Targeted Maximum Likelihood Estimation for Causal Inference With Observational Data-The Example of Private Tutoring.

Area of Science:

• Causal Inference
• Observational Data Analysis
• Educational Research Methodology

Background:

• Traditional causal inference methods for observational data rely on strong assumptions, risking misspecification bias.
• Advanced techniques like Targeted Maximum Likelihood Estimation (TMLE) aim to improve robustness and efficiency.
• Machine learning, including super learning, can enhance the estimation of data distribution components in causal models.

Purpose of the Study:

• To introduce Targeted Maximum Likelihood Estimation (TMLE) as a robust causal inference method.
• To estimate the causal effect of private mathematics tutoring in Year 7 on student outcomes using observational data.
• To compare TMLE estimates with those from Ordinary Least Squares, the parametric G-formula, and augmented inverse-probability weighting.

Main Methods:

• Utilized Targeted Maximum Likelihood Estimation (TMLE), a doubly robust, semiparametric substitution estimator.
• Employed super learning (a machine learning ensemble method) to estimate outcome and treatment models nonparametrically.
• Analyzed observational data from the National Education Panel Study (starting cohort 3, N=4,167) on mathematics tutoring effects.

Main Results:

• Close agreement was observed between TMLE and other methods (OLS, G-formula, AIPW) for end-of-year mathematics grades.
• Significant variations in estimates emerged when mathematics proficiency was the outcome, indicating sensitivity to analytical approach.
• The choice of causal inference methodology influenced the substantive conclusions drawn regarding the impact of private tutoring.

Conclusions:

• Advanced causal inference methods like TMLE are crucial for addressing complexities in observational data analysis.
• Methodological choices in causal inference can substantially affect the interpretation of research findings in education.
• The study highlights the importance of employing robust statistical techniques to ensure valid causal claims from observational studies.