Survey sampling methods applied to experiments
Author(s)
Editor(s)
Publisher
Neuchâtel : Université de Neuchâtel
Date issued
2025
Number of pages
92 p.
Subjects
Balanced sampling Design of experiments Design variance matrix Exact inference Mixture of designs Echantillonnage équilibré Inférence exacte Matrice de variance du plan Plan d’expériences
Abstract
Cette thèse étudie différents aspects de la conception des enquêtes et des expériences randomisées. Ces deux sujets présentent des similitudes et des différences particulières qui nous permettent d’établir des liens entre leurs théories. Une partie importante de cette thèse porte sur les compromis associés à l’équilibrage des covariables entre les groupes dans le but d’améliorer l’efficacité des estimateurs. Nous adaptons également des méthodes d´échantillonnage à des contextes expérimentaux. Nous fournissons des solutions spécifiques à des problèmes tels que l’affectation des participants à plusieurs groupes de traitement où les moyennes des covariables sont équilibrées entre les groupes, même avec des groupes de taille inégale ou des probabilités d’affectation à un groupe. Nous proposons également une méthode permettant d’effectuer des tests d’hypothèse exacts pour des modèles tels que la régression de Poisson, ce qui réduit la dépendance à l’égard des approximations sur de grands échantillons.
ABSTRACT
This thesis studies different aspects of the design of surveys and randomized experiments. Those two subjects have particular similarities and differences that allow us to establish links between their theories. An important part of this thesis is the trade-offs that occur when balancing covariates across groups to improve the efficiency of the estimators. We also adapt methods from survey sampling to experimental settings. We provide specific solutions to problems such as assigning participants to multiple treatment groups where the covariate means are balanced between groups even, with groups of unequal size, or assignment to group probabilities. We also propose a method to perform exact hypothesis tests for models like Poisson regression, which reduce reliance on large-sample approximations.
ABSTRACT
This thesis studies different aspects of the design of surveys and randomized experiments. Those two subjects have particular similarities and differences that allow us to establish links between their theories. An important part of this thesis is the trade-offs that occur when balancing covariates across groups to improve the efficiency of the estimators. We also adapt methods from survey sampling to experimental settings. We provide specific solutions to problems such as assigning participants to multiple treatment groups where the covariate means are balanced between groups even, with groups of unequal size, or assignment to group probabilities. We also propose a method to perform exact hypothesis tests for models like Poisson regression, which reduce reliance on large-sample approximations.
Notes
The dissertation Committee:
Prof. Yves Tillé, Thesis director University of Neuchâtel
Prof. Alina Matei, President of jury University of Neuchâtel
Prof. David Haziza, Examiner University of Ottawa
Prof. Camélia Goga, Examiner University of Bourgogne Franche-Comté
Dr. Caren Hasler, Examiner University of Neuchˆatel and University of Zürich
Thesis defended on 24 June 2025
No de thèse : 3195
Prof. Yves Tillé, Thesis director University of Neuchâtel
Prof. Alina Matei, President of jury University of Neuchâtel
Prof. David Haziza, Examiner University of Ottawa
Prof. Camélia Goga, Examiner University of Bourgogne Franche-Comté
Dr. Caren Hasler, Examiner University of Neuchˆatel and University of Zürich
Thesis defended on 24 June 2025
No de thèse : 3195
Publication type
doctoral thesis
File(s)![Thumbnail Image]()
Loading...
Name
00003195.pdf
Type
Main Article
Size
1.36 MB
Format
Adobe PDF
