This thesis deals with causal inference, with a particular focus on estimating the average treatment effect (ATE) in the context of both complete data and survival data. After introducing the concept of potential outcomes and the main assumptions required to identify causal effects, two approaches based on M-estimation are presented: Q-models and IPTW. The thesis then extends these methods to the setting of survival data, where the issue of right-censoring arises. Some causal quantities of interest are introduced, and additional assumptions needed for identification in this context are discussed. The final part of the work is dedicated to an application on real data.
In questa tesi viene affrontato lo studio dell’inferenza causale, con particolare attenzione alla stima dell’effetto medio del trattamento (ATE) in presenza di dati completi e di sopravvivenza. Dopo aver introdotto il concetto di outcome potenziali e le principali assunzioni necessarie per identificare effetti causali, vengono analizzati due approcci noti come M-estimation: i Q-models e l’IPTW. La tesi prosegue estendendo questi metodi al contesto dei dati di sopravvivenza, in cui si presenta il problema della censura a destra. Vengono introdotte alcune quantità causali d’interesse e discusse le assunzioni aggiuntive necessarie per garantire l’identificabilità in questo contesto. L'ultima parte del lavoro è dedicata a un’applicazione su dati reali.
Modelli causali per dati completi e di sopravvivenza
BIZZOTTO, MATTEO
2024/2025
Abstract
This thesis deals with causal inference, with a particular focus on estimating the average treatment effect (ATE) in the context of both complete data and survival data. After introducing the concept of potential outcomes and the main assumptions required to identify causal effects, two approaches based on M-estimation are presented: Q-models and IPTW. The thesis then extends these methods to the setting of survival data, where the issue of right-censoring arises. Some causal quantities of interest are introduced, and additional assumptions needed for identification in this context are discussed. The final part of the work is dedicated to an application on real data.| File | Dimensione | Formato | |
|---|---|---|---|
|
Bizzotto_Matteo.pdf
accesso aperto
Dimensione
378.73 kB
Formato
Adobe PDF
|
378.73 kB | Adobe PDF | Visualizza/Apri |
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/20.500.12608/92930