As I write, this year’s Clarivate journal rankings have been released. I am pleased to report that the impact factor of CJS has increased to 0.8. This has bumped our journal one category up, bringing it on a par with The Scandinavian Journal of Statistics and the Annals of the Institute of Statistical Mathematics, among others. Of course, these numbers do not reflect everything. However, many universities use them, and so I hope that the higher ranking of our journal will encourage you to consider it as a venue for your work. After all, it is the quality of published articles that make out a journal’s reputation. Given our vibrant and talented statistical community, I’m sure we can make it better still. If you have suggestions for improvement, don’t hesitate to reach out to me.
My editorial team and I continue to work hard to improve your publishing experience with CJS. Articles are now beginning to appear in Early View in our new look, designed by Professor Vincent Goulet from Université Laval. Wiley has now completed the emulation of our new design, which means that articles are again appearing online without delay. Sincere thanks to all affected authors for their patience with this somewhat tedious transition process.
The June issue will be the penultimate one in the old design. It is already available electronically on the journal’s website and consists of 14 research articles. Here is a digest of their contributions.
In the opening article, Gao and Wakefield [1] propose a model-assisted estimator that not only leverages covariate information and spatial smoothing but also accounts for the study design. Their method can achieve both design and model consistency and helps generate accurate subnational estimates of health and demographic indicators in countries with limited population census data.
The next two articles contribute to survival analysis. Maller, Resnick, and Shemehsavar [2] address the important issue of whether there is sufficient follow-up to detect the presence of immune or cured individuals. To test the hypothesis of insufficient follow-up, they find the exact finite and large-sample distribution of the statistic previously proposed by Maller and Zhou. To model multivariate survival data, He, Yi, and Yuan [3] use copulas to join marginal semiparametric linear transformation models. They propose step-wise likelihood inference techniques and investigate the impact of model misspecification on the covariate effect estimates.
Pang, Liu, Zhao, and Zhou [4] consider longitudinal data where the responses depend not only on the past observation history but also on the terminal event. They propose a quantile regression model and develop a non-smoothing estimation equation approach for inference.
The multivariate normal distribution is often too simplistic in practice. Fishbone and Mili [5] consider the wider class of elliptical distributions and propose novel, tunable estimators for location and dispersion. Their proposal is robust to outliers, more stable to initial conditions, and able to achieve higher efficiency than existing high-breakdown estimators.
Stringer [6] investigates optimal identifiability constraints in generalized additive models. As he shows, optimality of the popular centring constraints depends on the response distribution and the parametrisation; for example, in natural exponential families with canonical parametrisation, centring is optimal only for the Gaussian response.
Measurement errors and misclassified data abound in some studies. Spicker, Wallace, and Yi [7] propose an extension of simulation extrapolation which avoids any specific distributional assumptions on the measurement error. Their technique provides reliable results in many settings, such as when the widely used normal additive measurement error model is not appropriate. To estimate parameters in measurement error models, Wang, Wang, and Wang [8] combine the Bayesian and the instrumental variable approach. They obtain an expression for the linear Bayes estimator and show its superiority to the two-stage least squares estimator under the mean squared error matrix criterion. Observational databases used in biomedical research can also be prone to errors. Lotspeich, Amorim, Shaw, Tao, and Shepherd [9] consider a cost-effective two-stage design to data validation that relies on a subset of records. They propose an optimal design for likelihood-based estimators in the setting of binary outcome and exposure misclassification. This optimal design is then located via an adaptive grid search algorithm and its unknown parameters are approximated through a multi-wave strategy.
Chen, Yuan, and Qin [10] contribute to causal inference. To guard against bias when the propensity score and outcome regression models are misspecified, they propose a calibrated version of the augmented inverse weighting estimator of the marginal mean. They show that it can better control extreme influence from these models and is robust to model misspecification.
Liu, Liu, Li, and Lin [11] also tackle a challenging estimation problem, this time in spatial autoregressive models. They propose an estimating equation approach, which is less computationally demanding than maximum likelihood and does not require significant exogenous covariates, unlike the generalized method of moments or spatial two-stage least squares.
Raïssi [12] studies serial correlation in financial returns, allowing for unconditional heteroscedascity and time-varying probabilities of zero returns. The author makes a parallel between the zeros and missing values, and relies on methods for time series to handle the latter.
The closing pair of articles targets big data issues and statistical learning. Xie, Ding, Jiang, Yan, and Kong [13] combine quantile regression and sequential model averaging to achieve robust prediction in ultra-high dimensional data. A sequential approach makes their method computationally feasible. Plante, Larocque, and Adès [14] consider feature selection in supervised learning. Encoding the model as genetic material allows them to use genetic algorithms. The authors derive an objective threshold for feature selection that relies on the null distribution of the importance scores and introduce an eradication strategy akin to forward selection.
Wishing you inspirational readings and a wonderful summer!
Johanna G. Nešlehová, Editor-in-Chief
The Canadian Journal of Statistics
Table of Contents of the June 2024 Issue of The Canadian Journal of Statistics
[1] Smoothed model-assisted small area estimation of proportions, by/par Peter A. Gao & Jon Wakefield
[2] Finite sample and asymptotic distributions of a statistic for sufficient follow-up in cure models, by/par Ross Maller, Sidney Resnick & Soudabeh Shemehsavar
[3] Analysis of multivariate survival data under semiparametric copula models, by/par Wenqing He, Grace Y. Yi & Ao Yuan
[4] Joint modelling of quantile regression for longitudinal data with information observation times and a terminal event, by/par Weicai Pang, Yutao Liu, Xingqiu Zhao & Yong Zhou
[5] New highly efficient high-breakdown estimator of multivariate scatter and location for elliptical distributions, by/par Justin Fishbone & Lamine Mili
[6] Identifiability constraints in generalized additive models, by/par Alex Stringer
[7] Nonparametric simulation extrapolation for measurement-error models, by/par Dylan Spicker, Michael P. Wallace & Grace Y. Yi
[8] Bayesian instrumental variable estimation in linear measurement error models, by/par Qi Wang, Lichun Wang & Liqun Wang
[9] Optimal multi-wave validation of secondary use data with outcome and exposure misclassification, by/par Sarah C. Lotspeich, Gustavo G. C. Amorim, Pamela A. Shaw, Ran Tao & Bryan E. Shepherd
[10] A calibration method to stabilize estimation with missing data, by/par Baojiang Chen, Ao Yuan & Jing Qin
[11] A combined moment equation approach for spatial autoregressive models, by/par Jiaxin Liu, Hongliang Liu, Yi Li & Huazhen Lin
[12] On the correlation analysis of stocks with zero returns, by/par Hamdi Raïssi
[13] High-dimensional model averaging for quantile regression, by/par Jinhan Xie, Xianwen Ding, Bei Jiang, Xiaodong Yan & Linglong Kong
[14] Objective model selection with parallel genetic algorithms using an eradication strategy, by/par Jean-François Plante, Maxime Larocque & Michel Adès