The University of Chicago is well-known for its rich economics program, producing rigorous academic research and six current Nobel Laureates (#notsohumblebrag). However, it’s not all just academia here. Steven Levitt, of Freakonomics fame, has been a professor at the University of Chicago since 1997. Levitt has arguably been the most publicly influential economist in the last 20 years through his best-selling popular economics literature, blog, and podcast.
A few weeks ago, through the Becker Brown Bag series, Levitt discussed a Booth student’s favorite topic—data—and how modern data science has introduced new approaches to data not traditionally seen in economics. Furthermore, Levitt introduced to the audience his next big project with ambitions to tangibly change the world.
Every conversation about data starts with correlation (if yours don’t, they should). Levitt asserted that correlation is the “single most important element of data,” yet it’s also not very useful. It does not tell us why. The economics discipline has historically been focused on studying the why through natural or randomized experiments and empirical studies. As such, economics is inherently backwards looking, illuminating potential causal relationships through observations of what actually happened in the past.
This traditional economic approach to data produces economic models and theories on how people should behave, but does not do a great job of actually predicting how people will behave.
Modern data science turns this approach to data on its head. It is “theory free,” focused almost exclusively on correlation and pattern recognition. The why is less important, what matters is only that the patterns and relationships exist. Because of this, many people see modern data science as a “black box.” Interestingly, despite not focusing on the theory and the why, Levitt showed that modern data science models have empirically proven to be more effective in prediction than economic models.
So what caused this divergence in approach to data, and is one inherently better than the other? This is ultimately what Levitt sought to address.
It comes down to the differing purposes of economics and modern data science. Traditional economics seeks to explain why things happened in the past, while modern data science seeks to predict the future. In a static environment (i.e., things are more or less the same from today to tomorrow), causality is extraneous, so the correlation and pattern recognition of modern data science often prevails over theory. However, in environments with limited data, economic theory can be useful in deriving insights.
Furthermore, Levitt argues that the most impactful insights can be derived from combining both approaches. He praised data scientists for promoting the idea that everything is data—words, images, etc. Additionally, modern AI can help economics discover natural experiments or hypotheses worth testing that a human may miss.
On the other hand, in a highly dynamic setting where change may occur faster than data science models can predict, the role of economic theory can be crucial. In fact, his research showed that using a weighted average of data science models and economic models produced better predictions than any single approach. Hence, big data and economic ideas are complements, not substitutes, and their combination can result in powerful insights.
This combination is the foundation of Levitt’s latest new venture. In a refreshing and humbling review of his past work, Levitt closed his lecture by admitting he hadn’t made the tangible impact on policy or societal behavior that he had hoped. In his next project, he’s challenging himself to tangibly change the world through the combination of economic ideas and modern data science, and welcomed interested and passionate audience members to join him on this journey.
Upon reflection, this lecture really was a microcosm of Booth and the University of Chicago’s culture and mission—deep inquiry using rigorous data to impact the world. Who wouldn’t want to be part of that mission?