Domino Effect

Methods Center

Center for Causal Inference

Photo by pupes1/Fotolia

Context

To inform decisions and public policy, researchers are often called on to draw inferences about causal connections between conditions or programs and outcomes of interest. The Center for Causal Inference (CCI) supports the development and application of methods of causal inference, particularly in policy research.

As policymaking becomes ever more complex, CCI is uniquely poised to advise decision makers on their best possible options. The need to understand relationships between cause and effect arises in almost every policy domain, including health, labor, education, environmental studies, public safety, and national security. Policymakers, stakeholders and researchers might be interested in answering questions such as: What is the effect of a new medical treatment on health? What are the major drivers of rising health care costs? How does teacher compensation affect student achievement? Does a classroom intervention impact student proficiency? What is the effect of disability insurance receipt on employment and earnings? How does military service affect subsequent earnings? Do infrastructure investments in conflict zones reduce local support of terrorists?

RAND and the Pardee RAND Graduate School have been at the forefront of causal inference methodology for decades. Today, complex new societal issues and advances in technology for collecting large amounts of data make development of new methods, and improvement of existing ones, imperative for decision making.

Methodologies & Tools

CCI researchers have expertise in a variety of causal inference methodologies, including randomized studies and propensity score adjustment. Randomized studies are utilized when feasible, but real world problems rarely offer the possibility of manipulating conditions randomly in order to infer their causal effect.

Researchers are innovating in the development and application of other research designs that yield estimates of causal relationships based on non-experimental or "observational" data, including difference-in-differences, instrumental variables, propensity score adjustment, and regression discontinuity. In each situation, selecting the right research design is critical, as is understanding which causal inferences can be drawn and which cannot. In some cases, new estimators need to be developed to overcome data and other limitations, in order to draw robust, accurate conclusions that can guide policy makers.

  • Difference-in-Difference
  • Instrumental variables
  • Propensity Scores
  • Randomized studies
  • Regression Discontinuity

Real-World Applications

Police car viewed in a vehicle side mirror

Photo by Stephen Fore/Fotolia

Assessing the effect of race bias in post-traffic stop outcomes using propensity scores

In response to community demands, court settlements, and state laws concerning racial profiling, police departments across the nation are collecting data on traffic stops. Numerous studies have used that data to show whether or not racial profiling is occurring in urban traffic stops. But such studies use flawed methods that cast doubt over their findings. In particular, most prior studies do not adequately control for meaningful differences on the profiles of traffic stops when trying to determine if racial profiling is occurring. CCI researchers applied causal inference methods in a study for the city of Oakland, California to demonstrate the effectiveness of more credible methods for assessing racial profiling, looking at both the decision to stop motorists and post-stop activities. The results showed how naive comparisons that do not use causal inference methods can misstate the magnitude of the problem.

The Causal Effects of Community-Based Treatments for Youth

Over 1.6 million adolescents in the United States meet criteria for substance use disorders (SUDs). However, evidence for the effectiveness of community-based psychosocial treatments for adolescent SUDs is mixed, at best. This study made significant contributions to estimation of the causal effects of community based treatment programs (CBTPs) for adolescent substance abusers. First, the project identified two methodological challenges to estimating causal effects of CBTPs when using observational data: First, there are pre-existing differences across a very wide range of characteristics among youth receiving different types of treatment. Second, some youth are institutionalized during the follow-up period and have outcomes measured which do not accurately reflect what their substance use might look like had they been free in the community. The project developed rigorous models and powerful tools for addressing these challenges and applied these tools to the substantive evaluations for this project. The tools were also used by the project team in related applications and were disseminated successfully to other researchers in a diverse set of fields. Additionally, the project was among the first to study the cumulative effects of multiple treatment episodes for youth. Over the course of two project periods, the results improved our understanding of how adolescent substance abuse treatment works and for whom it is most effective. The project also resulted in several new state-of-the-art methodological developments related to propensity score weight estimation.

"...When the use of discretion seems to side consistently against minorities, trust in the fairness of 'the system' degrades."

Greg Ridgeway

Expertise

The RAND Center for Causal Inference Co-Directors

Work with Us

Want to learn how state-of-the-art causal inference methods can further your research? Fill out the form below and we'll get back to you soon.

You can also contact: causalinference@rand.org

Methods Centers at Pardee RAND