Case Studies In Business, Industry And Government Statistics

Editor's foreword

Christine Thomas-Agnan — 2021-05-20

Editor's foreword

A case study of non-inferiority testing with survival outcomes

Hsin-wen Chang — 2021-04-29

This is a case study for a new class of nonparametric tests designed to assess evidence of non-inferiority ordering among multiple survival functions. The tests are devised for tree-structured orderings, as needed for the comparison of an experimental treatment to one or more alternative treatments. Applications to data from two non-inferiority trials are developed: 1) a two-armed trial for the treatment of liver cancer in which we find strong evidence of the non-inferiority of an experimental treatment (lenvatinib) to a standard treatment (sorafenib), and 2) a three-armed trial for the treatment of major depression in which we find strong evidence that an experimental treatment is both superior to placebo and non-inferior to a standard treatment. We implement the approach in R, and explain in detail how to carry out the analyses for 1) and 2).

Digit analysis for Covid-19 reported data

Jean-François Coeurjolly — 2021-04-29

Thee coronavirus which appeared in December 2019 in Wuhan has spread out worldwide and caused the death of more than 330,000 people (as of May 29, 2020, submission date for the present article). Since February 2020, doubts were raised about the numbers of confirmed cases and deaths reported by the Chinese government. In this paper, we examine data available from China at the city and provincial levels and we compare them with Canadian provincial data, US state data and French regional data. We consider cumulative and daily numbers of confirmed cases and deaths and examine these numbers through the lens of their first two digits and in particular we measure departures of these first two digits to the Newcomb-Benford distribution, often used to detect frauds. Our finding is that there is no evidence that cumulative and daily numbers of confirmed cases and deaths for all these countries have different first or second digit distributions. We also show that the Newcomb-Benford distribution cannot be rejected for these data.

Modelling repeated paired phoneticmeasures using linear mixed models withcorrelated errors

Marie-José Martinez — 2021-04-29

In Phonetic Sciences, statistical analysis from experimental data have to be carried out to confirm or disconfirm hypotheses. In this paper, a phonetic data set is considered and phonetic research questions are addressed. To answer these questions, a mixed model is built using a complex random effects structure and a non-diagonal residual variance-covariance matrix. Then, it is validated on the data. Finally, we focus on statistical tests in the final model allowing to compare the means between two groups of subjects, and a single mean to a reference value. The paper is accessible to an audience experienced with linear models. Some familiarity with the R software is also helpful.

Exploring the distribution of conditional quantiles estimation ranges: an application to specific costs of pig production in the European Union

Dominique Desbois — 2021-04-29

This communication uses symbolic data analysis tools to visualize conditional quantile estimation intervals, applying it to the problem of cost allocation in agriculture. After recalling the conceptual framework of the estimation of agricultural production costs, the first part presents the empirical model, the quantile regression approach and the interval data processing techniques used as symbolic data analysis tools. The second part presents the comparative analysis of the econometric results between twelve European Member States, using the principal components analysis and the hierarchical grouping of the estimation intervals, by discussing the relevance of the exploratory graphs obtained for the international comparisons.