This thesis investigates likelihood-based inference in meta-analysis risk regression models, with particular emphasis on the behavior of the likelihood ratio test in small-sample settings. In such contexts, the asymptotic approximation of the test statistic distribution may be inaccurate, motivating the consideration of alternative inferential approaches. The analysis is conducted within a likelihood framework for risk regression in meta-analysis, which models the relationship between risk and study-level covariates while accounting for the uncertainty associated with the observed estimates. Inference on parameters of interest is based on the likelihood ratio test, both in its standard form and in a Bartlett-type corrected version obtained through parametric bootstrap under the null hypothesis. Through a Monte Carlo simulation study, the thesis evaluates the performance of the different testing procedures in terms of level accuracy and power, as sample size and model characteristics vary. The aim is to assess whether, and under which conditions, the Bartlett correction can improve upon the traditional asymptotic approach within the framework of meta-analysis risk regression.
Questa tesi analizza l’inferenza basata sulla verosimiglianza in modelli di meta-analysis risk regression, con particolare attenzione al comportamento del test del rapporto di verosimiglianza in campioni di dimensione ridotta. In tali contesti, l’approssimazione asintotica della distribuzione della statistica di test può risultare imprecisa, rendendo necessaria la valutazione di metodologie alternative. L’analisi è condotta all’interno di un modello di verosimiglianza per la risk regression in meta-analisi, che descrive la relazione tra il rischio e le covariate a livello di studio, tenendo conto dell’incertezza associata alle stime osservate. L’inferenza sui parametri di interesse è basata sul test del rapporto di verosimiglianza, sia nella sua forma standard sia in una versione corretta mediante una correzione di tipo Bartlett ottenuta tramite bootstrap parametrico sotto l’ipotesi nulla. Attraverso uno studio di simulazione Monte Carlo, la tesi valuta il comportamento delle diverse procedure di test in termini di accuratezza del livello e potenza, al variare della dimensione campionaria e delle caratteristiche del modello. L’obiettivo è comprendere se e in quali condizioni la correzione di Bartlett possa fornire un miglioramento rispetto all’approccio asintotico tradizionale nell’ambito della meta-analysis risk regression.
Approccio di verosimiglianza e correzione di Bartlett nella meta-analisi con rischio di base
BENETELLO, GIORGIA
2025/2026
Abstract
This thesis investigates likelihood-based inference in meta-analysis risk regression models, with particular emphasis on the behavior of the likelihood ratio test in small-sample settings. In such contexts, the asymptotic approximation of the test statistic distribution may be inaccurate, motivating the consideration of alternative inferential approaches. The analysis is conducted within a likelihood framework for risk regression in meta-analysis, which models the relationship between risk and study-level covariates while accounting for the uncertainty associated with the observed estimates. Inference on parameters of interest is based on the likelihood ratio test, both in its standard form and in a Bartlett-type corrected version obtained through parametric bootstrap under the null hypothesis. Through a Monte Carlo simulation study, the thesis evaluates the performance of the different testing procedures in terms of level accuracy and power, as sample size and model characteristics vary. The aim is to assess whether, and under which conditions, the Bartlett correction can improve upon the traditional asymptotic approach within the framework of meta-analysis risk regression.| File | Dimensione | Formato | |
|---|---|---|---|
|
Benetello_Giorgia.pdf
Accesso riservato
Dimensione
2.67 MB
Formato
Adobe PDF
|
2.67 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/105762