In the 25 years since its creation, the Institute of Education Sciences (IES) has helped build a field where research is now more rigorous than ever before. But that work is not done. As the Trump Administration “reimagines” the future of IES and, hopefully, rebuilds it, IES must pay more attention to replicating research findings, a cornerstone of strong science.
A September 2025 House Appropriations Subcommittee report on the FY2026 budget, focused on the National Institutes of Health, encourages funding for replication due to concerns that many biomedical studies are irreproducible or fraudulent. While the report does not mention education, the subcommittee’s concerns apply; although, compared to biomedical work, education research is likely less prone to fraud and more prone to poor practices.
I have often joked that education research does not have a replication crisis—but only because replications are so rare in education research. Accumulating evidence suggests the joke is on me. Perry and colleagues report that, between 2011 and 2020, replication studies account for around 0.20% (yes, that’s 1/5 of one percent) of published papers in education journals, leading them to conclude that “the scientific status of the evidence base for educational policy and practices is currently in question.”
A recent Nature forum reinforces this finding. Miske et al. started with a stratified random sample of 600 papers published from 2009 to 2018 in 62 journals spanning the social and behavioral sciences. Only 144 (24%) of the authors of these papers made data available for testing and 38 more papers had source data Miske et al. could access. Using the truncated dataset they created, Miske et al. found that over half of education studies reviewed had never been reproduced, and none were precisely replicated. This was a worse record than in the other fields studied (and we don’t know how the low rate of participation may have skewed results).
The authors of another article in the forum (Tyner et al.) attempted to replicate 274 claims of positive results in 54 social and behavioral science journals. They were able to replicate the original pattern of results for 55% of the published claims. About 60% of education findings replicated—slightly better than the 50% cross-discipline rate, but hardly reassuring.
Much of the discussion of the replication crisis focuses on fraud; however, many of the problems of research flow from poor research practices rather than intentional deception. As evident in the Nature articles, one of the biggest limits on replication is that the published authors did not share data or programming code, making replication hard or impossible. How could fewer than one-quarter of published authors get away with not sharing data with Miske et al.? Why don’t all journals require that data and code be filed with a data repository, such as ICPSR?
Other poor research practices are widespread in education studies: small sample sizes, the misuse of p values including p-hacking (running multiple analyses until finding significant results), HARKing (hypothesizing after results are known), and selective reporting of favorable outcomes. This is all aided and abetted by academic journals, which pursue “novel” research findings rather than the litmus test inherent in replications.
Education research also faces many field-specific challenges. Unlike laboratory sciences, education interventions occur in classroom environments that resist standardization. Teacher quality, school culture, student demographics, and implementation fidelity all vary. Additionally, many educational outcomes take years to measure, making replication studies expensive and slow. Still, difficulty is not an excuse. As Billie Holiday puts it: “The difficult I’ll do right now/The impossible will take a little while.”
IES has a responsibility to support research that identifies and curbs poor research practices and fraud. Its Standards for Excellence in Education Research (SEER) help in that task. Pre-registration of study designs and analysis plans, which is encouraged by SEER, is becoming more common, reducing researcher opportunities to engage in questionable analytical practices. SEER also calls for open science practices, including sharing data and materials, which facilitates independent verification. For many IES-funded projects, adherence to SEER principles is already required. As IES rebuilds, policing adherence to SEER requires hard work, resources, and effective communications to researchers, educators, and policy makers.
Just as important, SEER principles need to be adapted to accommodate the growing use of Digital Learning Platforms (DLPs) to implement experiments in classrooms. IES-supported researchers will need to police or adapt themselves to meet new standards and expectations. An IES grant supporting such platforms for delivering experiments and replicating them created SEERNet to coordinate the work of multiple large scale DLPs. The goal of SEERNet mirrors IES’s mission: to learn about what works for whom under what circumstances to improve student outcomes. But what makes DLPs so important is that they can do this in a way that is faster and cheaper using larger and more inclusive samples than traditional education research. As IES continues to support a more modern education infrastructure, it needs to adapt SEER to fit that new world.
IES can also draw on strategies proposed by Stuart Buck for increasing replication, including working with journals and scientific societies to elevate the status of replication studies.
Replication’s greatest value is not necessarily uncovering fraud but identifying errors, inflated effects, and context-specific findings. Strengthening replication will ensure that education policy and practice rest on solid evidence rather than statistical artifacts or weak methods. A rebuilt IES should make this a central priority.
