values for any number of observations. Examples include log-linear models and multivariate normal models. In other cases the parameter estimate obtained by maximizing a composite likelihood can be viewed as an approximation to the full maximum likelihood estimate. An application is given to an example in directional data based on a bivariate von Mises distribution.

A note on a conjectured sharpness principle for probabilistic forecasting with calibration

Pal, S. — Wed, 28 Oct 2009 21:59:46 PDT

This note proves a weak sharpness principle as conjectured by Gneiting et al. (2007) in connection with probabilistic forecasting subject to calibration constraints.

Construction of orthogonal Latin hypercube designs

Sun, F., Liu, M.-Q., Lin, D. K. J. — Mon, 26 Oct 2009 20:42:25 PDT

Latin hypercube designs have found wide application. Such designs guarantee uniform samples for the marginal distribution of each input variable. We propose a method for constructing orthogonal Latin hypercube designs in which all the linear terms are orthogonal not only to each other, but also to the quadratic terms. This construction method is convenient and flexible, and the resulting designs can accommodate many more factors than can existing ones.

Sinh-arcsinh distributions

Jones, M. C., Pewsey, A. — Sun, 25 Oct 2009 19:47:56 PDT

We introduce the sinh-arcsinh transformation and hence, by applying it to a generating distribution with no parameters other than location and scale, usually the normal, a new family of sinh-arcsinh distributions. This four-parameter family has symmetric and skewed members and allows for tailweights that are both heavier and lighter than those of the generating distribution. The central place of the normal distribution in this family affords likelihood ratio tests of normality that are superior to the state-of-the-art in normality testing because of the range of alternatives against which they are very powerful. Likelihood ratio tests of symmetry are also available and are very successful. Three-parameter symmetric and asymmetric subfamilies of the full family are also of interest. Heavy-tailed symmetric sinh-arcsinh distributions behave like Johnson S_U distributions, while their light-tailed counterparts behave like sinh-normal distributions, the sinh-arcsinh family allowing a seamless transition between the two, via the normal, controlled by a single parameter. The sinh-arcsinh family is very tractable and many properties are explored. Likelihood inference is pursued, including an attractive reparameterization. Illustrative examples are given. A multivariate version is considered. Options and extensions are discussed.

Adaptive approximate Bayesian computation

Beaumont, M. A., Cornuet, J.-M., Marin, J.-M., Robert, C. P. — Mon, 12 Oct 2009 22:44:16 PDT

Sequential techniques can enhance the efficiency of the approximate Bayesian computation algorithm, as in Sisson et al.’s (2007) partial rejection control version. While this method is based upon the theoretical works of Del Moral et al. (2006), the application to approximate Bayesian computation results in a bias in the approximation to the posterior. An alternative version based on genuine importance sampling arguments bypasses this difficulty, in connection with the population Monte Carlo method of Cappé et al. (2004), and it includes an automatic scaling of the forward kernel. When applied to a population genetics example, it compares favourably with two other versions of the approximate algorithm.

A note on adaptive Bonferroni and Holm procedures under dependence

Guo, W. — Mon, 12 Oct 2009 22:44:15 PDT

Hochberg & Benjamini (1990) first presented adaptive procedures for controlling familywise error rate. However, until now, it has not been proved that these procedures control the familywise error rate. We introduce a simplified version of Hochberg & Benjamini’s adaptive Bonferroni and Holm procedures. Assuming a conditional dependence model, we prove that the former procedure controls the familywise error rate in finite samples while the latter controls it approximately.

Semiparametric methods for evaluating risk prediction markers in case-control studies

Huang, Y., Pepe, M. S. — Mon, 12 Oct 2009 22:44:14 PDT

The performance of a well-calibrated risk model for a binary disease outcome can be characterized by the population distribution of risk and displayed with the predictiveness curve. Better performance is characterized by a wider distribution of risk, since this corresponds to better risk stratification in the sense that more subjects are identified at low and high risk for the disease outcome. Although methods have been developed to estimate predictiveness curves from cohort studies, most studies to evaluate novel risk prediction markers employ case-control designs. Here we develop semiparametric methods that accommodate case-control data. The semiparametric methods are flexible, and naturally generalize methods previously developed for cohort data. Applications to prostate cancer risk prediction markers illustrate the methods.

Bayesian analysis of matrix normal graphical models

Wang, H., West, M. — Fri, 09 Oct 2009 06:36:40 PDT

We present Bayesian analyses of matrix-variate normal data with conditional independencies induced by graphical model structuring of the characterizing covariance matrix parameters. This framework of matrix normal graphical models includes prior specifications, posterior computation using Markov chain Monte Carlo methods, evaluation of graphical model uncertainty and model structure search. Extensions to matrix-variate time series embed matrix normal graphs in dynamic models. Examples highlight questions of graphical model uncertainty, search and comparison in matrix data contexts. These models may be applied in a number of areas of multivariate analysis, time series and also spatial modelling.

Generalized fiducial inference for wavelet regression

Hannig, J., Lee, T. C. M. — Sat, 03 Oct 2009 22:18:03 PDT

We apply Fisher’s fiducial idea to wavelet regression, first developing a general methodology for handling model selection problems within the fiducial framework. We propose fiducial-based methods for wavelet curve estimation and the construction of pointwise confidence intervals. We show that these confidence intervals have asymptotically correct coverage. Simulations demonstrate that they possess promising empirical properties.

Sliced space-filling designs

Qian, P. Z. G., Wu, C. F. J. — Sat, 03 Oct 2009 22:18:03 PDT

We propose an approach to constructing a new type of design, a sliced space-filling design, intended for computer experiments with qualitative and quantitative factors. The approach starts with constructing a Latin hypercube design based on a special orthogonal array for the quantitative factors and then partitions the design into groups corresponding to different level combinations of the qualitative factors. The points in each group have good space-filling properties. Some illustrative examples are given.

A note on the variance of doubly-robust G-estimators

Moodie, E. E. M. — Thu, 01 Oct 2009 21:39:40 PDT

A recursive variance calculation is derived for doubly-robust G-estimators for dynamic treatment regimes in a multi-interval setting. Treatment decision parameters are not assumed to be shared across treatment intervals; this independence of parameters permits sequential estimation of the G-estimators’ variance when G-estimation is performed in a sequential fashion. The recursive variance calculation is both natural and computationally feasible. This development can easily be adapted to other complex estimating procedures that require nuisance parameter estimation.

Some design properties of a rejective sampling procedure

Fuller, W. A. — Thu, 01 Oct 2009 21:39:39 PDT

Occasionally, a selected probability sample may appear undesirable with respect to the available auxiliary information. In such a situation, the practitioner might consider rejecting the sample and selecting a new set of sample elements. We consider a procedure in which the probability sample is rejected unless the sample mean of an auxiliary vector is within a specified distance of the population mean. It is proven that the large sample mean and variance of the regression estimator for the rejective sample are the same as those of the regression estimator for the original selection procedure. Likewise, the usual estimator of variance for the regression estimator is appropriate for the rejective sample. In a Monte Carlo experiment, the large sample properties hold for relatively small samples and the Monte Carlo results are in agreement with the theoretical orders of approximation. The efficiency effect of the described rejective sampling is o(n_N^–1, where n_N is the expected sample size, but the effect can be important for particular samples. For example, rejective sampling can be used to eliminate those samples that give negative weights for the regression estimator.

Nonparametric estimation of the probability of illness in the illness-death model under cross-sectional sampling

Mandel, M., Fluss, R. — Wed, 30 Sep 2009 23:50:52 PDT

Cross-sectional sampling is an attractive design that saves resources but results in biased data. For proper inference, one should first discover the bias function and then weigh observations appropriately. We consider cross-sectioning of the illness-death model with the aim of estimating the probability of visiting the illness state before death. We develop simple consistent and asymptotically normal estimators under various assumptions on the model and data collection and, in particular, compare designs with and without a follow-up. These designs are common in surveillance of hospital acquired infections, but estimators currently in use do not properly correct the bias.

Bayesian lasso regression

Hans, C. — Thu, 24 Sep 2009 14:27:56 PDT

The lasso estimate for linear regression corresponds to a posterior mode when independent, double-exponential prior distributions are placed on the regression coefficients. This paper introduces new aspects of the broader Bayesian treatment of lasso regression. A direct characterization of the regression coefficients’ posterior distribution is provided, and computation and inference under this characterization is shown to be straightforward. Emphasis is placed on point estimation using the posterior mean, which facilitates prediction of future observations via the posterior predictive distribution. It is shown that the standard lasso prediction method does not necessarily agree with model-based, Bayesian predictions. A new Gibbs sampler for Bayesian lasso regression is introduced.

Nested Latin hypercube designs

Qian, P. Z. G. — Tue, 22 Sep 2009 08:59:34 PDT

We propose an approach to constructing nested Latin hypercube designs. Such designs are useful for conducting multiple computer experiments with different levels of accuracy. A nested Latin hypercube design with two layers is defined to be a special Latin hypercube design that contains a smaller Latin hypercube design as a subset. Our method is easy to implement and can accommodate any number of factors. We also extend this method to construct nested Latin hypercube designs with more than two layers. Illustrative examples are given. Some statistical properties of the constructed designs are derived.

A unified approach to linearization variance estimation from survey data after imputation for item nonresponse

Kim, J. K., Rao, J. N. K. — Wed, 16 Sep 2009 09:46:40 PDT

Variance estimation after imputation is an important practical problem in survey sampling. When deterministic imputation or stochastic imputation is used, we show that the variance of the imputed estimator can be consistently estimated by a unifying linearize and reverse approach. We provide some applications of the approach to regression imputation, fractional categorical imputation, multiple imputation and composite imputation. Results from a simulation study, under a factorial structure for the sampling, response and imputation mechanisms, show that the proposed linearization variance estimator performs well in terms of relative bias, assuming a missing at random response mechanism.