It is my pleasure to present the June 2023 issue of The Canadian Journal of Statistics, which comprises 15 research articles. The issue already available online on the journal’s website.

In the digital era, data are abundant and additional sources of information may be easily accessible. The first four articles of the June issue address some of the resulting statistical challenges. Han, Taylor, and Mukherjee [1] use the empirical likelihood framework to improve prediction in individual regression models by incorporating input from the so-called risk calculators, which typically contain rich information from previous studies, but may be largely black-box with few model details available. Liu and Li [2] consider the situation when data need to be stored on multiple machines due to their massive size. They develop two estimators in these distributed computer settings by combining empirical likelihood methods with the divide-and-conquer strategy. Su, Yin, Zhang, and Zhao [3] study massive survival data with censoring when the assumptions of the Cox proportional hazards model are violated. They too rely on the divide-and-conquer idea: once the accelerated failure time model has been fitted to each subsample, the resulting estimators are integrated using an approximative weighted least squares loss and adaptive LASSO. Shao, Song, and Zhou [4] consider reducing the volume of massive data through downscaling. Using their theoretical results and optimality criteria in experimental design, they develop an algorithm to draw the most informative subsample within the quantile regression framework.

The June issue further presents several contributions to causal inference. Pashley and Bind [5] study multiple treatment comparisons in settings where a full factorial design is infeasible, or where certain treatment combinations may be absent. They show how to design and analyze fractional factorial and incomplete factorial designs using the potential outcomes framework. Jiang, Wallace, and Thompson [6] focus on precision medicine whose aim is to tailor treatment plans to individual patient characteristics. Relying on a network propensity function, they develop dynamic treatment regime estimation that remains doubly robust even under interference stemming from, e.g., links in social networks. Manuel, Sinha, and Wang [7] consider cohort data with a misclassified, multicategory exposure and a binary response in the absence of validation data. They use instrumental variables to reduce bias due to misclassification and formulate sufficient conditions for parameter identification in the logistic model. Finally, Dai, Shen, and Stern [8] propose a nonparametric test of treatment effect heterogeneity in observational studies. Their method does not require any parametric assumptions on the outcomes and uses propensity score weighting to account for potential confounders.

Functional data are at the heart of the work of Ma, Liu, Xu, and Yang [9], who aim to quantify the effect of infinitely-dimensional functional predictors and finite-dimensional scalar covariates on a scalar response in settings where this effect may vary across subgroups of a heterogeneous population. Their method automatically identifies the subgroups and estimates the parameters of the resulting partial functional linear regression with subgroup specific means.

Prediction accuracy of a subject-specific characteristic can be much improved if one can identify a class to which the subject belongs and utilize the information available for that class. Ma and Jiang [10] extend classified mixed model prediction to generalized linear mixed models using a novel matching strategy and show consistency both in terms of prediction and class matching.

Three articles tackle high-dimensional problems. Wang, Liu, Zhang, and Liu for the Alzheimer’s Disease Neuroimaging Initiative [11] develop a method to estimate the number and location of change-points in the regression coefficients of a high-dimensional generalized linear model. Zheng, Wan, and Zhou [12] propose a sufficient dimension reduction procedure in a system of estimating equations with a high number of covariates when there are more moment conditions than unknown parameters and when the responses are missing at random. The so-called epsilon-admissible subset methodology rooted in generalized fiducial inference has been successfully employed to perform model selection in high-dimensional regression settings. In their open access article, Williams, Xie, and Hannig [14] extend this approach to high-dimensional vector autoregression models and prove pairwise, as well as strong, model selection consistency results.

Cholaquidis, Fraiman, Gamboa, and Moreno [13] study the extension of the so-called lens depth to any metric space and establish large-sample properties of its empirical version. They also introduce a weighted version of this depth for Riemannian manifolds and use the depth-depth approach for, e.g., pattern recognition in phylogenetic trees.

The issue closes with a contribution to extreme-value analysis by Stupfler and Usseglio-Carleve [15]. In their paper, the authors construct composite biased-reduced estimators of two widely used risk measures, quantiles and expectiles, by embedding them into the Lp-quantile framework.

Wishing you inspirational readings,

Johanna G. Nešlehová

Editor-in-Chief, The Canadian Journal of Statistics

Table of Contents of the June 2023 Issue of The Canadian Journal of Statistics

1. Integrating information from existing risk prediction models with no model details by Peisong Han, Jeremy M. G. Taylor, and Bhramar Mukherjee
2. Distributed estimation with empirical likelihood by Qianqian Liu and Zhouping Li
3. Divide and conquer for accelerated failure time model with massive time-to-event data by Wen Su, Guosheng Yin, Jing Zhang, and Xingqiu Zhao
4. Optimal subsampling for large-sample quantile regression with massive data by Li Shao, Shanshan Song, and Yong Zhou
5. Causal inference for multiple treatments using fractional factorial designs by Nicole E. Pashley and Marie-Abèle C. Bind
6. Dynamic treatment regimes with interference by Cong Jiang, Michael P. Wallace, and Mary E. Thompson
7. Reducing bias due to misclassified exposures using instrumental variables by Christopher Manuel, Samiran Sinha, and Suojin Wang
8. Nonparametric tests for treatment effect heterogeneity in observational studies by Maozhu Dai, Weining Shen, and Hal S. Stern
9. Subgroup analysis for functional partial linear regression model by Haiqiang Ma, Chao Liu, Sheng Xu, and Jin Yang
10. Classified generalized linear mixed model prediction incorporating pseudo-prior information by Haiqiang Ma and Jiming Jiang
11. Efficient multiple change point detection for high-dimensional generalized linear models by Xianru Wang, Bin Liu, Xinsheng Zhang, Yufeng Liu, for the Alzheimer's Disease Neuroimaging Initiative
12. Missing data analysis with sufficient dimension reduction by Siming Zheng, Alan T. K. Wan, and Yong Zhou
13. Weighted lens depth: Some applications to supervised classification by Alejandro Cholaquidis, Ricardo Fraiman, Fabrice Gamboa, and Leonardo Moreno
14. The EAS approach for graphical selection consistency in vector autoregression models by Jonathan P. Williams, Yuying Xie, and Jan Hannig
15. Composite bias-reduced Lp-quantile-based estimators of extreme quantiles and expectiles by Gilles Stupfler and Antoine Usseglio-Carleve