Testing goodness of fit for the distribution of errors in multivariate linear models

In this paper, to test goodness of fit to any fixed distribution of errors in multivariate linear models, we consider a weighted integral of the squared modulus of the difference between the empirical characteristic function of the residuals and the characteristic function under the null hypothesis. We study the limiting behaviour of this test statistic under the null hypothesis and under alternatives. In the asymptotics, the rank of the design matrix is allowed to grow with the sample size.     

滑动最小二乘插值函数配点法

给出了利用滑动最小二乘法构造加权残值法中试函数的方法，对试函数中的基函数以及权函数的选取提出了建议；该试函数适用于任何定解问题，采用配点法求出试函数中的系数，进而可得到定解问题的近似解，利用该试函数对简支板的挠曲，悬臂梁的弯曲，以及中心具有小圆孔的大板的均匀拉伸等三个例子进行了数值计算，并与理论结果进行对比，同时还检验了该法的精度对结点数，配点数，以及结点影响半径的依赖情况，结果表明，该试函数适用于多种边值问题，且精度高，该法简化了选择试函数的过程，尤其适用于工程中的各种数值计算。     

Fitting models to correlated data III: A comparison between residual analysis and other methods

Applications of the χ² test, the F test, the Durbin-Watson d test, and the f (or Sign) test, to examples of correlated data treatment, show important drawbacks with the d test and (apparently) with the f test. An analytical approach based on residual analysis suggests an improvement in their use that leads to better results at lowest order; it also points out a distinction between goodness-of-fit tests, as the f test, and goodness-of-modeling tests, as the χ² and F tests. The residual analysis method is applied to the same examples; it looks faster, simpler, and often more accurate than the classical ones.     

Fitting models to correlated data (large samples)

The study of the ordered series of residuals of a fit proved to be useful in evaluating separately the pure experimental error and the model bias leading to a possible improvement of the modeling [J. Mol. Spectrosc. 217 (2003) 32]. In the present work this procedure is extended to homogeneous correlated data. This new method allows a separate estimation of pure experimental error, model bias, and data correlation; furthermore, it brings a new insight into the difference between goodness of fit and model relevance. It can be considered either as a study of ‘random systematic errors’ or as an extended approach of the Durbin-Watson problem [Biometrika 37 (1950) 409] taking into account the model error. In the present work an empirical approach is proposed for large samples (n?500) where numerical tests are done showing the accuracy and the limits of the method.     

Benchmarking and validating algorithms that estimate p<Emphasis Type="Italic">K</Emphasis><Subscript>a</Subscript> values of drugs based on their molecular structures

The REGDIA regression diagnostics algorithm in S-Plus is introduced in order to examine the accuracy of pK _a predictions made with four updated programs: PALLAS, MARVIN, ACD/pKa and SPARC. This report reviews the current status of computational tools for predicting the pK _a values of organic drug-like compounds. Outlier predicted pK _a values correspond to molecules that are poorly characterized by the pK _a prediction program concerned. The statistical detection of outliers can fail because of masking and swamping effects. The Williams graph was selected to give the most reliable detection of outliers. Six statistical characteristics (F _exp, R ², , MEP, AIC, and s(e) in pK _a units) of the results obtained when four selected pK _a prediction algorithms were applied to three datasets were examined. The highest values of F _exp, R ², , the lowest values of MEP and s(e), and the most negative AIC were found using the ACD/pK _a algorithm for pK _a prediction, so this algorithm achieves the best predictive power and the most accurate results. The proposed accuracy test performed by the REGDIA program can also be applied to test the accuracy of other predicted values, such as log P, log D, aqueous solubility or certain physicochemical properties of drug molecules.     

On the convergence of Schröder iteration functions for pth roots of complex numbers

In this work a condition on the starting values that guarantees the convergence of the Schröder iteration functions of any order to a pth root of a complex number is given. Convergence results are derived from the properties of the Taylor series coefficients of a certain function. The theory is illustrated by some computer generated plots of the basins of attraction.     

Fitting models to correlated data II (small samples, large correlations, and wrong models)

The estimation of experimental error, data correlation and model error by the analysis of the fit residuals in an ordered sequence of data [J. Mol. Spectrosc. 224 (2004) 73] is improved by a better insight in the calculation of the statistical relevant quantities. The biases previously observed for too small samples of data, very wrong models or highly correlated data, are partially corrected and this leads to better estimates in some extreme cases provided these unfavourable conditions are not combined. However, such results are always obtained with a decrease in accuracy.     

Rank-based inference tools for copula regression,with property and casualty insurance applications

Rank-based procedures are commonly used for inference in copula models for continuous responses whose behavior does not depend on covariates. This paper describes how these procedures can be adapted to the broader framework in which (possibly non-linear) regression models for the marginal responses are linked by a copula that does not depend on covariates. The validity of many of these techniques can be derived from the asymptotic equivalence between the classical empirical copula process and its analog based on suitable residuals from the marginal models. Moment-based parameter estimation and copula goodness-of-fit tests are shown to remain valid under weak conditions on the marginal error term distributions, even when the residual-based empirical copula process fails to converge weakly. The performance of these procedures is evaluated through simulation in the context of two general insurance applications: micro-level multivariate insurance claims, and dependent loss triangles.     

Graphical Diagnostics for Markov Models for Categorical Data

Markov models are widely used as a method for describing categorical data that exhibit stationary and nonstationary autocorrelation. However, diagnostic methods are a largely overlooked topic for Markov models. We introduce two types of residuals for this purpose: one for assessing the length of runs between state changes, and the other for assessing the frequency with which the process moves from any given state to the other states. Methods for calculating the sampling distribution of both types of residuals are presented, enabling objective interpretation through graphical summaries. The graphical summaries are formed using a modification of the probability integral transformation that is applicable for discrete data. Residuals from simulated datasets are presented to demonstrate when the model is, and is not, adequate for the data. The two types of residuals are used to highlight inadequacies of a model posed for real data on seabed fauna from the marine environment.

Supplemental materials, including an R-package RMC with functions to perform the diagnostic measures on the class of models considered in this article, are at the journal’s website. The R-package is also available at CRAN.     

Multi-way partial least squares modeling of water quality data

A 10 years surface water quality data set pertaining to a polluted river was analyzed using partial least squares (PLS) regression models. Both the unfold-PLS and N-PLS (tri-PLS and quadri-PLS) models were calibrated through leave-one out cross-validation method. These were applied to the multivariate, multi-way data array with a view to assess and compare their predictive capabilities for biochemical oxygen demand (BOD) of river water in terms of their relative mean squares error of cross-validation, prediction and variance captured. The sum of squares of residuals and leverages were computed and analyzed to identify the sites, variables, years and months which may have influence on the constructed model. Both the tri- and quadri-PLS models yielded relatively low validation error as compared to unfold-PLS and captured high variance in model. Moreover, both of these methods produced acceptable model precision and accuracy. In case of tri-PLS the root mean squares errors were 1.65 and 2.17 for calibration and prediction, respectively; whereas these were 2.58 and 1.09 for quadri-PLS. At a preliminary level it seems that BOD can be predicted but a different data arrangement is needed. Moreover, analysis of the scores and loadings plots of the N-PLS models could provide information on time evolution of the river water quality.