Volume 17, Issue 1

When addressing semiparametric problems with parametric restrictions (assumptions on the distribution), the efficient score (ES) of a parameter is often important for generating useful estimates. However, usual derivation of ES, although conceptually simple, is often lengthy and with many steps that do not help in understanding why its final form arises. This drawback often casts onto semiparametric estimation a mantle that can turn away otherwise able doctoral students or researchers. Here we show that many ESs can be obtained as a one-step derivation after we characterize those features (envelopes) of the unrestricted problem that are constrained in the restricted problem. We demonstrate our arguments in three problems with known ES but whose usual derivations are lengthy. We show that the envelope-based derivation is dramatically explanatory and compact, needing essentially two lines where the standard approach needs 10 or more pages. This suggests that the envelope method can add useful intuition and exegesis to both teaching and research of semiparametric estimation.

We propose a method for summarizing the strength of association between a set of variables and a multivariate outcome. Classical summary measures are appropriate when linear relationships exist between covariates and outcomes, while our approach provides an alternative that is useful in situations where complex relationships may be present. We utilize machine learning to detect nonlinear relationships and covariate interactions and propose a measure of association that captures these relationships. A hypothesis test about the proposed associative measure can be used to test the strong null hypothesis of no association between a set of variables and a multivariate outcome. Simulations demonstrate that this hypothesis test has greater power than existing methods against alternatives where covariates have nonlinear relationships with outcomes. We additionally propose measures of variable importance for groups of variables, which summarize each groups’ association with the outcome. We demonstrate our methodology using data from a birth cohort study on childhood health and nutrition in the Philippines.

Stein-type shrinkage techniques are applied to the parametric components of a semi-nonparametric regression model recently proposed by (Ma et al. 2015: 285–303). On the basis of an uncertain prior information (restrictions) about the parameters of interest, shrinkage techniques are shown to improve the accuracy of the model. The effectiveness of the proposed estimators are corroborated by a simulation study.

We propose a multivariate regression model to deal with multiple continuous bounded data. The proposed model is based on second-moment assumptions, only. We adopted the quasi-score and Pearson estimating functions for estimation of the regression and dispersion parameters, respectively. Thus, the proposed approach does not require a multivariate probability distribution for the variable response vector. The multivariate quasi-beta regression model can easily handle multiple continuous bounded outcomes taking into account the correlation between the response variables. Furthermore, the model allows us to analyze continuous bounded data on the interval [0, 1], including zeros and/or ones. Simulation studies were conducted to investigate the behavior of the NORmal To Anything (NORTA) algorithm and to check the properties of the estimating function estimators to deal with multiple correlated response variables generated from marginal beta distributions. The model was motivated by a data set concerning the body fat percentage, which was measured at five regions of the body and represent the response variables. We analyze each response variable separately and compare it with the fit of the multivariate proposed model. The multivariate quasi-beta regression model provides better fit than its univariate counterparts, as well as allows us to measure the correlation between response variables. Finally, we adapted diagnostic tools to the proposed model. In the supplementary material, we provide the data set and R code.

In integrative analysis parametric or nonparametric methods are often used. The former is easier for interpretation but not robust, while the latter is robust but not easy to interpret the relationships among the different types of variables. To combine the advantages of both methods and for flexibility, here a system of semiparametric projection non-linear regression models is proposed for the integrative analysis, to model the innate coordinate structure of these different types of data, and a diagnostic tool is constructed to classify new subjects to the case or control group. Simulation studies are conducted to evaluate the performance of the proposed method, and shows promising results. Then the method is applied to analyze a real omics data from The Cancer Genome Atlas study, compared the results with those from the similarity network fusion, another integrative analysis method, and results from our method are more reasonable.

Linear regression with measurement error in the covariates is a heavily studied topic, however, the statistics/econometrics literature is almost silent to estimating a multi-equation model with measurement error. This paper considers a seemingly unrelated regression model with measurement error in the covariates and introduces two novel estimation methods: a pure Bayesian algorithm (based on Markov chain Monte Carlo techniques) and its mean field variational Bayes (MFVB) approximation. The MFVB method has the added advantage of being computationally fast and can handle big data. An issue pertinent to measurement error models is parameter identification, and this is resolved by employing a prior distribution on the measurement error variance. The methods are shown to perform well in multiple simulation studies, where we analyze the impact on posterior estimates for different values of reliability ratio or variance of the true unobserved quantity used in the data generating process. The paper further implements the proposed algorithms in an application drawn from the health literature and shows that modeling measurement error in the data can improve model fitting.

Progression-free survival (PFS), defined as the time from randomization to progression of disease or death, has been indicated as an endpoint to support accelerated approval of certain cancer drugs by the U.S. FDA. The standard Kaplan–Meier (KM) estimator of PFS, however, can result in significantly biased estimates. A major source for the bias results from the substitution of censored progression times with death times. Currently, to ameliorate this bias, several sensitivity analyses based on rather arbitrary definitions of PFS censoring are usually conducted. In addition, especially in the advanced cancer setting, patients with censored progression and observed death times have the potential to experience disease progression between those two times, in which case their true PFS time is actually between those times. In this paper, we present two alternative nonparametric estimators of PFS, which statistically incorporate survival data often available for those patients who are censored with respect to progression to obtain less biased estimates. Through extensive simulations, we show that these estimators greatly reduce the bias of the standard KM estimator and can also be utilized as alternative sensitivity analyses with a solid statistical basis in lieu of the arbitrarily defined analyses currently used. An example is also given using an ECOG-ACRIN Cancer Research Group advanced breast cancer study.

The receiver operating-characteristic (ROC) curve is a graphical statistical tool routinely used for studying the classification accuracy in both, diagnostic and prognosis problems. Given the different nature of these situations, ROC curve estimation has been separately considered for binary (diagnostic) and time-to-event (prognosis) outcomes, even for data coming from the same study design. In this work, the authors propose a two-stage ROC curve estimator which allows to link both contexts through a general prediction model (first-stage) and the empirical cumulative estimator of the distribution function (second-stage) of the considered test (marker) on the total population. The so-called two-stage Mixed-Subject (sMS) approach proves its behavior on both, large-samples (theoretically) and finite-samples (via Monte Carlo simulations). Besides, a useful asymptotic distribution for the concomitant area under the curve is also computed. Results show the ability of the proposed estimator to fit non-standard situations by considering flexible predictive models. Two real-world examples, one with binary and one with time-dependent outcomes, help us to a better understanding of the proposed methodology on usual practical circumstances. The R code used for the practical implementation of the proposed methodology and its documentation is provided as supplementary material.

Ranked set sampling (RSS), known as a cost-effective sampling technique, requires that the ranker gives a complete ranking of the units in each set. Frey (2012) proposed a modification of RSS based on partially ordered sets, referred to as RSS-t in this paper, to allow the ranker to declare ties as much as he/she wishes. We consider the problem of estimating the area under a receiver operating characteristics (ROC) curve using RSS-t samples. The area under the ROC curve (AUC) is commonly used as a measure for the effectiveness of diagnostic markers. We develop six nonparametric estimators of the AUC with/without utilizing tie information based on different approaches. We then compare the estimators using a Monte Carlo simulation and an empirical study with real data from the National Health and Nutrition Examination Survey. The results show that utilizing tie information increases the efficiency of estimating the AUC. Suggestions about when to choose which estimator are also made available to practitioners.

We analyse data from the Southall And Brent REvisited (SABRE) tri-ethnic study, where measurements of metabolic and anthropometric variables have been recorded. In particular, we focus on modelling the distribution of insulin resistance which is strongly associated with the development of type 2 diabetes. We propose the use of a Bayesian nonparametric prior to model the distribution of Homeostasis Model Assessment insulin resistance, as it allows for data-driven clustering of the observations. Anthropometric variables and metabolites concentrations are included as covariates in a regression framework. This strategy highlights the presence of sub-populations in the data, characterised by different levels of risk of developing type 2 diabetes across ethnicities. Posterior inference is performed through Markov Chains Monte Carlo (MCMC) methods.

Co-localization analysis is a popular method for quantitative analysis in fluorescence microscopy imaging. The localization of marked proteins in the cell nucleus allows a deep insight into biological processes in the nucleus. Several metrics have been developed for measuring the co-localization of two markers, however, they depend on subjective thresholding of background and the assumption of linearity. We propose a robust method to estimate the bivariate distribution function of two color channels. From this, we can quantify their co- or anti-colocalization. The proposed method is a combination of the Maximum Entropy Method (MEM) and a Gaussian Copula, which we call the Maximum Entropy Copula (MEC). This new method can measure the spatial and nonlinear correlation of signals to determine the marker colocalization in fluorescence microscopy images. The proposed method is compared with MEM for bivariate probability distributions. The new colocalization metric is validated on simulated and real data. The results show that MEC can determine co- and anti-colocalization even in high background settings. MEC can, therefore, be used as a robust tool for colocalization analysis.