Associate Professor in Statistics

Yves Berger's research focuses on foundations of statistical inference from complex sample surveys. Yves Berger established a world-class research related to various fundamental issues of statistical inference from complex sample surveys. These issues include: variance estimation; repeated surveys; non-response; imputation; inference for complex parameters (empirical likelihood). His publications include articles in the Journals of the Royal Statistical Society (Series A, B and C), Biometrika and the Canadian Journal of Statistics.
 
Yves Berger teaches a wide range of courses at undergraduate and postgraduate levels: statistical modelling, generalised linear models, sample survey theory, non-response adjustment, multivariate analysis, multilevel models, longitudinal data and repeated measures, statistical computing, computer intensive statistical methods, statistical consulting, communication and research skills and demography.

Yves Berger is an Associate Editor of Computational Statistics & Data Analysis.

"An empirical likelihood approach under cluster sampling with missing observations"

Posted on 25th June 2018


Accepted for publication in the Annals of the Institute of Mathematical Statistics.

Follow this link to download the paper

Review paper: "Empirical likelihood approaches under complex sampling designs" (Statsref 2018)

Posted on 6th November 2017


This is a short review paper which compares empirical likelihood with pseudoempirical likelihood.

Follow this link to download the paper

"modelling complex survey data with population level information: an empirical likelihood approach" (Biometrika 2016)

Posted on 24th July 2017


Survey data are often collected with unequal probabilities from a stratified population. In many modelling situations, the parameter of interest is a subset of a set of parameters, with the others treated as nuisance parameters. We show that in this situation the empirical likelihood ratio statistic follows a chi-squared distribution asymptotically, under stratified single and multi-stage unequal probability sampling, with negligible sampling fractions. Simulation studies show that the empirical likelihood confidence interval may achieve better coverages and has more balanced tail error rates than standard approaches involving variance estimation, linearization or re-sampling.

Follow this link to download the paper