Looking to build a “test, learn and adapt” culture, federal agencies have conducted dozens of rapid, low-cost experiments over the last few years.
What’s one of the least known success stories in recent years, in terms of evidence-based decision-making within the federal government? It’s the more than 70 rigorous, often rapid experiments conducted by agencies since 2015.
Those experiments, undertaken with the help of the General Services Administration’s Office of Evaluation Sciences, have used data that agencies already collect, so they are low-cost. All of the results are shared online.
An example: OES worked with the Centers for Medicare and Medicaid Services to try to reduce overprescribing of a common antipsychotic drug. The experiment drew on behavioral research to design a letter to top-prescribing doctors that included a graphic comparing their prescribing rates to those of their peers.
The results, based on the randomized experiment (doctors were randomly assigned either a letter with the peer comparison or a letter without it) showed an 11.1% reduction after nine months among those who received the peer-comparison letter. The experiment provided a valuable, relatively quick, low-cost insight about what works in addressing a priority problem. After all, “It costs just as much to send an effective letter as an ineffective one,” notes OES Director Kelly Bidwell.
Examples like these were highlighted in May at a gathering of federal agency officials and analysts organized by the Association of Government Accountants. An audience poll of questions related to rapid experimentation included the following. I hope my brief answers help more agencies embrace this powerful strategy for learning and improvement.
How is rapid experimentation different from “agile”?
Agile is project management philosophy that chunks a project into quick “sprints” and emphasizes being open to adjustments along the way. Rapid experimentation is about testing what works, but it could easily fit into an agile process. For example, experiments could produce insights that inform needed adjustments in an agile project.
How do you square the goal of experimentation with the need to treat everyone equally?
Government rightfully has a strong emphasis on equal treatment and fairness. However, that doesn’t mean, in most cases, that you can’t try different approaches to figure out how to serve people better. After all, you may find that the group receiving the experimental intervention did not fare better than the “control” group. Moreover, if the intervention does prove effective, you then would provide it to the control group.
Can rapid experimentation really work in government, where rules and processes can slow down the ability to act and react quickly?
Rapid experiments rely on processes that agencies can control and change quickly. For example, many students fail to submit a renewal Free Application for Federal Student Aid form each year, potentially missing out on valuable financial aid. The Department of Education’s Office of Federal Student Aid sent several versions of emails to students who had filed a FAFSA the previous year, encouraging them to submit a renewal FAFSA for the upcoming year. The results showed which messages were the most effective.
One piece of advice: Take a careful look at what your agency is actually allowed to do, since often times there are de facto rules or requirements that have calcified over time, but aren't actually a part of statute or regulations.
What does “rigorous” mean when applied to evidence?
It means the evidence is strong enough to have confidence in the results. In many cases, the easiest way to run a rapid experiment that produces rigorous evidence is to use a randomized trial.
How does this type of experimentation compare with other program evaluations, quality assurance processes, or inspectors general reports?
Unlike most evaluations and assessments, rapid experiments focus within programs—that is, on operations—so they’re particularly useful to program managers. Moreover, they are typically run as randomized trials, designed to produce credible insights about specific questions. And they generally use existing data, making them very low cost.
Where should we start?
Once agencies, offices or programs have identified the most important challenges their organizations are facing, they should ask: Are there ways to test out potential improvements quickly and cost-effectively? Given that making change in government isn’t easy, experiments can be a valuable way to say, “Let’s try it on a small scale and see if it works.” Hard evidence can help build momentum for broader changes. Moreover, experiments can help overcome gridlock when there are competing theories within an agency about how to address a challenge. Rapid experiments, in other words, can sometimes enable multiple approaches and theories to be tested simultaneously.
This movement toward a “test, learn and adapt” culture in government is nascent, but it’s something to be celebrated. That includes experiments that have shown null or even negative results, underscoring their value in testing assumptions. The experiments that did show positive results have helped those agencies scale up effective approaches.
Knowledge is power. It’s also good customer service and respect for taxpayers. For all these reasons, agency leaders should ensure that their offices and programs are building an experimental mindset.
Andrew Feldman is a director in the public sector practice at Grant Thornton and also hosts the Gov Innovator podcast. He served as a special adviser on the evidence team at the White House Office of Management and Budget in the Obama administration.