Daniel Jeske, Editor, The American Statistician

The February 2020 issue of The American Statistician is available online and features 14 articles and one letter to the editor. Recall that one of the benefits of ASA membership is free access to the online issues of TAS.

The General section has three articles. The first examines the asymptotic behavior of the Bayes factor for comparing two models. Considerations to dependent data and mis-specified models are included. The results provide a practical extension to the use of Bayes factors. The second article provides an easy-to-use algorithm for generating correlation matrices that have specified eigenvalues. Applications for the algorithm are discussed. The third article proposes two methods for constructing confidence intervals for a bivariate correlation coefficient when the underlying joint distribution is unknown. The tools used are generalized pivot quantities and empirical likelihood.

The Statistical Practice section has two articles. The first is concerned with the problem of binary responses in spatial modeling. Properties of, and relationships between, various models are explored. The second paper offers a detailed discussion on two-tailed p-values, with particular attention to situations in which the null distribution is asymmetric. A modified p-value is introduced that purports to provide a measure of evidence for a null hypothesis.

There are three articles in the Teacher’s Corner. The first discusses an R Shiny app that implements a game for teaching response surface modeling. Detailed suggestions as to how to integrate the app into classrooms are provided. The second paper provides multiple examples that help to understand when the distributions of X/(X+Y) and Y/(X+Y) are identical. The examples have pedagogical value for some concepts taught in mathematical statistics classes. The third article revisits the conditionality principle. The scaled uniform distribution provides an interesting illustrative example.

The Data Science section has two articles. The first describes a data science program offered at The Johns Hopkins University. The program consists of nine four-week courses and has had more than 4 million initial enrollments over the past three years. The second article offers suggestions for teaching data science at the undergraduate level. The approach taken uses a rich variety of computational tools.

An article in the Interdisciplinary section looks into ordinal item response models. The impact of a particular constraint in the parameter space is investigated and a case study with customer ratings data is presented.

An article in the History Corner presents graphical depictions and interpretations of Norman L. Johnson’s proposed (1949) families of transformation functions.

The Short Technical Note section has two articles. The first shows that least squares estimators have a certain type of loss function robustness by demonstrating they are optimal with respect to a general divergence loss function. The second article contributes to an ongoing discussion on how standardization affects collinearity diagnostics, with a particular focus on variance inflation factors.

The February issue concludes with a Letter to the Editor, which investigates the sharpness of an upper bound on Bayes factors.