Search

The authors provide a rigorous large sample theory for linear models whose response variable has been subjected to the Box-Cox transformation. They provide a continuous asymptotic approximation to the distribution of estimators of natural parameters of the model. They show, in particular, that the maximum likelihood estimator of the ratio of slope to residual standard deviation is consistent and relatively stable. The authors further show the importance for inference of normality of the errors and give tests for normality based on the estimated residuals. For non-normal errors, they give adjustments to the log-likelihood and to asymptotic standard errors.

We prove that the standard EWMA mean chart with asymptotic control limits and the EWMA mean chart with time-varying control limits for monitoring mean changes in a normal process with known mean and known variance are ARL-unbiased. Using the results derived we discuss the effects of estimation of the process mean on ARL.

This paper studies general sufficient conditions for the geometric ergodicity and the existence of moments for a class of nonlinear autoregressive models with nonlinear ARCH errors. Applications of these conditions to various well-known nonlinear time series models yield specific sufficient conditions, many of which are new or generalizations of existing conditions.

Threshold autoregressive models are widely used in time-series applications. When building or using such a model, it is important to know whether conditional heteroscedasticity exists. The authors propose a nonparametric test of this hypothesis. They develop the large-sample theory of a test of nonlinear conditional heteroscedasticity adapted to nonlinear autoregressive models and study its finite-sample properties through simulations. They also provide percentage points for carrying out this test, which is found to have very good power overall.

Two EWMA charts are usually required to monitor both process mean and variability. In this paper, we propose a new EWMA chart which effectively combines the usual two EWMA charts into one chart. In...

A least squares non-parametric discovery process model has been developed recently for assessing petroleum resources in mature plays with complex size distribution characteristics. Applications of this model to oil plays in the Colorado Group suggests that these techniques can be used to assess petroleum plays composed of a mixture of pools comprising both conventional and unconventional petroleum accumulations. We use the Cardium oil play of the Colorado Group in the Western Canada Sedimentary Basin to illustrate use of the non-parametric approach for such a mixed petroleum play. Our Colorado Group resource assessment indicates a large remaining oil resource potential in this petroleum system, largely in tight unconventional reservoirs.

The authors give easy-to-check sufficient conditions for the geometric ergodicity and the finiteness of the moments of a random process xt = ϕ(xt-1,…, xt-p) + ϵtσ(xt-1,…, xt-q) in which ϕ: Rp R, σ Rq R and (ϵt) is a sequence of independent and identically distributed random variables. They deduce strong mixing properties for this class of nonlinear autoregressive models with changing conditional variances which includes, among others, the ARCH(p), the AR(p)-ARCH(p), and the double-threshold autoregressive models. Les auteurs enoncent des conditions faciles a verifier qui garantissent l'ergodicite geometrique et la finitude des moments d'un processus aleatoire xt = ϵ(xt-1,…, xt-p) + ϵtσ(xt-1,…, xt-q) defini a partir de fonctions ϕ: Rp R, σ: Rq R et d'une suite (ϵt) d'aleas independants de měme loi. Ils en deduisent des proprietes de melangeance forte pour cette classe de modeles autoregressifs non lineaires a variances conditionnelles variables qui englobe, entre autres, les modeles ARCH(p), les modeles AR(p)-ARCH(p), ainsi que les modeles autoregressifs a double seuil.

We give simple and concise expressions of influence measures for general linear models with correlated errors. We show that the commonly used influence measures can be expressed as functions of the generalized leverages and residuals and have the same forms as those in standard linear regression models.

The p chart for monitoring the fraction, p, of nonconforming products requires very large sample sizes to achieve any power in detecting decreases in p, and the false alarm for probabilities differ considerably from the nomial normal values. Two alterna..

The u chart for monitoring λ, the average number of defects per inspection unit, requires λ to be large to have positive lower control limits, and the false alarm probabilities differ considerably from the nominal normal values. Based on nonlinear transformations, some alternative charts, such as the chart, can achieve large improvement over the u chart. We show in this paper that with very simple adjustments to the control limits of the u chart, one can achieve equal or even better improvement while still working on the original scale. In particular, using the very simple adjustments is equivalent to using the exact Poisson distribution to design a chart. Some guidelines for choosing an appropriate chart are given.

A Box-Cox transformed linear model usually has the form y(λ) = μ + β1x1 +… + βpxp + oe, where y(λ) is the power transform of y. Although widely used in practice, the Fisher information matrix for the unknown parameters and, in particular, its inverse have not been studied seriously in the literature. We obtain those two important matrices to put the Box-Cox transformed linear model on a firmer ground. The question of how to make inference on β = (β1,…,βp)T when λ; is estimated from the data is then discussed for large but finite sample size by studying some parameter-based asymptotics. Both unconditional and conditional inference are studied from the frequentist point of view. Un modele lineaire transforme suivant Box-Cox a habituellement la forme y(λ) = μ+β1x1 +…+ βpxp + oϵ, ou y(λ) est est la transformation puissance de y. Malgre le fait qu'elle est communement utilisee dans la pratique, la matrice d'information de Fisher pour les parametres inconnus et, en particulier, son inverse n'ont pas ete etudiees serieusement dans la litterature. Dans cet article, nous obtenons ces deux importantes matrices ce qui permet de placer le modele lineaire transforme suivant Box-Cox sur des bases plus solides. Nous etudions I'inference au sujet de β = (β1,….βp)T lorsque λ est estime a partir des donnees, cela pour un echantillon fini de grande taille et en considerant certaines asymptotiques fondees sur des parametres. Nous etudions, d'un point de vue frequentiel, les inferences inconditionnelle et conditionnelle.

Time series modelling methods can be loosely classified as (i) parametric methods and (ii) non-parametric methods. Within a usually quite flexible but well structured family of models, the parametric modelling process typically consists of model identification, parameter estimation, model diagnostic checking, and forecasting. On the other hand, within a much less structured framework, different non-parametric smoothing techniques are usually used to bring out the features of the observed time series, however, few serious and systematic attempts have been made to model time series non-parametrically. We concentrate on a non-parametric method based on multivariate adaptive regression splines (MARS). Parallel to the parametric modelling process, we systemize a non-parametric modelling process as (i) model perception (where a very large spline expansion of a very large family of models is specified), (ii) model search (forward plus backward search to come up with a model), (iii) model diagnostic checking, and (iv) forecasting. The major difference between the MARS and the parametric methods is that the potential models for the MARS method form a family which is much larger than any family of parametric time series models, and the local structures found in the data are used to guide the search for a fitted model. Also, unlike most non-parametric methods, MARS time series models can be analytically written down. In this paper, we present the results of an empirical comparison of parametric (ARIMA) and non-parametric (MARS) time series modelling methods. Eight environmental time series are used for the comparison. © 1997 by John Wiley & Sons Ltd.

Very often in regression analysis, a particular functional form connecting known covariates and unknown parameters is either suggested by previous work or demanded by theoretical considerations so that the deterministic part of the responses has a known form. However, the underlying error structure is often less well understood. In this case, the transform-both-sides (TBS) models are appropriate. In this paper we generalize the usual TBS models and develop tests to assess goodness of fit when fitting TBS or GTBS models. Parameter estimation is discussed, and tests based on the Cramer-von Mises statistic and the Anderson-Darling statistic are presented with a table suitable for finite-sample applications. Tres souvent, lors de l'analyse de la regression, une forme fonctionnelle particuliere reliant des “covariates” connus et des parametres inconnus est soit suggeree par des travaux anterieurs, soit exigee par des considerations theoriques afin que la partie deterministe des reponses ait une forme connue. Toutefois, la structure d'erreur sous-jacente est souvent moins bien comprise. Dans ce cas, les modeles TBS (“transform-both-sides”) sont appropries. Cet article generalise les modeles TBS habituels et developpe des tests pour evaluer la validite de l'ajustement lorsqu'on ajuste des modeles TBS ou GTBS. Nous discutons l'estimation des parametres et presentons des tests fondes sur la statistique Cramer-von Mises et la statistique Anderson-Darling avec un tableaux convenant a des applications a echantillon fini.

Skewed distributions play an important role in the analysis of data from quality and reliability experiments. Very often unknown parameters must be estimated from the sample data in order to test whether the data has come from a certain family of distri..

A specific scale must be used to collect data in an experiment, but this scale may or may not be the appropriate scale to analyze the data thus collected. In this paper, we study a family of three-parameter distributions which includes the log-normal distribution as a special case, and one of the parameters plays the role of choosing a suitable scale for subsequent analysis. Some basic properties and maximum likelihood estimation of the unknown parameters are discussed, and an easy-to-use goodness-of-fit test is developed. A real data set, to which the usual Weibull, log-normal and gamma distributions cannot be fitted, is used to demonstrate the potential applicability of this family of distributions.

In this paper we consider the estimation of the degree of differencing d in the fractionally integrated autoregressive moving-average time series model ARFIMA (p, d, q). Using lag window spectral density estimators we develop a regression type estimator of d which is easy to calculate and does not require prior knowledge of p and q. Some large sample properties of the estimator are studied and the performance of the estimator for small samples is investigated using the simulation method for a range of commonly used lag windows. Some practical recommendations on the choice of lag windows and the choice of the window parameters are provided.

Weak convergence results are obtained in this paper for the residual empirical processes in linear models, when the normal error assumption is replaced by a number of other error assumptions such as the Gamma family and the inverse Gaussian family. Application to asymptotic power calculations for a class of tests of fit is discussed.

The shape of the commonly used distributions in extreme value analysis is determined by either a specific functional form or a shape parameter. In this paper, we develop a family of distributions which uses two parameters to determine the shape: a balance parameter to control symmetry and a tail order parameter to control the thickness of the tail and modes. Each member in this family has closed forms for its density, cumulative distribution, and quantile function. We discuss the maximum likelihood and minimum distance estimation of the unknown parameters and illustrate the usefulness of the new distribution in extreme value analysis with a real data set.

The performance of an X-bar chart is usually studied under the assumption that the process standard deviation is well estimated and does not change. This is, of course, not always the case in practice. We find that X-bar charts are not robust against errors in estimating the process standard deviation or changing standard deviation. In this paper we discuss the use of a t chart and an exponentially weighted moving average (EWMA) t chart to monitor the process mean. We determine the optimal control limits for the EWMA t chart and show that this chart has the desired robustness property. Copyright © 2009 John Wiley & Sons, Ltd.

When monitoring the process variability, it is a common practice that a Phase I data set is used to estimate the unknown in-control process standard deviation σ0 or variance [Formula: see text] to set up the control limits, then monitoring proceeds. Once the process is considered to be in-control, the estimated control limits are assumed as fixed. This practice ignores the effect of estimating the unknown in-control process variance [Formula: see text]. In this paper, we derive the exact run length distribution of the S2 control chart when the in-control process variance [Formula: see text] is estimated and find that m = 200 or more Phase I samples are needed to neglect the effect of using estimated control limits. New control limits when m is small are also derived.

This paper studies the estimation of a varying-coefficient partially linear regression model which is a generalization of the partially linear regression model and varying-coefficient regression model [Fan and Huang, Manuscript, University of North Carolina, Chapel Hill, USA, 2002]. We focus on the case where some covariates are measured with additive errors. The usual profile least squares and local polynomial estimations lead to biased estimators of the parametric and nonparametric components, respectively, when measurement errors are ignored. By correcting the attenuation we propose a modified profile least squares estimator for the parametric component and a local polynomial estimator for the nonparametric component. We show that the former is consistent, asymptotically normal and achieves the rate in the law of the iterated logarithm, and the latter achieves the optimal strong convergence rate of the usual nonparametric regression. In addition, a consistent estimator is also developed for the error variance. These results can be used to make asymptotically valid statistical inferences. Some simulation studies are conducted to illustrate the finite sample performance of the proposed estimators.

When fitting, by least squares, a linear model (with an intercept term) with $p$ parameters to $n$ data points, the asymptotic behavior of the residual empirical process is shown to be the same as in the single sample problem provided $p^3 \log^2 (p) /n \to 0$ for any error density having finite variance and a bounded first derivative. No further conditions are imposed on the sequence of design matrices. The result is extended to more general estimates with the property that the average error and average squared error in the fitted values are on the same order as for least squares.

The importance of the independence between a standard normal and a chi-square random variable in constructing a t random variable is explored. When independence is absent it is demonstrated that the usual t approximation can lead to misleading results.

In this paper jackknifing technique is examined for functions of the parametric component in a partially linear regression model with serially correlated errors. By deleting partial residuals a jackknife-type estimator is proposed. It is shown that the jackknife-type estimator and the usual semiparametric least-squares estimator (SLSE) are asymptotically equivalent. However, simulation shows that the former has smaller biases than the latter when the sample size is small or moderate. Moreover, since the errors are correlated, both the Tukey type and the delta type jackknife asymptotic variance estimators are not consistent. By introducing cross-product terms, a consistent estimator of the jackknife asymptotic variance is constructed and shown to be robust against heterogeneity of the error variances. In addition, simulation results show that confidence interval estimation based on the proposed jackknife estimator has better coverage probability than that based on the SLSE, even though the latter uses the information of the error structure, while the former does not.

This paper is concerned with a partial linear regression model with serially correlated random errors which are unobservable and modeled by a moving-average process of infinite order. We study a class of estimators for the linear regression coefficients as well as the function characterizing the non-linear part of the model, constructed based on general kernel smoothing and least squares methods. The law of iterated logarithm and strong convergence rates of these estimator are derived by truncating the moving-average error process, a procedure widely applied in the analysis of time series. Our results can be used to establish uniform strong convergence rate of the estimators of autocovariance and autocorrelation functions of the error process.