The study of law and economics has a long history that has contributed much to our understanding of how different regulatory structures impact economic activity. But much of this work has focused on the structure of regulations rather than the actors within the legal systems that enforce the regulation. As one example, the American Economic Review has published just about 100 articles under the law and economics JEL code in the last ten years. Of these, fewer than ten have focused on the behavior of judges (or other actors) within the legal system. The absence is even more stark if one considers data from low- and middle-income countries: only one uses non-U.S. data.
The lack of attention to the economic behavior of actors in the legal system is understandable – it is hard to get data – but belies their importance. Actors in the legal system respond to incentives (and shocks) the same as actors in any of the dozens of other settings economists study. Yet the responses of these actors to these incentives and shocks may be disproportionately important for the practical implementation of the policies, laws, and regulations that economists have devoted substantial time to study.
That’s why I was excited to participate in a workshop at the Becker Friedman Institute a few months ago. The workshop was devoted to changes in the study of actors in legal systems driven by the recent revolution that has occurred in data access. Access to data on the behavior of actors in the legal system has been a major limitation to doing work in this area because data on much of what economists might like to analyze – patterns of judicial decision making for example – either doesn’t exist or has been contained in unstructured (or non-digitized) text that had been hard to analyze.
Advances in both methods for analyzing unstructured text and in the operation of court systems themselves are rapidly ‘revitalizing’ the field of law and political economy. Based on the papers presented at the workshop, this revitalization has three main themes: (1) the increasing availability of digitized court data, (2) the use of machine learning to understand non-quantitative digital data, and (3) the creation of new tools to assess the behavior of legal actors. And one bonus theme, the rapid expansion of this work to non-U.S. settings: roughly half the papers presented used data from low- and middle-income countries.
Digitized court data
The systematic digitization of data from courts has been crucial in the empirical revitalization of this literature. The digitization includes both the text of court cases – for example the Indian Kanoon system – as well as meta-data about the activities of courts. Almost all of the papers presented relied on some kind of newly digitized data but I’d like to highlight three.
The first, by a team of authors and presented by Matthieu Chemin, used newly digitized data on processing times in Kenyan courts to conduct an RCT on the importance of having efficient courts. The authors use the newly digitized data to create an algorithm that identifies delays in processing a case that are due to court – as opposed to other actors – actions. Using this algorithm they provided information on delays to only courts or to courts and the public. They find that information provided to both results in a decline in court driven delays and that has subsequent benefits for the enforcement and use of contracts in the broader community.
The second, by Emma Harrington and Hannah Shaffer, combines survey data on prosecutors with digitized records of their actual cases to assess the role of discrimination in sentencing. As the authors point out, the impact of bias on decision-making depends not only on the bias of the decision-maker, in this case the prosecutor, but also on their beliefs about the bias of previous decision-makers involved in the process, in this case the police. With a survey of prosecutors they show that prosecutors vary widely in their beliefs about police bias, and by linking these surveys to digitized cases that the prosecutors actually handle, that these beliefs have important implications for how they prosecute cases.
The third, also by a large team but presented by Shaoda Wang, uses a large dataset on judges, cases, law firms, and lawyers in China that the authors have compiled and digitized so examine how the practice of ‘revolving door’ judges – those who leave the judiciary to rejoin law firms – influence case outcomes. They find that these judges exert substantial influence in obtaining favorable outcomes for their clients and that this is not only because of superior legal ability. Because they tend to go to the largest law firms, employed by the richest clients, this may exacerbate inequality.
Use of machine learning
The increase in the amount of digitized data on legal actions and outcomes has only been part of the revolution in empirical law and economics research. The ability to systematically analyze large amounts of unstructured text data has also been important. Increasingly capable ML and LLM models have played a major role in these changes.
Daniel Chen’s presentation highlighted some of the potential in this space. His talk encompassed evidence from many papers but here is one example that nicely encapsulates the potential of these tools. One challenge in estimating the impacts of the actions of actors in the legal system (or other aspects of the system) is determining the consequences of a particular case outcome. In the past this has required hiring law students (or other trained readers) to review cases individually and hand label them based on outcomes. Using one such hand labelled dataset the authors created a semi-supervised machine learning model that can classify cases and show that it can successfully classify cases in a much larger, unlabeled data set.
New tools for assessing behavior
The final papers I want to mention both offer new tools to researchers in their evaluation of behavior of actors in the legal process. The first of these, by Jens Ludwig and co-authors, proposes a tool for more formal hypothesis generation among researchers. Prof. Ludwig started his talk by noting that a huge amount of attention – rightly so! – has been devoted to making the empirical methods that we employ more rigorous. But the initial stage of research, hypothesis generation, is still mostly ad hoc and informal. (How many of your own research projects are the result of something you happened to see in the news or an off-hand comment in a seminar?) They offer one attempt to make this process of hypothesis generation more formal (with potential application to fields beyond just law and economics).
The last paper, by Evan Rose and co-authors, offers a new procedure for ranking and grading performance that balances the information content of the ranks with noise reduction in the ranks. They apply their method to questions of discrimination in employment but the method is applicable in a much broader array of settings.
There were more interesting papers presented than I have space to cover fully here – the workshop agenda lists many of those I didn’t discuss. But it is an exciting time to work on empirical questions of law and political economy. The revolution in data availability and processing capability that we are currently going through suggests that this is a field that will continue to grow and yield important insights in the coming years.
Join the Conversation