Summary
I propose a cutting-edge and transformative paradigm for statistical modelling that is crucial to enhance the quality of data analyses. Leveraging my expertise in causal inference and semiparametric statistics, I will establish the fundamental principles of a comprehensive estimation theory, which maps model parameters onto generic, interpretable, model-free estimands (e.g., association or effect measures) with favourable efficiency bound, and harnesses the power of debiased (statistical/machine) learning techniques to estimate these. My core objective is to develop a flexible and accessible data modelling framework, called ‘assumption-lean modelling’. This framework will deliver minimal bias and maximal interpretability, even in the presence of model misspecification, along with honest confidence bounds that account for model uncertainty.
Debiased learning is at the core of this research. While gaining popularity, a rigorous scientific optimality theory is lacking. I shall draw on my expertise in (bias-reduced) double robust estimation to develop optimal debiased learning estimators. These utilize learners that optimize strategically chosen loss functions to achieve low variance and high stability, along with confidence intervals that are valid under weak conditions on the learners.
I will connect to timely, exciting developments in statistics, such as debiased learning of function-valued parameters and the construction of confidence bounds for such parameters. I will offer novel avenues into these problems by incorporating the assumption-lean modelling principles and connecting to real-world needs.
I will develop assumption-lean modelling strategies to tackle significant challenges in causal modelling, including target trial emulation, causal mediation analysis, and statistical modelling of dependent outcomes. I will deliver methods with potential impact on all empirical sciences, as well as on the foundations of the discipline of statistical modelling.
Debiased learning is at the core of this research. While gaining popularity, a rigorous scientific optimality theory is lacking. I shall draw on my expertise in (bias-reduced) double robust estimation to develop optimal debiased learning estimators. These utilize learners that optimize strategically chosen loss functions to achieve low variance and high stability, along with confidence intervals that are valid under weak conditions on the learners.
I will connect to timely, exciting developments in statistics, such as debiased learning of function-valued parameters and the construction of confidence bounds for such parameters. I will offer novel avenues into these problems by incorporating the assumption-lean modelling principles and connecting to real-world needs.
I will develop assumption-lean modelling strategies to tackle significant challenges in causal modelling, including target trial emulation, causal mediation analysis, and statistical modelling of dependent outcomes. I will deliver methods with potential impact on all empirical sciences, as well as on the foundations of the discipline of statistical modelling.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101141305 |
| Start date: | 01-10-2024 |
| End date: | 30-09-2029 |
| Total budget - Public funding: | 2 445 063,00 Euro - 2 445 063,00 Euro |
Cordis data
Original description
I propose a cutting-edge and transformative paradigm for statistical modelling that is crucial to enhance the quality of data analyses. Leveraging my expertise in causal inference and semiparametric statistics, I will establish the fundamental principles of a comprehensive estimation theory, which maps model parameters onto generic, interpretable, model-free estimands (e.g., association or effect measures) with favourable efficiency bound, and harnesses the power of debiased (statistical/machine) learning techniques to estimate these. My core objective is to develop a flexible and accessible data modelling framework, called ‘assumption-lean modelling’. This framework will deliver minimal bias and maximal interpretability, even in the presence of model misspecification, along with honest confidence bounds that account for model uncertainty.Debiased learning is at the core of this research. While gaining popularity, a rigorous scientific optimality theory is lacking. I shall draw on my expertise in (bias-reduced) double robust estimation to develop optimal debiased learning estimators. These utilize learners that optimize strategically chosen loss functions to achieve low variance and high stability, along with confidence intervals that are valid under weak conditions on the learners.
I will connect to timely, exciting developments in statistics, such as debiased learning of function-valued parameters and the construction of confidence bounds for such parameters. I will offer novel avenues into these problems by incorporating the assumption-lean modelling principles and connecting to real-world needs.
I will develop assumption-lean modelling strategies to tackle significant challenges in causal modelling, including target trial emulation, causal mediation analysis, and statistical modelling of dependent outcomes. I will deliver methods with potential impact on all empirical sciences, as well as on the foundations of the discipline of statistical modelling.
Status
SIGNEDCall topic
ERC-2023-ADGUpdate Date
26-11-2024
Images
No images available.
Geographical location(s)
