Whose Law
Legal Overfitting
Ian Ayres & Yair Listokin
American Law and Economics Review, forthcoming
Abstract:
A central concern of machine learning is overfitting -- which occurs when a prediction model includes too many explanatory variables and predicts noise. The problem with overfitting is that it leads the model to poorer out-of-sample predictions because it misattributes causal significance to irrelevant variables. We argue that a similar phenomenon reduces the quality of precedential reasoning. A judge or lawyer trying to reconcile prior authoritative opinions that are decided with some noise to make the best prediction about the outcome of a new case may misattribute causal significance to factors that do not offer precedential guidance. Machine learning has developed a series of estimation and diagnostic techniques to reduce the likelihood of overfitting. For example, it is standard to train models on a subset of the available data and then test how well the model predicts out-of-sample. This article argues that the quality of precedential reasoning would be improved if judges used analogous techniques in deciding cases. Lawyers might also use these techniques to improve their ability to predict what the law is. We present a decision tree estimation of copyright law's fair use defense to illustrate the phenomenon of overfitting and how it might be limited.
Access to Justice in the Age of AI: Evidence from U.S. Federal Courts
Anand Shah & Joshua Levy
MIT Working Paper, March 2026
Abstract:
This paper studies how generative AI has reshaped entry into the federal civil court system. Drawing on administrative records covering more than 4.5 million non-prisoner federal civil court cases from FY2005-FY2026 and 46 million PACER docket entries matched to those cases, we document three sets of findings. First, the number of pro se cases -- or self-represented cases -- is increasing dramatically, rising from a long-term steady-state average of 11% to 16.8% in FY2025. This increase is concentrated in case types characterized by formulaic document production and absent from more complex, attorney-intensive categories. Second, we argue these cases are placing larger burden on federal district courts. Pro se cases are not terminating faster, and this combined with the increased case numbers suggests more cases for judges to process. Moreover, intra-case activity is up, with the total volume of docket entries per court generated by pro se cases in their first 180 days up 158% from pre-AI means to 2025. Third, we directly validate that AI use is increasing in federal courts. Using a random sample of 1,600 complaints drawn from an 8-year period (2019-2026), we find that a large and growing share of complaints are flagging positive for AI-generated text, from essentially zero in the pre-AI period to more than 18% in 2026.
State Attorneys General, Revolving Doors and the Market for Regulatory Compliance
Colin Provost
Law & Social Inquiry, forthcoming
Abstract:
Revolving door research on government lawyers moving into the private sector shows multiple potential outcomes. Such lawyers may utilize connections in government to secure special favors for their new clients, in a manner consistent with regulatory capture (the Quid Pro Quo Hypothesis). Conversely, they may utilize their knowledge and expertise to improve the regulatory compliance of their clients (the Regulatory Schooling Hypothesis). I examine these hypotheses in the context of state attorneys general (state AGs) and multi-state litigation. Multi-state litigation is an important and prominent form of state and federal regulation. Law firms have responded by creating "State AG Practices," which help companies cope with the regulatory threat presented by state AGs. I utilize sixteen semi-structured interviews, career data of 194 state AGs, and legal media secondary sources to analyze whether AGs move through the revolving door in order to improve regulatory compliance or facilitate regulatory capture. Contrary to some media accounts, former AGs primarily bring expertise, rather than government connections, to the private sector, but using this expertise is often easier when well-known former colleagues are across the negotiating table.
The Structure of a Federal Appeal
Tyler Lindley
Vanderbilt Law Review, forthcoming
Abstract:
Federal appellate procedural law is in large part unwritten, including doctrines like preservation, appellate standing, vacatur, divestiture, and relief pending appeal. Courts have developed these doctrines in a piecemeal fashion, without recognizing the source of or justification for this body of law. The result is a confusing and sometimes contradictory body of law that can feel more like pretext than principle. This Article uncovers a unifying theory of unwritten appellate procedure. Traditionally, appeals were less a dispute between two parties and more a challenge in an appellate court to the lower court's judgment. This basic structure still shapes modern doctrine. Once a court issues a judgment, it becomes the legal status quo. In a sense, the lower court (or at least its judgment) becomes a party in interest to the appeal, and appellate procedure works to protect that interest. This insight carries both explanatory and prescriptive force. It explains why the Court in United States v. Sineneng-Smith demanded adherence to issue preservation, how district and appellate courts should treat the limited-record principle articulated in New York State Rifle & Pistol Ass'n v. Bruen, why appellate standing hinges on injury from the judgment, and why courts still apply the divestiture rule invoked in Moore v. Harper and Coinbase, Inc. v. Bielski. It even helps resolve the Supreme Court's ongoing struggle in defining the status quo for relief pending appeal. Although not calling for a doctrinal overhaul, this article provides a framework for courts to make sense of the chaos. By refocusing attention on these underappreciated doctrines, appellate courts would shift some of their discretion and power to district courts. And even if the old way of thinking about appellate procedure is too far gone, understanding the modern system's roots might provide impetus for reform through legislation or rulemaking.
Judicial Institutions, Independence, and the Separation of Powers
Teena Wilhelm & David Hughes
Political Research Quarterly, forthcoming
Abstract:
We study the separation of powers between judicial and legislative institutions among the American states. Building upon previous studies, we consider how judicial selection/accountability mechanisms affect the judicial-legislative separation of powers game. We specify game theoretic models that consider these issues. Our formal models demonstrate that judicial elections can empower courts by providing them with direct signals of legitimacy from voters. We then examine the use of judicial review in state supreme courts between 2007 and 2017 both at the institutional and justice level. Our results suggest that electoral institutions empower courts in the separation of powers game, as these institutions are associated with significantly higher odds of invalidating legislative acts compared to unelected courts. Furthermore, we find no evidence that legislative court-curbing efforts constrain judicial independence and argue the time may be ripe to revisit the role court-curbing plays in judicial preemption in American politics.
When Do Judges Throw the Book at Companies? The Influence of Partisanship in Corporate Prosecutions
Todd Gormley, Mahsa Kaviani & Hosein Maleki
Review of Financial Studies, forthcoming
Abstract:
We document that judges' political affiliations are strongly associated with the level of judicial penalties levied against companies. For example, Republican-appointed judges impose larger fines for hiring illegal immigrants, while Democrat-appointed judges impose larger fines for pollution- and environment-related violations. Time-series variation suggests that political partisanship, not fixed ideological differences, drives these findings. The differences become amplified when higher-court judicial vacancies exist and in the months before national elections. Our findings highlight the importance of political polarization for U.S. companies and illustrate how judicial composition can affect firms' incentive to avoid violating laws connected to partisan issues.
Racial Disparities in Federal Sentencing: Evidence from Drug Mandatory Minimums
Cody Tuttle
Review of Economic Studies, forthcoming
Abstract:
I study racial disparities in the criminal justice system by analyzing abnormal bunching in the distribution of crack-cocaine amounts used in federal sentencing. I compare cases sentenced before and after the Fair Sentencing Act, a 2010 law that changed the 10-year mandatory minimum threshold for crack-cocaine from 50g to 280g. First, I find that after 2010, there is a sharp increase in the fraction of cases sentenced at 280g (the point that now triggers a 10-year mandatory minimum), and that this increase is disproportionately large for black and Hispanic offenders. I then explore several possible explanations for the observed racial disparities, including racial discrimination that occurs after entry into the criminal justice system. I analyze data from multiple stages in the criminal justice system and find that the increased bunching for minority offenders is driven by prosecutorial discretion, specifically as used by about 20-30% of prosecutors. Moreover, the fraction of cases at 280g falls in 2013 when evidentiary standards become stricter. Finally, the racial disparity in the increase cannot be explained by differences in education, sex, age, criminal history, seized drug amount, or other elements of the crime, but it can be largely explained by a measure of state-level racial animus. These results shed light on the role of prosecutorial discretion and racial discrimination as causes of racial disparities in sentencing.
"You're Trying to Do What Now?" A Mystery Shopper Study on Judicial Bypass in Alabama, Georgia, and Mississippi
Erin Carroll et al.
Perspectives on Sexual and Reproductive Health, forthcoming
Methods: Between July 2019 and February 2020, we conducted mystery shopper calls to county courthouses in three states mandating parental involvement: Alabama, Georgia, and Mississippi. Research assistants called courthouses and asked about the bypass process, using a semi-structured script. We calculated the percentage of counties where courthouse staff could not provide callers with sufficient information about the process and conducted a content analysis of the call transcripts.
Results: Courthouse staff in 50% of the 214 contacted counties suggested bypass might be possible, but could not provide specific information and 28% communicated bypass would not be possible. Lacking familiarity with the process, staff often referred callers to places that did not handle judicial bypass (e.g., health departments) and sometimes conflated bypass with emancipation, or said abortion was not legal. In 21 counties, courthouse staff stepped outside their professional role and engaged with callers in a personal manner that was often unsupportive of a minor's abortion decision.
The impact of working conditions on productivity: Evidence from the U.S. public defense system
Zhihan Liu & Amy Mahler
Journal of Public Economics, May 2026
Abstract:
The Sixth Amendment guarantees the right to effective legal representation for all defendants, regardless of income. In practice, this constitutional promise depends critically on the working conditions of the public defenders who represent low-income clients. This paper examines how the workloads public defenders face affect the outcomes of the cases they handle. Using the universe of public defense case records from two U.S. states and quasi-random variation in workloads induced by strict case assignment rules, we find robust evidence that a one-standard-deviation increase in an attorney's workload (about 25 additional cases) increases incarceration sentence length by 12 percent, with suggestive evidence of higher probability of conviction. We explore mechanisms using granular attorney time-use data and find that rising workloads reduce time spent across all tasks by 18 percent, while some independent tasks, such as reviewing evidence and conducting legal research, may be dropped entirely.
Learning About Police Bias: Prosecutors and Police Before and After Body-Worn Cameras
Emma Harrington & Hannah Shaffer
University of Virginia Working Paper, April 2026
Abstract:
Decision-makers often rely on earlier actors but fail to correct for their biases. We model and measure two mechanisms: underestimating upstream bias and treating subjective information as ground truth. We link an original survey of 203 North Carolina prosecutors to their 505,787 cases. Exploiting the rollout of police body-worn cameras (BWC), we show monitoring reduces incarceration disparities by 14 percent, little of which is driven by arrests. About one quarter of this effect reflects learning: prosecutors with greater BWC exposure view police as more biased and unreliable. Monitoring reduces disparities most for prosecutors who treat police reports as ground truth.