Canadian Journal of Statistics-Revue Canadienne De Statistique最新文献

We present methods for estimating loss-based measures of the performance of a prediction model in a target population that differs from the source population in which the model was developed, in settings where outcome and covariate data are available from the source population but only covariate data are available on a simple random sample from the target population. Prior work adjusting for differences between the two populations has used various weighting estimators with inverse odds or density ratio weights. Here, we develop more robust estimators for the target population risk (expected loss) that can be used with data-adaptive (e.g., machine learning-based) estimation of nuisance parameters. We examine the large-sample properties of the estimators and evaluate finite-sample performance in simulations. Last, we apply the methods to data from lung cancer screening using nationally representative data from the National Health and Nutrition Examination Survey (NHANES) and extend our methods to account for the complex survey design of the NHANES.

Estimating the COVID-19 infection fatality rate, inferring the latent incidence and predicting the future epidemic evolution are critical to public health surveillance, but often challenging due to limited data availability or quality. Recently, a Bayesian framework combining time series deconvolution of deaths with a parametric Susceptible–Infectious–Recovered (SIR) model was proposed by Irons and Raftery, 2021. We assess the parameter identifiability of the model using the profile likelihood approach and simulations, when only the time series of deaths and seroprevalence survey data are available. The robustness of the model to the more complex but also more realistic Susceptible–Exposed–Infectious–Recovered (SEIR)-based epidemics is evaluated through simulations; the influence of potential biases in the serosurveys on the inference is also investigated. We use a stationary first-order autoregressive prior to account for the variability of transmission rate over time. The results suggest that the model is relatively robust to SEIR-based epidemics, especially when the reproductive number is low, given sufficient information from serosurveys or priors. However, the lack of parameter identifiability under limited data availability cannot be neglected. We apply the model to infer the COVID-19 infections in Ontario and Quebec, Canada during the Omicron era.

We consider estimation of the mean squared prediction error (MSPE) for observed best prediction (OBP) in small area estimation with count data. The OBP method has been previously developed in this context by Chen et al. (Journal of Survey Statistics and Methodology, 3, 136–161, 2015). However, estimation of the MSPE remains a challenging problem due to potential model misspecification that is considered in this setting. The latter authors proposed a bootstrap method for estimating the MSPE, whose theoretical justification is not clear. We propose to use a Prasad–Rao-type linearization method to estimate the MSPE. Unlike the traditional linearization approaches, our method is computationally oriented and easier to implement in the same regard. Theoretical properties and empirical performance of the proposed method are studied. A real-data application is considered.

Order-restricted hypothesis testing problems frequently arise in practice, including studies involving regression models for longitudinal data. These tests are known to be more powerful than tests that ignore such restrictions. In this article, we consider order-restricted tests for nonlinear mixed-effects models with measurement errors in time-dependent covariates. We propose to use a multiple imputation method to address measurement errors, since this approach allows us to use existing complete-data methods for order-restricted tests. Some theoretical results are presented. We evaluate our proposed methods via simulation studies that demonstrate they are more powerful than either a competing naive method or a two-step approach to testing hypotheses. We illustrate the use of our proposed approach by analyzing data from an HIV/AIDS study.

We propose a general mixture of Markov jump processes. The key novel feature of the proposed mixture is that the generator matrices of the Markov processes comprising the mixture are entirely unconstrained. The Markov processes are mixed with distributions that depend on the initial state of the mixture process. The maximum likelihood (ML) estimates of the mixture's parameters are obtained from continuous realizations of the mixture process and their standard errors from an explicit form of the observed Fisher information matrix, which simplifies the Louis (Journal of the Royal Statistical Society Series B, 44:226–233, 1982) general formula for the same matrix. The asymptotic properties of the ML estimators are also derived. A simulation study verifies the estimates' accuracy. The proposed mixture provides an exploratory tool for identifying the homogeneous subpopulations in a heterogeneous population. This is illustrated with an application to a medical dataset.

We study the first-order stochastic dominance (SD) test in the context of two independent random samples. We introduce several test statistics that effectively capture violations of the dominance relationship, particularly in the tail regions. Additionally, we develop a resampling procedure to compute the $��p�$-values or critical values for these tests. The proposed tests have asymptotic type I error rates for frontal configurations equal to the nominal level $��\alpha �$. Furthermore, their powers approach 1 for any fixed alternatives. Through simulation experiments, we demonstrate that our SD tests outperform the recentring test proposed by Donald and Hsu (2016) as well as the integral-type test presented by Linton et al. (2010) in various scenarios discussed in existing literature. We also employ the proposed tests to analyze changes in the distribution of household income in the United Kingdom over time. The proposed tests offer some insights into potential dominance relationships within this context.

A tolerance band for a functional response provides a region that is expected to contain a given fraction of observations from the sampled population at each point in the domain. This band is a functional analogue of the tolerance interval for a univariate response. Although the problem of constructing functional tolerance bands has been considered for a Gaussian response, it has not been investigated for non-Gaussian responses, which are common in biomedical applications. We describe a methodology for constructing tolerance bands for two non-Gaussian members of the exponential family: binomial and Poisson. The approach is to first model the data using the framework of generalized functional principal components analysis. Then, a parameter is identified in which the marginal distribution of the response is stochastically monotone. We show that the tolerance limits can be readily obtained from confidence limits for this parameter, which in turn can be computed using large-sample theory and bootstrapping. Our proposed methodology works for both dense and sparse functional data. We report the results of simulation studies designed to evaluate its performance and get recommendations for practical applications. We illustrate our proposed method using two actual biomedical studies, and also provide computer source code that implements our method.

Interval-censored data arise when a failure process is under intermittent observation and failure status is only known at assessment times. We consider the development of predictive algorithms when training samples involve interval censoring. Using censoring unbiased transformations and pseudo-observations, we define observed data loss functions, which are unbiased estimates of the corresponding complete data loss functions. We show that regression trees based on these loss functions can recover the tree structure and yield good predictive accuracy. An application is given to a study involving individuals with psoriatic arthritis where the aim is to identify genetic markers useful for the prediction of axial disease within 10 years of a baseline assessment.

Consider the problem of benchmarking small-area estimates under multiplicative models with positive parameters. The goal is to propose a loss function that guarantees positive constrained estimates of small-area parameters in this situation. The weighted precautionary loss function is introduced to solve the problem. Compared with the weighted Kullback–Leibler (KL) loss function, our proposed loss function penalizes underestimation of the small-area parameters of interest more for small values of parameters. This property is appealing when we estimate disease rates. It tends to give larger estimates of small-area parameters compared with those obtained under the KL loss function. The hierarchical empirical Bayes and constrained hierarchical empirical Bayes estimates of small-area parameters and their corresponding risk functions under the new proposed loss function are obtained. The performance of the proposed methods is investigated using simulation studies and a real dataset.

Data collected over time are common in applications and may contain censored or missing observations, making it difficult to use standard statistical procedures. This article proposes an algorithm to estimate the parameters of a censored linear regression model with errors serially correlated and innovations following a Student-$��t�$ distribution. This distribution is widely used in the statistical modelling of data containing outliers because its longer-than-normal tails provide a robust approach to handling such data. The maximum likelihood estimates of the proposed model are obtained through a stochastic approximation of the EM algorithm. The methods are applied to an environmental dataset regarding ammonia-nitrogen concentration, which is subject to a limit of detection (left censoring) and contains missing observations. Additionally, two simulation studies are conducted to examine the asymptotic properties of the estimates and the robustness of the model. The proposed algorithm and methods are implemented in the R package ARCensReg.