Parameter approximations for quantile regressions with measurement error

For fixed $\alpha \epsilon (0, 1)$, the quantile regression function gives the $\alpha$th quantile $\theta_{\alpha}(\mathbf{x})$ in the conditional distribution of a response variable Y given the value $\mathbf{X} = \mathbf{x}$ of a vector of covariates. It can be used to measure the effect of covariates not only in the center of a population, but also in the upper and lower tails. A functional that summarizes key features of the quantile specific relationship between X and Y is the vector $\beta_{\alpha}$ of weighted expected values of the vector of partial derivatives of the quantile function $\theta_{\alpha}(\mathbf{x})$. In a nonparametric setting, $\beta_{\alpha}$ can be regarded as a vector of quantile specific nonparametric regression coefficients. In survival analysis models (e.g., Cox's proportional hazard model, proportional odds rate model, accelerated failure time model) and in monotone transformation models used in regression analysis, $\beta_{\alpha}$ gives the direction of the parameter vector in the parametric part of the model. $\beta_{\alpha}$ can also be used to estimate the direction of the parameter vector in semiparametric single index models popular in econometrics. We show that, under suitable regularity conditions, the estimate of $\beta_{\alpha}$ obtained by using the locally polynomial quantile estimate of Chaudhuri (1991a) is $n^{1/2}$-consistent and asymptotically normal with asymptotic variance equal to the variance of the influence function of the functional $\beta_{\alpha}$. We discuss how the estimate of $\beta_{\alpha}$ can be used for model diagnostics and in the construction of a link function estimate in general single index models.

Quantile regression is an important statistical tool for statistical modeling. It has been widely used in various fields including econometrics, medicine, and bioinformatics. Despite its popularity in practice, individually estimated quantile regression functions often cross each other and consequently violate the basic properties of quantiles. In this paper we propose a new method for estimating multiple quantile regression functions without crossing. Both linear and kernel quantile regression models are considered. Several numerical examples are presented to illustrate competitive performance of the proposed method.

Introduction: Quantile regression is a valuable alternative for survival data analysis, enabling flexible evaluations of covariate effects on survival outcomes with intuitive interpretations. It offers practical computation and reliability. However, challenges arise when applying quantile regression to censored data, particularly for upper quantiles. The minimum distance approach, utilizing dual-kernel estimation and the inverse cumulative distribution function, shows promise in addressing these challenges, especially with higher-dimensional covariates. Methods: This study contrasts two methods within the realm of quantile linear regression for survival analysis: check-based modeling and the minimum distance approach. Effectiveness is assessed across various scenarios through comprehensive simulation. Results: The simulation results showed that using the quantile regression model with the minimum distance approach reduces the percentage of root mean square error in parameter estimation compared to the quantile regression models based on the check loss function. Additionally, a larger sample size and reduced censoring percentage led to decreased root mean square error in parameter estimation. Conclusion: The research highlights the benefits of using the minimum distance approach for quantile regression. It reduces errors, improves model predictions, captures patterns, and optimizes parameters even with complete data. However, this approach has limitations. The accuracy of estimated quantiles can be influenced by the choice of distance metric and weighting scheme. The assumption of independence between censoring mechanism and survival time may not hold in real-world scenarios. Additionally, dealing with large datasets can be computationally complex.

In many applications, data are imperfectly collected. Variables are often measured with error and data are missing for various reasons. Regression calibration is often a simple solution to correct measurement error bias for mean regressions. The measurement error-induced bias could be corrected by solving the corrected estimating equations, as in, or, equivalently, by minimizing a corrected quantile regression loss function. As for the mean regression calibration approach designed to correct measurement errors, simple conditional mean imputation is only valid when the estimation equations are linear, but that is not the case for quantile regression. Likewise, ignoring missing observations in the data can lead to efficiency loss or biased estimation. While there is an abundant literature on measurement errors and missing data, there have been little attention devoted to quantile methods directly, primarily due to the lack of parametric likelihood in quantile regression. Ignoring the missing data will undermine study efficiency and sometimes introduce substantial bias.

We consider the problem of estimating quantile regression coefficients in errors-in-variables models. When the error variables for both the response and the manifest variables have a joint distribution that is spherically symmetric but is otherwise unknown, the regression quantile estimates based on orthogonal residuals are shown to be consistent and asymptotically normal. We also extend the work to partially linear models when the response is related to some additional covariate. Regression analysis is routinely carried out in all areas of statistical appli- cations to explain how a dependent variable Y relates to independent variables X. Most authors consider the estimation or inference problems based on data observed on both X and Y variables. However, the covariates are not always observable without error. If X is observed subject to random error, the regres- sion model is usually called the errors-in-variables (EV) model. A careful study of such models is often needed, as the standard results on regression models do not carry over. The best-known is the effect of attenuation for the likelihood- based estimators without correction for the measurement error in X. A detailed coverage of linear errors-in-variables models can be found in Fuller (1987). More recent work on nonlinear models with measurement errors can be found in Car- roll, Ruppert and Stefanski (1995). The literature on EV models are mainly confined to estimating the conditional mean function of Y given X, assuming Gaussian errors. In the present paper we attempt to consider conditional me- dian and other quantile functions, as pioneered by Koenker and Bassett (1978), for a class of unspecified error distributions. For the usefulness of conditional quantiles, see examples and discussions found in Efron (1991) and He (1997), among many others. Let us start with the EV model Yi = X T

In this paper, we consider variable selection for ultra-high dimensional quantile regression model with missing data and measurement errors in covariates. Specifically, we correct the bias in the loss function caused by measurement error by applying the orthogonal quantile regression approach and remove the bias caused by missing data using the inverse probability weighting. A nonconvex Atan penalized estimation method is proposed for simultaneous variable selection and estimation. With the proper choice of the regularization parameter and under some relaxed conditions, we show that the proposed estimate enjoys the oracle properties. The choice of smoothing parameters is also discussed. The performance of the proposed variable selection procedure is assessed by Monte Carlo simulation studies. We further demonstrate the proposed procedure with a breast cancer data set.

Quantile regression neural networks: Implementation in R and application to precipitation downscaling

This paper develops a model for repeated binary regression when a covariate is measured with error. The model allows for estimating the effect of the true value of the covariate on a repeated binary response. The choice of a probit link for the effect of the error-free covariate, coupled with normal measurement error for the error-free covariate, results in a probit model after integrating over the measurement error distribution. We propose a two-stage estimation procedure where, in the first stage, a linear mixed model is used to fit the repeated covariate. In the second stage, a model for the correlated binary responses conditional on the linear mixed model estimates is fit to the repeated binary data using generalized estimating equations. The approach is demonstrated using nutrient safety data from the Diet Intervention of School Age Children (DISC) study.

In the framework of quantile regression, local linear smoothing techniques have been studied by several authors, particularly by Yu and Jones (J Am Stat Assoc 93:228–237, 1998). The problem of bandwidth selection was addressed in the literature by the usual approaches, such as cross-validation or plug-in methods. Most of the plug-in methods rely on restrictive assumptions on the quantile regression model in relation to the mean regression, or on parametric assumptions. Here we present a plug-in bandwidth selector for nonparametric quantile regression that is defined from a completely nonparametric approach. To this end, the curvature of the quantile regression function and the integrated squared sparsity (inverse of the conditional density) are both nonparametrically estimated. The new bandwidth selector is shown to work well in different simulated scenarios, particularly when the conditions commonly assumed in the literature are not satisfied. A real data application is also given.

On choice and effect of weight matrix for response sensitivity-based damage identification with measurement and model errors

This thesis presents a comprehensive exploration of linear mixed models that incorporate measurement errors in specific covariates for longitudinal data with non-ignorable and non-monotone missing responses. The primary objective is to estimate mean response parameters and variance components utilizing a combined methodology of Regression Calibration (RC) and Monte Carlo EM. The investigation begins with an in-depth examination of the RC method applied to linear mixed models, particularly its adaptation to longitudinal data with covariates affected by measurement errors. A thorough simulation study is conducted, involving various mean response functions, revealing that the RC method produces unbiased and efficient estimators in scenarios where the true underlying model includes covariates with measurement errors. Subsequently, linear mixed models are introduced for longitudinal data with non-ignorable missing responses. The thesis proposes a semi-parametric Monte Carlo EM algorithm for the simultaneous estimation of regression parameters and variance components in linear mixed models with non-ignorable and non-monotone missing responses, combining the Monte Carlo EM algorithm of Ibrahim et al. and the RC method. The simulation study demonstrates the effectiveness of the proposed method, even in the presence of a substantial proportion of non-ignorable missing responses. The thesis concludes with an application of the proposed RC-MCEM method to actual longitudinal data from the Canadian Community Health Survey (CCHS), focusing on individuals with alcoholism and exploring how measurement errors in certain covariates can introduce biases in parameter estimation. The complexity of the problem is further elucidated when missingness occurs in the response variable.

In this study, we propose a smoothed weighted quantile regression (SWQR), which combines convolution smoothing with a weighted framework to address the limitations. By smoothing the non-differentiable quantile regression loss function, SWQR can improve computational efficiency and allow for more stable model estimation in complex datasets. We construct an efficient optimization process based on gradient-based algorithms by introducing weight refinement and iterative parameter estimation methods to minimize the smoothed weighted quantile regression loss function. In the simulation studies, we compare the proposed method with two existing methods, including martingale-based quantile regression (MartingaleQR) and weighted quantile regression (WeightedQR). The results emphasize the superior computational efficiency of SWQR, outperforming other methods, particularly WeightedQR, by requiring significantly less runtime, especially in settings with large sample sizes. Additionally, SWQR maintains robust performance, achieving competitive accuracy and handling the challenges of right censoring effectively, particularly at higher quantiles. We further illustrate the proposed method using a real dataset on primary biliary cirrhosis, where it exhibits stable coefficient estimates and robust performance across quantile levels with different censoring rates. These findings highlight the potential of SWQR as a flexible and robust method for analyzing censored data in survival analysis, particularly in scenarios where computational efficiency is a key concern.

Radiomics involves the study of tumor images to identify quantitative markers explaining cancer heterogeneity. The predominant approach is to extract hundreds to thousands of image features, including histogram features comprised of summaries of the marginal distribution of pixel intensities, which leads to multiple testing problems and can miss out on insights not contained in the selected features. In this paper, we present methods to model the entire marginal distribution of pixel intensities via the quantile function as functional data, regressed on a set of demographic, clinical, and genetic predictors to investigate their effects of imaging-based cancer heterogeneity. We call this approach quantile functional regression, regressing subject-specific marginal distributions across repeated measurements on a set of covariates, allowing us to assess which covariates are associated with the distribution in a global sense, as well as to identify distributional features characterizing these differences, including mean, variance, skewness, heavy-tailedness, and various upper and lower quantiles. To account for smoothness in the quantile functions, account for intrafunctional correlation, and gain statistical power, we introduce custom basis functions we call quantlets that are sparse, regularized, near-lossless, and empirically defined, adapting to the features of a given dataset and containing a Gaussian subspace so non-Gaussianness can be assessed. We fit this model using a Bayesian framework that uses nonlinear shrinkage of quantlet coefficients to regularize the functional regression coefficients and provides fully Bayesian inference after fitting a Markov chain Monte Carlo. We demonstrate the benefit of the basis space modeling through simulation studies, and apply the method to Magnetic resonance imaging (MRI)-based radiomic dataset from Glioblastoma Multiforme to relate imaging-based quantile functions to various demographic, clinical, and genetic predictors, finding specific differences in tumor pixel intensity distribution between males and females and between tumors with and without DDIT3 mutations. Supplementary materials for this article, including a standardized description of the materials available for reproducing the work, are available as an online supplement.

This paper study two stratified quantile regression models of the marginal and the conditional varieties. We estimate the quantile functions of these models by using two nonparametric methods of smoothing spline (B-spline) and kernel regression (Nadaraya-Watson). The estimates can be obtained by solve nonparametric quantile regression problem which means minimizing the quantile regression objective functions and using the approach of varying coefficient models. The main goal is discussing the comparison between the estimators of the two nonparametric methods and adopting the best one between them.

We propose a notion of conditional vector quantile function and a vector quantile regression. A conditional vector quantile function (CVQF) of a random vector Y, taking values in ℝd given covariates Z=z, taking values in ℝk, is a map u↦QY∣Z(u,z), which is monotone, in the sense of being a gradient of a convex function, and such that given that vector U follows a reference non-atomic distribution FU, for instance uniform distribution on a unit cube in ℝd, the random vector QY∣Z(U,z) has the distribution of Y conditional on Z=z. Moreover, we have a strong representation, Y=QY∣Z(U,Z) almost surely, for some version of U. The vector quantile regression (VQR) is a linear model for CVQF of Y given Z. Under correct specification, the notion produces strong representation, Y=β(U)⊤f(Z), for f(Z) denoting a known set of transformations of Z, where u↦β(u)⊤f(Z) is a monotone map, the gradient of a convex function, and the quantile regression coefficients u↦β(u) have the interpretations analogous to that of the standard scalar quantile regression. As f(Z) becomes a richer class of transformations of Z, the model becomes nonparametric, as in series modelling. A key property of VQR is the embedding of the classical Monge-Kantorovich's optimal transportation problem at its core as a special case. In the classical case, where Y is scalar, VQR reduces to a version of the classical QR, and CVQF reduces to the scalar conditional quantile function. Several applications to diverse problems such as multiple Engel curve estimation, and measurement of financial risk, are considered.