On Generalized Linear Model Method for Detecting Dispersion Effects in Unreplicated Factorial Designs

There are numerous approaches to screen location effects for unreplicated experiments, but only a handful to screen dispersion effects. Generalized linear models, popular in analyses of non-normal data, were recently proposed to screen both location and dispersion effects simultaneously. This paper illustrates and explains the impact of unidentified location effects on dispersion effects identification for such procedures. A remedy is proposed to recover the loss of power of the GLM method due to such impact.     

D-Optimal Two-Level Fractional Factorial Designs of Resolution III with Two Dispersion Factors

The problem of finding D-optimal designs, with two dispersion factors, for the estimation of all location main effects is investigated in the class of regular unreplicated two-level fractional factorial designs of resolution III. Designs having length three words involving both of the dispersion factors in the defining relation are shown to be inferior in terms of D-optimality. Tables of factors that are named as the two dispersion factors so that the resulting design is either D-optimal or has the largest determinant of the information matrix are provided. Rank-order of designs is studied when the number of length three words involving either one of the dispersion factors and the number of length four words involving both of the dispersion factors are fixed. Rank-order of designs when the numbers of aforementioned words are less than or equal to ten is given.     

Rejoinder to 'Ahmed, M.S. (1998). A note on regression-type estimators using multiple auxiliary information.'

In the estimators t ₃ , t ₄ , t ₅ of Mukerjee, Rao & Vijayan (1987), b _{y x} and b _{y z} are partial regression coefficients of y on x and z , respectively, based on the smaller sample. With the above interpretation of b _{y x} and b _{y z} in t ₃ , t ₄ , t ₅ , all the calculations in Mukerjee at al. (1987) are correct. In this connection, we also wish to make it explicit that b _{x z} in t ₅ is an ordinary and not a partial regression coefficient. The 'corrected' MSEs of t ₃ , t ₄ , t ₅ , as given in Ahmed (1998 Section 3) are computed assuming that our b _{y x} and b _{y z} are ordinary and not partial regression coefficients. Indeed, we had no intention of giving estimators using the corresponding ordinary regression coefficients which would lead to estimators inferior to those given by Kiregyera (1984). We accept responsibility for any notational confusion created by us and express regret to readers who have been confused by our notation. Finally, in consideration of the above, it may be noted that Tripathi & Ahmed's (1995) estimator t ₀ , quoted also in Ahmed (1998), is no better than t ₅ of Mukerjee at al. (1987).     

Median treatment effect in randomized trials

For two response variables y _t and y _c corresponding to two treatments for two policies) T and C , we wish to learn about quantiles of y _t− y _c from the marginal quantiles of y _t and y _c; only one of y _t and y _c is observed for an individual. We find that, in general, this is difficult for quantiles other than the median unless strong assumptions are imposed on how y _t is related to y _c. For the median, we present conditions under which the sign of the median treatment effect is identified.     

Optimal 2n-p fractional factorial designs for dispersion effects under a location-diapersion model

We consider main effects models for 2-level experiments that also include. Parameters characterizing potential dispersion effects due to specified factors. One special case is considered. In this case only a single specified factor is responsible for the dispersion effects. We determine the connection between alias relations and Optimality of a design for estimation of dispersion effects in the class of regu!ar fractional Y - P factorial designs of resolution III or higher. This rmectioil heips US identify those designs that are A-optimal for estimating dispersion effects by a suitable choice of defining contrasts. in particuiar, we show that an increase in efficiency with respect to dispersion effects is accompanied by a loss iii efficiency for estimating the location effects. In practice, one mmt thcrcfcre accept a trade& between the efficiencies associated with estirnates of location effects and dispersion effects.     

The beta distribution as a latent response model for ordinal data (I): Estimation of location and dispersion parameters

Ordinal data are often modeled using a continuous latent response distribution, which is partially observed through windows of adjacent intervals defined by cutpoints. In this paper we propose the beta distribution as a model for the latent response. The beta distribution has several advantages over the other common distributions used, e.g. , normal and logistic. In particular, it enables separate modeling of location and dispersion effects which is essential in the Taguchi method of robust design. First, we study the problem of estimating the location and dispersion parameters of a single beta distribution (representing a single treatment) from ordinal data assuming known equispaced cutpoints. Two methods of estimation are compared: the maximum likelihood method and the method of moments. Two methods of treating the data are considered: in raw discrete form and in smoothed continuousized form. A large scale simulation study is carried out to compare the different methods. The mean square errors of the estimates are obtained under a variety of parameter configurations. Comparisons are made based on the ratios of the mean square errors (called the relative efficiencies). No method is universally the best, but the maximum likelihood method using continuousized data is found to perform generally well, especially for estimating the dispersion parameter. This method is also computationally much faster than the other methods and does not experience convergence difficulties in case of sparse or empty cells. Next, the problem of estimating unknown cutpoints is addressed. Here the multiple treatments setup is considered since in an actual application, cutpoints are common to all treatments, and must be estimated from all the data. A two-step iterative algorithm is proposed for estimating the location and dispersion parameters of the treatments, and the cutpoints. The proposed beta model and McCullagh's (1980) proportional odds model are compared by fitting them to two real data sets.     

On Smooth Statistical Tail Functionals

Many estimators of the extreme value index of a distribution function F that are based on a certain number k _n of largest order statistics can be represented as a statistical tail function al, that is a functional T applied to the empirical tail quantile function Q _n. We study the asymptotic behaviour of such estimators with a scale and location invariant functional T under weak second order conditions on F . For that purpose first a new approximation of the empirical tail quantile function is established. As a consequence we obtain weak consistency and asymptotic normality of T ( Q _n) if T is continuous and Hadamard differentiable, respectively, at the upper quantile function of a generalized Pareto distribution and k _pn tends to infinity sufficiently slowly. Then we investigate the asymptotic variance and bias. In particular, those functionals T re characterized that lead to an estimator with minimal asymptotic variance. Finally, we introduce a method to construct estimators of the extreme value index with a made-to-order asymptotic behaviour     

A small-sample correction factor for S-estimators

S-estimators are frequently used as robust estimators of regression and of location and dispersion. Under certain differentiability conditions, S-estimators of multivariate location and dispersion parameters are consistent [Davies PL. Asymtotic behaviour of S-estimators of multivariate location parameters and dispersion matrices. Ann Stat. 1987;15(3):1269–1292]. However, it has been observed that the S-estimators of dispersion parameters give biased results in the case of small-sample data sets. In this work, we constructed formulas based on simulation studies, which allow us to compute small-sample correction factors for all sample sizes and dimensions for S-estimators of dispersion parameters without having to carry out any new simulations. We considered real data to illustrate the effects of the small-sample correction factor.     

Prior elicitation, variable selection and Bayesian computation for logistic regression models

Bayesian selection of variables is often difficult to carry out because of the challenge in specifying prior distributions for the regression parameters for all possible models, specifying a prior distribution on the model space and computations. We address these three issues for the logistic regression model. For the first, we propose an informative prior distribution for variable selection. Several theoretical and computational properties of the prior are derived and illustrated with several examples. For the second, we propose a method for specifying an informative prior on the model space, and for the third we propose novel methods for computing the marginal distribution of the data. The new computational algorithms only require Gibbs samples from the full model to facilitate the computation of the prior and posterior model probabilities for all possible models. Several properties of the algorithms are also derived. The prior specification for the first challenge focuses on the observables in that the elicitation is based on a prior prediction y ₀ for the response vector and a quantity a ₀ quantifying the uncertainty in y ₀. Then, y ₀ and a ₀ are used to specify a prior for the regression coefficients semi-automatically. Examples using real data are given to demonstrate the methodology.     

Simultaneous estimation of location and dispersion in two-level fractional factorial designs

The reduction of variation is one of the obvious goals in quality improvement. The identification of factors aff ecting the dispersion is a step towards this goal. In this paper, the problem of estimating location effects and dispersion eff ects simultaneously in unreplicated factorial experiments is considered. By making a one-to-one transformation of the response variables, the study of the quadratic functions becomes clearer. The transformation also gives a natural motivation to the model of the variances of the original variables. The covariances of the transformed responses appear as parameters in the variances of the original variables. Results of Hadamard products are used for deriving these covariances. The method of estimating dispersion effects is shown in two illustrations. In a 24 factorial design, the essential covariance matrix of the transformed variables is also presented. The method is also illustrated in a 25-1 fractional design with a model which is saturated in this context.     

Data Dispersion: Now You See It… Now You Don't

Poisson regression is the most well-known method for modeling count data. When data display over-dispersion, thereby violating the underlying equi-dispersion assumption of Poisson regression, the common solution is to use negative-binomial regression. We show, however, that count data that appear to be equi- or over-dispersed may actually stem from a mixture of populations with different dispersion levels. To detect and model such a mixture, we introduce a generalization of the Conway-Maxwell-Poisson (COM-Poisson) regression model that allows for group-level dispersion. We illustrate mixed dispersion effects and the proposed methodology via semi-authentic data.     

IMPROVING UPON THE BEST INVARIANT ESTIMATOR IN MULTIVARIATE LOCATION PROBLEMS

We are concerned with estimators which improve upon the best invariant estimator, in estimating a location parameter θ. If the loss function is L(θ - a) with L convex, we give sufficient conditions for the inadmissibility of δ₀(X) = X. If the loss is a weighted sum of squared errors, we find various classes of estimators δ which are better than δ₀. In general, δ is the convolution of δ₁ (an estimator which improves upon δ₀ outside of a compact set) with a suitable probability density in R^p. The critical dimension of inadmissibility depends on the estimator δ₁ We also give several examples of estimators δ obtained in this way and state some open problems.     

Partitioning Pearson's chi-squared statistic for a partially ordered three-way contingency table

This paper presents a generalization of the partition of the chi-squared statistic presented in Beh & Davy (1998). For a three-way contingency table with one or two sets of ordered categories, the chi-squared statistic partition is defined using orthogonal polynomials. Using this partition, information about the relationship between the variables can be obtained by identifying important associations in terms of the location (linear), dispersion (quadratic) and higher order components. The paper compares these partitions with log-linear models for ordinal data.     

Empirical Likelihood for Non-Smooth Criterion Functions

Abstract.  Suppose that X ₁ ,…,  X _n is a sequence of independent random vectors, identically distributed as a d -dimensional random vector X . Let     be a parameter of interest and     be some nuisance parameter. The unknown, true parameters ( μ ₀ , ν ₀ ) are uniquely determined by the system of equations E { g ( X , μ ₀ , ν ₀ )} =   0 , where g  =  ( g ₁ ,…, g _{p + q} ) is a vector of p + q functions. In this paper we develop an empirical likelihood (EL) method to do inference for the parameter μ ₀ . The results in this paper are valid under very mild conditions on the vector of criterion functions g . In particular, we do not require that g ₁ ,…, g _{p + q} are smooth in μ or ν . This offers the advantage that the criterion function may involve indicators, which are encountered when considering, e.g. differences of quantiles, copulas, ROC curves, to mention just a few examples. We prove the asymptotic limit of the empirical log-likelihood ratio, and carry out a small simulation study to test the performance of the proposed EL method for small samples.     

Analysis of large sets of ranking data

Various procedures, mainly graphical are presented for analyzing large sets of ranking data in which the permutations are not equally likely. One method is based on box plots, the others are motivated by a model originally proposed by Mallows. The model is characterised by two parameters corresponding to location and dispersion. Graphical methods based on Q-Q plots are also discussed for comparing two groups of judges. The proposed methods are illustrated on an empirical data set.     

ε-Repetitions of the Maximum Residuals in an AR(1) Model

This paper considers record values of residuals or prediction errors in a one-parameter autoregressive process and the statistic Z _n = number of ε -repetitions of this record. When the parameter of the autoregression is unknown, the prediction errors, and therefore Z _n , are unobservable. Here an observable analogue ̂ n of Z _n is considered. It is proved that under special conditions the difference Z _n − unobservable. Here an observable analogue ̂ n converges to zero in probability and therefore that unobservable. Here an observable analogue ̂ n has the same asymptotic behaviour as Z _n .     

Semiparametric Likelihood Based Method for Goodness of Fit Tests and Estimation in Upgraded Mixture Models

We use Owen's (1988, 1990) empirical likelihood method in upgraded mixture models. Two groups of independent observations are available. One is z ₁, ..., z _n which is observed directly from a distribution F ( z ). The other one is x ₁, ..., x _m which is observed indirectly from F ( z ), where the x _i s have density ∫ p ( x | z ) dF ( z ) and p ( x | z ) is a conditional density function. We are interested in testing H ₀: p ( x | z ) = p ( x | z ; θ ), for some specified smooth density function. A semiparametric likelihood ratio based statistic is proposed and it is shown that it converges to a chi-squared distribution. This is a simple method for doing goodness of fit tests, especially when x is a discrete variable with finitely many values. In addition, we discuss estimation of θ and F ( z ) when H ₀ is true. The connection between upgraded mixture models and general estimating equations is pointed out.     

Bayesian model selection for join point regression with application to age-adjusted cancer rates

Summary.  The method of Bayesian model selection for join point regression models is developed. Given a set of K +1 join point models M ₀,  M ₁, …,  M _K with 0, 1, …,  K join points respec-tively, the posterior distributions of the parameters and competing models M _k are computed by Markov chain Monte Carlo simulations. The Bayes information criterion BIC is used to select the model M _k with the smallest value of BIC as the best model. Another approach based on the Bayes factor selects the model M _k with the largest posterior probability as the best model when the prior distribution of M _k is discrete uniform. Both methods are applied to analyse the observed US cancer incidence rates for some selected cancer sites. The graphs of the join point models fitted to the data are produced by using the methods proposed and compared with the method of Kim and co-workers that is based on a series of permutation tests. The analyses show that the Bayes factor is sensitive to the prior specification of the variance σ ², and that the model which is selected by BIC fits the data as well as the model that is selected by the permutation test and has the advantage of producing the posterior distribution for the join points. The Bayesian join point model and model selection method that are presented here will be integrated in the National Cancer Institute's join point software ( http://www.srab.cancer.gov/joinpoint/ ) and will be available to the public.     

Non-parametric tests for distributional treatment effect for randomly censored responses

Summary.  For a binary treatment ν =0, 1 and the corresponding 'potential response' Y ⁰ for the control group ( ν =0) and Y ¹ for the treatment group ( ν =1), one definition of no treatment effect is that Y ⁰ and Y ¹ follow the same distribution given a covariate vector X . Koul and Schick have provided a non-parametric test for no distributional effect when the realized response (1− ν ) Y ⁰+ ν Y ¹ is fully observed and the distribution of X is the same across the two groups. This test is thus not applicable to censored responses, nor to non-experimental (i.e. observational) studies that entail different distributions of X across the two groups. We propose ' X -matched' non-parametric tests generalizing the test of Koul and Schick following an idea of Gehan. Our tests are applicable to non-experimental data with randomly censored responses. In addition to these motivations, the tests have several advantages. First, they have the intuitive appeal of comparing all available pairs across the treatment and control groups, instead of selecting a number of matched controls (or treated) in the usual pair or multiple matching. Second, whereas most matching estimators or tests have a non-overlapping support (of X ) problem across the two groups, our tests have a built-in protection against the problem. Third, Gehan's idea allows the tests to make good use of censored observations. A simulation study is conducted, and an empirical illustration for a job training effect on the duration of unemployment is provided.