Stanford Computational Policy Lab Stanford Computational Policy Lab Menu

In partnership with government agencies and public interest groups, we work on projects spanning the spectrum of data collection, data analysis, and tool building. Below we describe some of work we’re doing at the intersection of data science and public policy.

Data Collection
Stanford Open Policing Project

More than 20 million Americans are stopped each year for traffic violations, making this one of the most common ways in which the public interacts with the police. But there is currently no comprehensive, national repository detailing these encounters.

Our team is working to change that. We’re gathering, analyzing, and releasing traffic stop records from dozens of state and local law enforcement agencies across the country. To date we’ve already released detailed data on more than 100 million stops. Our goal is to help researchers, journalists, and policymakers investigate and improve interactions between police and the public.

As part of this effort, we carried out a detailed analysis of traffic stops in Nashville, commissioned by the Nashville Mayor’s Office.

For more information, visit the Stanford Open Policing Project website.

Uncovering Debtors' Prisons

Debtors’ prisons have quietly reemerged across the country. Many municipalities have become reliant on fines and fees for minor offenses, like traffic violations, to fund their operations. And in some cases, failure to pay such debts can result in imprisonment.

In theory, debtors are protected from imprisonment unless they willfully fail to pay. In reality, however, the safeguards against jailing indigent individuals can be weak.

Despite the deleterious consequences of debtors’ prisons on individuals and communities, little is known about the nature and extent of this practice. To address this data gap, we’re building the first national database to rigorously document, understand, and ultimately help remedy this potentially widespread problem with our criminal justice system.

Data Analysis
One Person, One Vote

Prominent officials have alleged that millions of people vote twice in presidential elections, calling into question the bedrock of democratic governance. Past investigations have found no indication of widespread voter fraud, but critics argue that it’s simply hard to detect.

In the most comprehensive study of voter fraud to date, we examined over 100 million voting records for the 2012 presidential election. We found that double voting is exceedingly rare. We further found that one popular effort to prevent double voting — the Interstate Crosscheck Program — can in practice burden hundreds of legitimate voters for every double vote prevented.

To learn more, see our scientific paper, read our op-ed in Slate, or listen to an in-depth discussion of our work on This American Life. / Alleged Murderer by lightest light / CC by 2.0
Quantifying Bias

Policymakers often seek to assess bias in human decisions, including the actions of lenders, judges, and police officers. But traditional measures of discrimination suffer from significant limitations, hampering rigorous analysis.

Drawing on the latest developments in machine learning, we’re developing more robust tests for discrimination. Our threshold test, for example, gives policymakers a powerful new way to identify and track potential bias in organizations. When we applied this test to records collected as part of our Open Policing Project, we found evidence of bias in police searches nationwide; and in many instances, traditional tests of discrimination would have left such bias uncovered.

For more information, see our scientific papers (here and here) describing the method. We have also released code to run the test on your own data.

Daniel Schwen / CC BY-SA 4.0
Algorithms for Bail Reform

One of the most consequential — and controversial — use of algorithms is in the criminal justice system. When deciding which defendants to release before their trials, judges in many jurisdictions now consult risk assessment algorithms that aim to quantify the likelihood a defendant will engage in violent crime or fail to appear in court if released.

When properly designed, such algorithmic tools can reduce jail populations without adversely affecting public safety; when poorly designed, they can exacerbate troubling, historical disparities. We’re working both to understand how existing algorithms work, and to build new, equitable solutions for criminal justice reform.

To learn more about this work, see our op-eds in the New York Times and the Washington Post. You can also read our scientific paper on defining fair algorithms.
Three Strikes

In 2011, the Supreme Court ruled that prison overcrowding in California violated inmates’ Eighth Amendment protections. Many believe that sentencing enhancements—such as the 1994 “three strikes” law that can dramatically increase prison terms—has contributed to such overcrowding. But a lack of data has hindered rigorous analysis.

In cooperation with the San Francisco District Attorney’s Office, we are analyzing hundreds of thousands of criminal sentencing hearings spanning four decades. For the first time, we can examine how frequently enhancements are charged by prosecutors, which enhancements are most common, how enhancements relate to base charges, and how enhancements have contributed to the size of the prison population.

By documenting the role of sentencing enhancements in the criminal justice system, we hope to spur more equitable outcomes.

Tool Building
AI for Education

The recent push toward online instruction has dramatically increased access to education. But too often online education is passive, lacking the interactivity of in-person experiences.

To bridge this gap, we’re building a chatbot to help high-school students learn math. Our chatbot, called Ada, augments in-class instruction by reviewing course material and helping students work through problems tailored to enhance their understanding of key ideas. The problems presented, and the feedback students receive, are personalized to each student’s needs. And it all works through a simple, text-based conversational interface.

By combining the latest developments in artificial intelligence with best practices from education research, our goal is to improve mathematical literacy by offering this tool to teachers and students worldwide free of charge.
The Criminal Justice Pipeline

More than 2 million people are currently incarcerated in the United States, more than in any other country in the world. Despite the large number of people affected, a lack of data has made it difficult to document and understand the flow of individuals through the criminal justice system.

As participants in the MacArthur Safety and Justice Challenge, and in partnership with the City of St. Louis, we are building a public dashboard to map and visualize this process, from initial arrest, to court hearings, to incarceration and eventual release. We hope both to bring transparency to this often opaque system and to identify interventions that promote equitable outcomes.

Smart Prosecution

Each year, the San Francisco District Attorney’s Office processes thousands of arrests, but not all ultimately lead to a criminal charge. Once an arrest is made, intake attorneys have only a brief window to decide whether or not to proceed with or to dismiss the case.

In collaboration with the SFDA, the Stanford Computational Policy Lab is developing algorithmic tools to ensure these charging decisions are transparent, consistent, and equitable.