Targeted maximum likelihood estimation for psychological research: From causal identification to statistical...

Area of Science:

• Psychology
• Statistics
• Causal Inference

Background:

• Causal questions are fundamental in psychology, traditionally addressed via randomized experiments.
• Observational and quasi-experimental data increasingly require formal causal inference methods.
• Existing methods like outcome regression and propensity scores are vulnerable to model misspecification.

Purpose of the Study:

• To provide practical guidance and reproducible code for implementing Targeted Maximum Likelihood Estimation (TMLE) in psychological research.
• To integrate causal-to-statistical parameter mapping and semiparametric explanations for applied researchers.
• To address fragmented guidance on modern causal inference implementation using TMLE.

Main Methods:

• Utilized Targeted Maximum Likelihood Estimation (TMLE) for doubly robust estimation and influence-function-based inference.
• Integrated causal-to-statistical parameter mapping, positivity diagnostics, and fully nested cross-fitting.
• Developed reproducible Python code and provided TMLE-specific reporting guidance.
• Employed simulations to assess TMLE performance under various conditions, including nuisance model misspecification and limited overlap.

Main Results:

• TMLE effectively combines outcome and treatment information for valid target-parameter estimation and inference.
• Cross-fitted Super Learner TMLE demonstrated robust performance compared to transparent parametric estimation.
• Simulations highlighted the impact of nuisance model misspecification and limited overlap on estimation accuracy.

Conclusions:

• TMLE serves as a crucial estimation stage within a broader causal inference workflow for psychological research.
• The study offers translational contributions, making advanced causal inference methods more accessible to applied researchers.
• Effective implementation of TMLE requires careful consideration of causal models, identification assumptions, and reporting standards.