Using LLMs to Directly Guess Conditional Expectations Can Improve Efficiency in Causal Estimation
This work addresses efficiency issues in causal inference for researchers and practitioners, but it is incremental as it builds on existing methods by incorporating LLM predictions.
The paper tackles the problem of improving causal estimation efficiency in double machine learning by using LLM-generated predictions as direct guesses for conditional expectations, demonstrating improved efficiency in a case study with online jewelry auctions.
We propose a simple yet effective use of LLM-powered AI tools to improve causal estimation. In double machine learning, the accuracy of causal estimates of the effect of a treatment on an outcome in the presence of a high-dimensional confounder depends on the performance of estimators of conditional expectation functions. We show that predictions made by generative models trained on historical data can be used to improve the performance of these estimators relative to approaches that solely rely on adjusting for embeddings extracted from these models. We argue that the historical knowledge and reasoning capacities associated with these generative models can help overcome curse-of-dimensionality problems in causal inference problems. We consider a case study using a small dataset of online jewelry auctions, and demonstrate that inclusion of LLM-generated guesses as predictors can improve efficiency in estimation.