The Three Types Of Economists

Using Machine Learning To Cluster Like-Minded Economists

Policymakers often turn to economists to answer all sorts of questions: appointed economists hold broad regulatory power over the US monetary system that impacts not only the US economy but the entire international trade system, presidents since Harry Truman have convened a council of economists to advise them on the economic ramifications of policies, and economic analysis tools like cost-benefit analysis have been a required component of the federal regulatory process since the early days of the Reagan administration.

Economics is the study of decision-making under conditions of scarcity, a condition that describes every public policy problem in one way or another. Economics strives to be a science of resources allocation in which theoretical models of how actors respond to scarcity are subject to rigorous empirical analysis.

The field’s utility to policymakers hinges on economics functioning as a scientific process of inquiry. If individual biases among economists lead to different findings, they start to look more like a politician than a scientist - and typically the last thing policymakers need to solve their problems is more politicians. Similarly, if there are “schools of thought” among economists, then policymakers can choose bodies of evidence from economists that align with their values or, more cynically, confirm their own bias. Thus, the degree to which expert opinion among economists converges is important to the policymaking process and the fault lines along which they diverge are equally important.

So do economists as a whole come with an adequate number of varying opinions and perspectives, or do they mostly fall along the left-right divide we see with politicians? We can assess the range of opinions in the economic community using survey data provided by the University of Chicago Booth School of Business. Booth has assembled a panel of economic experts - aptly dubbed the IGM Economic Experts Panel - and have regularly been asking the panelists about their opinions on timely public policy issues. In total, the panel gives us 25 economists’ opinions on 63 questions.

We’re not the first people to study the Economic Experts Panel. As a matter of fact, two different teams of researches have looked at this same dataset and come to opposite conclusions. One 2013 study could not find any partisan bias, though they did find evidence that economists who were older, male, and who worked for elected officials tended to exhibit more certainty in their answers than those who were younger, female, and more academic. On the other hand, a 2018 study found that economists who used phrases associated with a certain ideological persuasion were more likely to endorse policy positions that aligned with that ideology.

In order to adjudicate between these competing conclusions, we decided to analyze the Economic Experts Panel data using a hierarchical clustering method, which automatically sorted the economists into three clusters based on how often they agree with one another.¹For any question, panelists have the following options: strongly disagree, disagree, unsure, agree, strongly agree, no opinion. We combined strongly disagree with disagree, strongly agree with agree, and no opinion with unsure.

The dendrogram chart below gives a more granular view of the clusters, going as far as showcasing how closely aligned specific panelists within a cluster are. For example, Edlin and Scheinkman are the most likeminded of all the economists in cluster 1; Fair and Thaler are the most aligned of all the possible pairs in cluster 2. The dendrogram also reveals that cluster 1 is the most separated - meaning cluster 2 and 3 may agree on some policy areas not discussed above (whether firm’s combined market share equates to market power, for example).

Finally, you may have noticed cluster 2 is significantly larger than the other two clusters - in fact, the lot is large enough to host significant subsets: although the 13 economists agree with each other enough to differentiate themselves from clusters 1 and 3, we witness a fissure within the cluster (likely caused by disagreement over less substantive issues, such as the effect of colleges admitting children of alumni and donors).

Overall, the data don’t reveal any immediately obvious patterns. The biggest takeaway is that these panelists don’t seem to fall along liberal-conservative ideological lines. While we guarantee there are liberal and conservative economists who sharply disagree over highly politicized issues, there are dozens of policy questions without defined sides, giving any two economists plenty of opportunities to disagree with each other one day and agree the next. In fact, even the most closely aligned panelists only agree with each other about half the time. In short, the lack of patterns suggests leading economists display a good deal of independence from one another in their opinions - there’s a reason we needed a machine to cluster the panel into like-minded groups for us: the task was beyond any human.

Extending past partisan ideologies and considering other popular dichotomies in economics, the panelists still don’t fall along any obvious ideological spectrum; we did not find evidence of “neoclassical economists” and “Keynesian economists.” Perhaps two of the panelists will agree with the Keynesian school of thought, but that fact alone doesn’t suggest they would agree on other questions. By contrast, given knowledge of just two or three of a politician’s policy stances, we could most likely predict their political party and their positions on any most other issues. For the time being, economists do not operate this way.

Code and data for this project available on Github.