ABSTRACT Fay–Herriot (FH) estimators are widely used to produce small‐area statistics when only area‐level aggregate data are available. This paper investigates the conditions under which bivariate FH models give useful reductions in approximate prediction mean squared error (APMSE) compared to separate univariate models. The APMSE is shown to be equal under these two approaches if the sampling errors and the area‐level random effects have proportional variance‐covariance matrices, even if there is high correlation between the two variables of interest. The ratio of APMSEs is calculated numerically for a range of settings, and this numerical study is summarised using a novel regression tree approach. Univariate and bivariate estimators are compared using data on 30 indicators from the 2011–2012 New Zealand Health Survey, with MSEs estimated by a parametric bootstrap approach. The results suggest that bivariate modelling can be worthwhile, but only for a minority of cases.

ABSTRACT The tidyverse is a popular meta‐package comprising several core R packages to aid in various data science tasks, including data import, manipulation and visualisation. Although functionalities offered by the tidyverse can generally be replicated using other packages, its widespread adoption in both teaching and practice indicates there are factors contributing to its preference, despite some debate over its usage. This suggests that particular aspects, such as interface design, may play a significant role in its selection. Examining the interface design can potentially reveal aspects that aid the design process for developers. While Tidyverse has been lauded for adopting a user‐centred design, arguably some elements of the design focus on the work domain instead of the end user. We examine the Tidyverse interface design via the lens of human–computer interaction, with an emphasis on data visualisation and data wrangling, to identify factors that might serve as a model for developers designing their packages. We recommend that developers adopt an iterative design that is informed by user feedback, analysis and complete coverage of the work domain, and ensure perceptual visibility of system constraints and relationships.

ABSTRACT Motivated by the need to estimate prevalence at multiple disaggregated level hierarchies, rather than only one, this study extends widely used area‐level models in Bayesian and frequentist framework. We propose adding additional unstructured random effects at higher level disaggregated domains to the traditional models. Using our extension, we find major benefits for unbiasedness and coverage. The penalty in using additional random effects can be slightly higher standard errors (SEs), but if small, this increase is warranted because it can improve coverage of the model‐based estimator. The proposed model is robust in the sense that it can better account for unexplained variation at the higher aggregation levels compared to traditional spatial and non‐spatial area‐level models. When applied to Australian smoking data, the extended model shows the benefit of including both unstructured random effects at the detailed target levels, that is, statistical areas level 3 and 4 (SA3 and SA4), and structured random effects at the more detailed (SA3) level. Using the extended model that has very strong fixed‐effect components confirms unbiasedness for the targeted domains at both SA3 and SA4 levels.

ABSTRACTVisual assessment of residual plots is a common approach for diagnosing linear models, but it relies on manual evaluation, which does not scale well and can lead to inconsistent decisions across analysts. The lineup protocol, which embeds the observed plot among null plots, can reduce subjectivity but requires even more human effort. In today's data‐driven world, such tasks are well suited for automation. We present a new R package that uses a computer vision model to automate the evaluation of residual plots. An accompanying Shiny application is provided for ease of use. Given a sample of residuals, the model predicts a visual signal strength (VSS) and offers supporting information to help analysts assess model fit.

ABSTRACTThe paper describes six examples of poor statistical practice in public policy. The first example is the lack of a COVID Information Plan for Australia resulting in deficient information being used to understand the progress of the pandemic and the best public policy responses. The second example is inappropriate criteria being used for determining when to ease COVID restrictions as vaccination rates increased because they ignored the impact of uncertainty in the modelling assumptions. The third example is the machine learning algorithms used in Robodebt, which were flawed, used inappropriate data and did not incorporate measures of uncertainty. The fourth example is the opinion polls used in the 2019 Australian election, which got the result wrong because they relied on unrepresentative samples with inadequate weighting adjustments for this deficiency. The fifth example is from the United States where the salaries of teachers (and even their continued employment) were based on the performance of their students using regression models that were inadequate for the purpose. The sixth and more positive example is the use of purchasing power parities to influence two different global debates on poverty reduction and climate change. The paper concludes with suggestions on what the Australian statistical profession should do to address the lack of statistical thinking in many policy areas.

ABSTRACTWe report on the first year of an ongoing pilot of large‐screen interactive whiteboards (IWBs) in small group tutorial and lab classes for a major second‐year mathematical statistics subject, exploring what they may bring to our teaching of both core theoretical principles and computation/software use. In the tutorials, students worked collaboratively in small teams, hand‐writing solutions to conceptual problems, following established previous practice, but a modality well supported by IWBs. However, the labs were completely revamped to focus primarily on open‐ended team exercises, which challenged students to be creative and think laterally together, including team presentations sharing approaches. Whereas our traditional computer labs have students working individually using rows of computers on benches, the new classroom structure with IWBs enabled open‐ended, engaging and collaborative team problem‐solving supported by shared visualisation tools. As assessed by experienced teachers, student interest, engagement, conceptual understanding, presentation skills, self‐insights into their own statistical thinking and team work were all enhanced. Indeed, students entered into a genuine co‐construction of meaning through dialogue with the staff. Key lessons for statistics educators wishing to follow this path are to bring in the university IT early in the process, but without underestimating the academic staff time required to reshape approaches to realise full potential and to ensure good training and support for the tutoring staff.

ABSTRACTConsiderations of collaboration and leadership are relevant in all disciplines but are of particular significance in the statistical and data sciences. No matter how theoretical or practical, all statistical endeavours have roots or motivations in real problems linked with other disciplines and, in turn, often drive endeavours in these disciplines. Statistical problem‐solving has also often been the driver of developments in mathematics and information technologies. The multi‐faceted roles of collaboration in statistics feed into good practice in teaching statistics in many ways, especially in higher education. They also feed into constructs and practicalities of leadership in statistics and hence into both its teaching and learning. In investigating these issues, this chapter discusses that the long‐held advocacy that the teaching of statistics and science of data should reflect its practice applies to far more than active learning of statistical and data investigations. Examples are used to illustrate how good practice in teaching statistics is an integral and professional segment in the totality of the practice of and leadership in statistics.

ABSTRACTWe present a normalising transformation of the Hill estimator to improve the convergence rate in finite‐sample performance. Our proposal for the normalising transformation is based on the higher order asymptotic expansion of the Hill estimator. The transformation is automatic and simple in computation. The resulting transformation theoretically improves the approximation to the standard normal distribution, achieving a lower error rate compared with the variance stabilisation or the Wilson and Hilferty approximation. The numerical results of simulations also align with our theoretical findings.