Susan Athey on Machine Learning, Big Data, and Causation

Predictive Modeling

Machine learning's role in predictive modeling within economics is both promising and limited. Susan Athey explains that while these models can predict outcomes like employment changes due to minimum wage adjustments, they should not be mistaken for causal interpretations 1. Predictive models excel at fitting data, but their reliability can falter when environmental conditions change. Athey highlights the importance of interpretable models, which are crucial for understanding the underlying mechanisms and ensuring predictions hold across different settings 2.

Predictive models shouldn't be given causal interpretations. But you tell the machine, find me the best predictive model.

--- Susan Athey

This distinction is vital for economists who seek to understand the causal relationships in economic data.

   

Big Data Prediction

Big data and machine learning offer new avenues for handling high-dimensional datasets in economics. Susan Athey discusses how these techniques allow economists to control for numerous variables simultaneously, enhancing the accuracy of predictions 3. By leveraging micro-level data, researchers can gain insights into specific economic phenomena, such as the impact of a recession on a particular industry or locality 4.

Modeling things at the micro level rather than the macro level can be incredibly powerful.

--- Susan Athey

This approach provides a more nuanced understanding of economic dynamics, although it requires careful consideration of the data's scope and limitations.

   

Causal Inference

Machine learning's potential in causal inference is significant, yet it faces challenges. Susan Athey notes that while tech firms excel in prediction, they often overlook causal inference, a gap that economists aim to bridge 5. Collaborations between social scientists and tech companies have yielded insights into the causal effects of technology on behavior, such as voting patterns influenced by social media 6.

The machine learning community has systematically ignored causal inference. They've gotten awesome at prediction.

--- Susan Athey

By integrating machine learning with traditional econometric methods, researchers can better understand the causal impacts of policies and innovations.